diff --git a/.gitattributes b/.gitattributes
index b0f67b10ceb944f1d564b620fb0f1b9ca7a5a9e7..07eaa1b767547fa6ea87cded4213162b1d20824f 100644
--- a/.gitattributes
+++ b/.gitattributes
@@ -166,3 +166,12 @@ pythonProject/.venv/Lib/site-packages/torch/lib/cupti64_2022.3.0.dll filter=lfs
pythonProject/.venv/Lib/site-packages/torch/bin/protoc.exe filter=lfs diff=lfs merge=lfs -text
pythonProject/.venv/Lib/site-packages/torch/lib/cudnn_graph64_9.dll filter=lfs diff=lfs merge=lfs -text
pythonProject/.venv/Lib/site-packages/torch/lib/cudnn_engines_runtime_compiled64_9.dll filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/cudnn_heuristic64_9.dll filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/cublas64_11.dll filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/libiomp5md.dll filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/libomp140.x86_64.dll filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/kineto.lib filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/libprotobuf-lite.lib filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/curand64_10.dll filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/fbgemm.lib filter=lfs diff=lfs merge=lfs -text
+pythonProject/.venv/Lib/site-packages/torch/lib/fmt.lib filter=lfs diff=lfs merge=lfs -text
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/__init__.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..09da42ea71c6054490b72f39567ddc9520464c50
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/__init__.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/pipeline_bria.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/pipeline_bria.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..4d391f5ab65b6f4f6355e5f533b2aa401a4c5db3
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/pipeline_bria.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/pipeline_output.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/pipeline_output.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..87acceff76e1e01dabe60f79dd6d6bf3ec47fa8b
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/bria/__pycache__/pipeline_output.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..d9238b735c4156f25a831671196f418bd565db60
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__init__.py
@@ -0,0 +1,49 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+ DIFFUSERS_SLOW_IMPORT,
+ OptionalDependencyNotAvailable,
+ _LazyModule,
+ get_objects_from_module,
+ is_torch_available,
+ is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_additional_imports = {}
+_import_structure = {"pipeline_output": ["ChromaPipelineOutput"]}
+
+try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+ from ...utils import dummy_torch_and_transformers_objects # noqa F403
+
+ _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+ _import_structure["pipeline_chroma"] = ["ChromaPipeline"]
+ _import_structure["pipeline_chroma_img2img"] = ["ChromaImg2ImgPipeline"]
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+ except OptionalDependencyNotAvailable:
+ from ...utils.dummy_torch_and_transformers_objects import * # noqa F403
+ else:
+ from .pipeline_chroma import ChromaPipeline
+ from .pipeline_chroma_img2img import ChromaImg2ImgPipeline
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
+
+ for name, value in _dummy_objects.items():
+ setattr(sys.modules[__name__], name, value)
+ for name, value in _additional_imports.items():
+ setattr(sys.modules[__name__], name, value)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/__init__.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..88479cd13a7a8f6bc8d65c0b6b96bbb287964e40
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/__init__.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_chroma.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_chroma.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..b4d42000b82d49699de124251845acf50f1f4722
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_chroma.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_chroma_img2img.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_chroma_img2img.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..ecd4c3947c3ade12522771281bfdbd894b5eb7a8
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_chroma_img2img.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_output.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_output.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..75abe80943f1c4ebc01f33513c12d11d646a607c
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/__pycache__/pipeline_output.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_chroma.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_chroma.py
new file mode 100644
index 0000000000000000000000000000000000000000..a3dd1422b876b4b456ff2cccac8cf74db6038a02
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_chroma.py
@@ -0,0 +1,949 @@
+# Copyright 2025 Black Forest Labs and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import numpy as np
+import torch
+from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection, T5EncoderModel, T5TokenizerFast
+
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import FluxIPAdapterMixin, FluxLoraLoaderMixin, FromSingleFileMixin, TextualInversionLoaderMixin
+from ...models import AutoencoderKL, ChromaTransformer2DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import (
+ USE_PEFT_BACKEND,
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import randn_tensor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import ChromaPipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> import torch
+ >>> from diffusers import ChromaPipeline
+
+ >>> model_id = "lodestones/Chroma"
+ >>> ckpt_path = "https://huggingface.co/lodestones/Chroma/blob/main/chroma-unlocked-v37.safetensors"
+ >>> transformer = ChromaTransformer2DModel.from_single_file(ckpt_path, torch_dtype=torch.bfloat16)
+ >>> pipe = ChromaPipeline.from_pretrained(
+ ... model_id,
+ ... transformer=transformer,
+ ... torch_dtype=torch.bfloat16,
+ ... )
+ >>> pipe.enable_model_cpu_offload()
+ >>> prompt = [
+ ... "A high-fashion close-up portrait of a blonde woman in clear sunglasses. The image uses a bold teal and red color split for dramatic lighting. The background is a simple teal-green. The photo is sharp and well-composed, and is designed for viewing with anaglyph 3D glasses for optimal effect. It looks professionally done."
+ ... ]
+ >>> negative_prompt = [
+ ... "low quality, ugly, unfinished, out of focus, deformed, disfigure, blurry, smudged, restricted palette, flat colors"
+ ... ]
+ >>> image = pipe(prompt, negative_prompt=negative_prompt).images[0]
+ >>> image.save("chroma.png")
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
+def calculate_shift(
+ image_seq_len,
+ base_seq_len: int = 256,
+ max_seq_len: int = 4096,
+ base_shift: float = 0.5,
+ max_shift: float = 1.15,
+):
+ m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+ b = base_shift - m * base_seq_len
+ mu = image_seq_len * m + b
+ return mu
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class ChromaPipeline(
+ DiffusionPipeline,
+ FluxLoraLoaderMixin,
+ FromSingleFileMixin,
+ TextualInversionLoaderMixin,
+ FluxIPAdapterMixin,
+):
+ r"""
+ The Chroma pipeline for text-to-image generation.
+
+ Reference: https://huggingface.co/lodestones/Chroma/
+
+ Args:
+ transformer ([`ChromaTransformer2DModel`]):
+ Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
+ scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representation
+ text_encoder ([`T5EncoderModel`]):
+ [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically
+ the [google/t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant.
+ tokenizer (`T5TokenizerFast`):
+ Second Tokenizer of class
+ [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast).
+ """
+
+ model_cpu_offload_seq = "text_encoder->image_encoder->transformer->vae"
+ _optional_components = ["image_encoder", "feature_extractor"]
+ _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+ def __init__(
+ self,
+ scheduler: FlowMatchEulerDiscreteScheduler,
+ vae: AutoencoderKL,
+ text_encoder: T5EncoderModel,
+ tokenizer: T5TokenizerFast,
+ transformer: ChromaTransformer2DModel,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ feature_extractor: CLIPImageProcessor = None,
+ ):
+ super().__init__()
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ transformer=transformer,
+ scheduler=scheduler,
+ image_encoder=image_encoder,
+ feature_extractor=feature_extractor,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ # Flux latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
+ # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
+ self.default_sample_size = 128
+
+ def _get_t5_prompt_embeds(
+ self,
+ prompt: Union[str, List[str]] = None,
+ num_images_per_prompt: int = 1,
+ max_sequence_length: int = 512,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ device = device or self._execution_device
+ dtype = dtype or self.text_encoder.dtype
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ batch_size = len(prompt)
+
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_sequence_length,
+ truncation=True,
+ return_length=False,
+ return_overflowing_tokens=False,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ attention_mask = text_inputs.attention_mask.clone()
+
+ # Chroma requires the attention mask to include one padding token
+ seq_lengths = attention_mask.sum(dim=1)
+ mask_indices = torch.arange(attention_mask.size(1)).unsqueeze(0).expand(batch_size, -1)
+ attention_mask = (mask_indices <= seq_lengths.unsqueeze(1)).long()
+
+ prompt_embeds = self.text_encoder(
+ text_input_ids.to(device), output_hidden_states=False, attention_mask=attention_mask.to(device)
+ )[0]
+
+ dtype = self.text_encoder.dtype
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+ attention_mask = attention_mask.to(dtype=dtype, device=device)
+
+ _, seq_len, _ = prompt_embeds.shape
+
+ # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ attention_mask = attention_mask.repeat(1, num_images_per_prompt)
+ attention_mask = attention_mask.view(batch_size * num_images_per_prompt, seq_len)
+
+ return prompt_embeds, attention_mask
+
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Union[str, List[str]] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ do_classifier_free_guidance: bool = True,
+ max_sequence_length: int = 512,
+ lora_scale: Optional[float] = None,
+ ):
+ r"""
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`
+ instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, FluxLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None and USE_PEFT_BACKEND:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ prompt_embeds, prompt_attention_mask = self._get_t5_prompt_embeds(
+ prompt=prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+
+ dtype = self.text_encoder.dtype if self.text_encoder is not None else self.transformer.dtype
+ text_ids = torch.zeros(prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)
+ negative_text_ids = None
+
+ if do_classifier_free_guidance:
+ if negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt = (
+ batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ )
+
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+
+ negative_prompt_embeds, negative_prompt_attention_mask = self._get_t5_prompt_embeds(
+ prompt=negative_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+
+ negative_text_ids = torch.zeros(negative_prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)
+
+ if self.text_encoder is not None:
+ if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ return (
+ prompt_embeds,
+ text_ids,
+ prompt_attention_mask,
+ negative_prompt_embeds,
+ negative_text_ids,
+ negative_prompt_attention_mask,
+ )
+
+ # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ return image_embeds
+
+ # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt
+ ):
+ image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != self.transformer.encoder_hid_proj.num_ip_adapters:
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {self.transformer.encoder_hid_proj.num_ip_adapters} IP Adapters."
+ )
+
+ for single_ip_adapter_image in ip_adapter_image:
+ single_image_embeds = self.encode_image(single_ip_adapter_image, device, 1)
+ image_embeds.append(single_image_embeds[None, :])
+ else:
+ if not isinstance(ip_adapter_image_embeds, list):
+ ip_adapter_image_embeds = [ip_adapter_image_embeds]
+
+ if len(ip_adapter_image_embeds) != self.transformer.encoder_hid_proj.num_ip_adapters:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` must have same length as the number of IP Adapters. Got {len(ip_adapter_image_embeds)} image embeds and {self.transformer.encoder_hid_proj.num_ip_adapters} IP Adapters."
+ )
+
+ for single_image_embeds in ip_adapter_image_embeds:
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for single_image_embeds in image_embeds:
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ negative_prompt=None,
+ prompt_embeds=None,
+ prompt_attention_mask=None,
+ negative_prompt_embeds=None,
+ negative_prompt_attention_mask=None,
+ callback_on_step_end_tensor_inputs=None,
+ max_sequence_length=None,
+ ):
+ if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
+ logger.warning(
+ f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}. Dimensions will be resized accordingly"
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and prompt_attention_mask is None:
+ raise ValueError("Cannot provide `prompt_embeds` without also providing `prompt_attention_mask")
+
+ if negative_prompt_embeds is not None and negative_prompt_attention_mask is None:
+ raise ValueError(
+ "Cannot provide `negative_prompt_embeds` without also providing `negative_prompt_attention_mask"
+ )
+
+ if max_sequence_length is not None and max_sequence_length > 512:
+ raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")
+
+ @staticmethod
+ def _prepare_latent_image_ids(batch_size, height, width, device, dtype):
+ latent_image_ids = torch.zeros(height, width, 3)
+ latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height)[:, None]
+ latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width)[None, :]
+
+ latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape
+
+ latent_image_ids = latent_image_ids.reshape(
+ latent_image_id_height * latent_image_id_width, latent_image_id_channels
+ )
+
+ return latent_image_ids.to(device=device, dtype=dtype)
+
+ @staticmethod
+ def _pack_latents(latents, batch_size, num_channels_latents, height, width):
+ latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
+ latents = latents.permute(0, 2, 4, 1, 3, 5)
+ latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)
+
+ return latents
+
+ @staticmethod
+ def _unpack_latents(latents, height, width, vae_scale_factor):
+ batch_size, num_patches, channels = latents.shape
+
+ # VAE applies 8x compression on images but we must also account for packing which requires
+ # latent height and width to be divisible by 2.
+ height = 2 * (int(height) // (vae_scale_factor * 2))
+ width = 2 * (int(width) // (vae_scale_factor * 2))
+
+ latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
+ latents = latents.permute(0, 3, 1, 4, 2, 5)
+
+ latents = latents.reshape(batch_size, channels // (2 * 2), height, width)
+
+ return latents
+
+ def enable_vae_slicing(self):
+ r"""
+ Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+ compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+ """
+ self.vae.enable_slicing()
+
+ def disable_vae_slicing(self):
+ r"""
+ Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+ computing decoding in one step.
+ """
+ self.vae.disable_slicing()
+
+ def enable_vae_tiling(self):
+ r"""
+ Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+ compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+ processing larger images.
+ """
+ self.vae.enable_tiling()
+
+ def disable_vae_tiling(self):
+ r"""
+ Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+ computing decoding in one step.
+ """
+ self.vae.disable_tiling()
+
+ # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.prepare_latents
+ def prepare_latents(
+ self,
+ batch_size,
+ num_channels_latents,
+ height,
+ width,
+ dtype,
+ device,
+ generator,
+ latents=None,
+ ):
+ # VAE applies 8x compression on images but we must also account for packing which requires
+ # latent height and width to be divisible by 2.
+ height = 2 * (int(height) // (self.vae_scale_factor * 2))
+ width = 2 * (int(width) // (self.vae_scale_factor * 2))
+
+ shape = (batch_size, num_channels_latents, height, width)
+
+ if latents is not None:
+ latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+ return latents.to(device=device, dtype=dtype), latent_image_ids
+
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
+
+ latent_image_ids = self._prepare_latent_image_ids(batch_size, height // 2, width // 2, device, dtype)
+
+ return latents, latent_image_ids
+
+ def _prepare_attention_mask(
+ self,
+ batch_size,
+ sequence_length,
+ dtype,
+ attention_mask=None,
+ ):
+ if attention_mask is None:
+ return attention_mask
+
+ # Extend the prompt attention mask to account for image tokens in the final sequence
+ attention_mask = torch.cat(
+ [attention_mask, torch.ones(batch_size, sequence_length, device=attention_mask.device)],
+ dim=1,
+ )
+ attention_mask = attention_mask.to(dtype)
+
+ return attention_mask
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def joint_attention_kwargs(self):
+ return self._joint_attention_kwargs
+
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def current_timestep(self):
+ return self._current_timestep
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ negative_prompt: Union[str, List[str]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 35,
+ sigmas: Optional[List[float]] = None,
+ guidance_scale: float = 5.0,
+ num_images_per_prompt: Optional[int] = 1,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ negative_ip_adapter_image: Optional[PipelineImageInput] = None,
+ negative_ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+ callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ max_sequence_length: int = 512,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ not greater than `1`).
+ height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The height in pixels of the generated image. This is set to 1024 by default for the best results.
+ width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The width in pixels of the generated image. This is set to 1024 by default for the best results.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+ their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+ will be used.
+ guidance_scale (`float`, *optional*, defaults to 3.5):
+ Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+ a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+ Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+ the [paper](https://huggingface.co/papers/2210.03142) to learn more.
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ negative_ip_adapter_image:
+ (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ negative_ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ prompt_attention_mask (torch.Tensor, *optional*):
+ Attention mask for the prompt embeddings. Used to mask out padding tokens in the prompt sequence.
+ Chroma requires a single padding token remain unmasked. Please refer to
+ https://huggingface.co/lodestones/Chroma#tldr-masking-t5-padding-tokens-enhanced-fidelity-and-increased-stability-during-training
+ negative_prompt_attention_mask (torch.Tensor, *optional*):
+ Attention mask for the negative prompt embeddings. Used to mask out padding tokens in the negative
+ prompt sequence. Chroma requires a single padding token remain unmasked. PLease refer to
+ https://huggingface.co/lodestones/Chroma#tldr-masking-t5-padding-tokens-enhanced-fidelity-and-increased-stability-during-training
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.flux.ChromaPipelineOutput`] instead of a plain tuple.
+ joint_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ callback_on_step_end (`Callable`, *optional*):
+ A function that calls at the end of each denoising steps during the inference. The function is called
+ with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+ callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+ `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+ max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.chroma.ChromaPipelineOutput`] or `tuple`: [`~pipelines.chroma.ChromaPipelineOutput`] if
+ `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the
+ generated images.
+ """
+
+ height = height or self.default_sample_size * self.vae_scale_factor
+ width = width or self.default_sample_size * self.vae_scale_factor
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ height,
+ width,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ prompt_attention_mask=prompt_attention_mask,
+ negative_prompt_embeds=negative_prompt_embeds,
+ negative_prompt_attention_mask=negative_prompt_attention_mask,
+ callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+ max_sequence_length=max_sequence_length,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._joint_attention_kwargs = joint_attention_kwargs
+ self._current_timestep = None
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ lora_scale = (
+ self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
+ )
+ (
+ prompt_embeds,
+ text_ids,
+ prompt_attention_mask,
+ negative_prompt_embeds,
+ negative_text_ids,
+ negative_prompt_attention_mask,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ prompt_attention_mask=prompt_attention_mask,
+ negative_prompt_attention_mask=negative_prompt_attention_mask,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ max_sequence_length=max_sequence_length,
+ lora_scale=lora_scale,
+ )
+
+ # 4. Prepare latent variables
+ num_channels_latents = self.transformer.config.in_channels // 4
+ latents, latent_image_ids = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 5. Prepare timesteps
+ sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
+ image_seq_len = latents.shape[1]
+ mu = calculate_shift(
+ image_seq_len,
+ self.scheduler.config.get("base_image_seq_len", 256),
+ self.scheduler.config.get("max_image_seq_len", 4096),
+ self.scheduler.config.get("base_shift", 0.5),
+ self.scheduler.config.get("max_shift", 1.15),
+ )
+
+ attention_mask = self._prepare_attention_mask(
+ batch_size=latents.shape[0],
+ sequence_length=image_seq_len,
+ dtype=latents.dtype,
+ attention_mask=prompt_attention_mask,
+ )
+ negative_attention_mask = self._prepare_attention_mask(
+ batch_size=latents.shape[0],
+ sequence_length=image_seq_len,
+ dtype=latents.dtype,
+ attention_mask=negative_prompt_attention_mask,
+ )
+
+ timesteps, num_inference_steps = retrieve_timesteps(
+ self.scheduler,
+ num_inference_steps,
+ device,
+ sigmas=sigmas,
+ mu=mu,
+ )
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+ self._num_timesteps = len(timesteps)
+
+ if (ip_adapter_image is not None or ip_adapter_image_embeds is not None) and (
+ negative_ip_adapter_image is None and negative_ip_adapter_image_embeds is None
+ ):
+ negative_ip_adapter_image = np.zeros((width, height, 3), dtype=np.uint8)
+ negative_ip_adapter_image = [negative_ip_adapter_image] * self.transformer.encoder_hid_proj.num_ip_adapters
+
+ elif (ip_adapter_image is None and ip_adapter_image_embeds is None) and (
+ negative_ip_adapter_image is not None or negative_ip_adapter_image_embeds is not None
+ ):
+ ip_adapter_image = np.zeros((width, height, 3), dtype=np.uint8)
+ ip_adapter_image = [ip_adapter_image] * self.transformer.encoder_hid_proj.num_ip_adapters
+
+ if self.joint_attention_kwargs is None:
+ self._joint_attention_kwargs = {}
+
+ image_embeds = None
+ negative_image_embeds = None
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ )
+ if negative_ip_adapter_image is not None or negative_ip_adapter_image_embeds is not None:
+ negative_image_embeds = self.prepare_ip_adapter_image_embeds(
+ negative_ip_adapter_image,
+ negative_ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ )
+
+ # 6. Denoising loop
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ self._current_timestep = t
+ if image_embeds is not None:
+ self._joint_attention_kwargs["ip_adapter_image_embeds"] = image_embeds
+
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = t.expand(latents.shape[0]).to(latents.dtype)
+
+ noise_pred = self.transformer(
+ hidden_states=latents,
+ timestep=timestep / 1000,
+ encoder_hidden_states=prompt_embeds,
+ txt_ids=text_ids,
+ img_ids=latent_image_ids,
+ attention_mask=attention_mask,
+ joint_attention_kwargs=self.joint_attention_kwargs,
+ return_dict=False,
+ )[0]
+
+ if self.do_classifier_free_guidance:
+ if negative_image_embeds is not None:
+ self._joint_attention_kwargs["ip_adapter_image_embeds"] = negative_image_embeds
+ neg_noise_pred = self.transformer(
+ hidden_states=latents,
+ timestep=timestep / 1000,
+ encoder_hidden_states=negative_prompt_embeds,
+ txt_ids=negative_text_ids,
+ img_ids=latent_image_ids,
+ attention_mask=negative_attention_mask,
+ joint_attention_kwargs=self.joint_attention_kwargs,
+ return_dict=False,
+ )[0]
+ noise_pred = neg_noise_pred + guidance_scale * (noise_pred - neg_noise_pred)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents_dtype = latents.dtype
+ latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+ if latents.dtype != latents_dtype:
+ if torch.backends.mps.is_available():
+ # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+ latents = latents.to(latents_dtype)
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ self._current_timestep = None
+
+ if output_type == "latent":
+ image = latents
+ else:
+ latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+ latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
+ image = self.vae.decode(latents, return_dict=False)[0]
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return ChromaPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_chroma_img2img.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_chroma_img2img.py
new file mode 100644
index 0000000000000000000000000000000000000000..233f4c43a1c235ca0f83d22c39f6ff00a5f16da7
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_chroma_img2img.py
@@ -0,0 +1,1034 @@
+# Copyright 2025 Black Forest Labs and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import numpy as np
+import torch
+from transformers import CLIPImageProcessor, CLIPVisionModelWithProjection, T5EncoderModel, T5TokenizerFast
+
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import FluxIPAdapterMixin, FluxLoraLoaderMixin, FromSingleFileMixin, TextualInversionLoaderMixin
+from ...models import AutoencoderKL, ChromaTransformer2DModel
+from ...schedulers import FlowMatchEulerDiscreteScheduler
+from ...utils import (
+ USE_PEFT_BACKEND,
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import randn_tensor
+from ..pipeline_utils import DiffusionPipeline
+from .pipeline_output import ChromaPipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> import torch
+ >>> from diffusers import ChromaTransformer2DModel, ChromaImg2ImgPipeline
+
+ >>> model_id = "lodestones/Chroma"
+ >>> ckpt_path = "https://huggingface.co/lodestones/Chroma/blob/main/chroma-unlocked-v37.safetensors"
+ >>> pipe = ChromaImg2ImgPipeline.from_pretrained(
+ ... model_id,
+ ... transformer=transformer,
+ ... torch_dtype=torch.bfloat16,
+ ... )
+ >>> pipe.enable_model_cpu_offload()
+ >>> init_image = load_image(
+ ... "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+ ... )
+ >>> prompt = "a scenic fastasy landscape with a river and mountains in the background, vibrant colors, detailed, high resolution"
+ >>> negative_prompt = "low quality, ugly, unfinished, out of focus, deformed, disfigure, blurry, smudged, restricted palette, flat colors"
+ >>> image = pipe(prompt, image=init_image, negative_prompt=negative_prompt).images[0]
+ >>> image.save("chroma-img2img.png")
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift
+def calculate_shift(
+ image_seq_len,
+ base_seq_len: int = 256,
+ max_seq_len: int = 4096,
+ base_shift: float = 0.5,
+ max_shift: float = 1.15,
+):
+ m = (max_shift - base_shift) / (max_seq_len - base_seq_len)
+ b = base_shift - m * base_seq_len
+ mu = image_seq_len * m + b
+ return mu
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class ChromaImg2ImgPipeline(
+ DiffusionPipeline,
+ FluxLoraLoaderMixin,
+ FromSingleFileMixin,
+ TextualInversionLoaderMixin,
+ FluxIPAdapterMixin,
+):
+ r"""
+ The Chroma pipeline for image-to-image generation.
+
+ Reference: https://huggingface.co/lodestones/Chroma/
+
+ Args:
+ transformer ([`ChromaTransformer2DModel`]):
+ Conditional Transformer (MMDiT) architecture to denoise the encoded image latents.
+ scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representation
+ text_encoder ([`T5EncoderModel`]):
+ [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically
+ the [google/t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant.
+ tokenizer (`T5TokenizerFast`):
+ Second Tokenizer of class
+ [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast).
+ """
+
+ model_cpu_offload_seq = "text_encoder->image_encoder->transformer->vae"
+ _optional_components = ["image_encoder", "feature_extractor"]
+ _callback_tensor_inputs = ["latents", "prompt_embeds"]
+
+ def __init__(
+ self,
+ scheduler: FlowMatchEulerDiscreteScheduler,
+ vae: AutoencoderKL,
+ text_encoder: T5EncoderModel,
+ tokenizer: T5TokenizerFast,
+ transformer: ChromaTransformer2DModel,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ feature_extractor: CLIPImageProcessor = None,
+ ):
+ super().__init__()
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ transformer=transformer,
+ scheduler=scheduler,
+ image_encoder=image_encoder,
+ feature_extractor=feature_extractor,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.latent_channels = self.vae.config.latent_channels if getattr(self, "vae", None) else 16
+
+ # Flux latents are turned into 2x2 patches and packed. This means the latent width and height has to be divisible
+ # by the patch size. So the vae scale factor is multiplied by the patch size to account for this
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor * 2)
+ self.default_sample_size = 128
+
+ def _get_t5_prompt_embeds(
+ self,
+ prompt: Union[str, List[str]] = None,
+ num_images_per_prompt: int = 1,
+ max_sequence_length: int = 512,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ device = device or self._execution_device
+ dtype = dtype or self.text_encoder.dtype
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ batch_size = len(prompt)
+
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_sequence_length,
+ truncation=True,
+ return_length=False,
+ return_overflowing_tokens=False,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ attention_mask = text_inputs.attention_mask.clone()
+
+ # Chroma requires the attention mask to include one padding token
+ seq_lengths = attention_mask.sum(dim=1)
+ mask_indices = torch.arange(attention_mask.size(1)).unsqueeze(0).expand(batch_size, -1)
+ attention_mask = (mask_indices <= seq_lengths.unsqueeze(1)).long()
+
+ prompt_embeds = self.text_encoder(
+ text_input_ids.to(device), output_hidden_states=False, attention_mask=attention_mask.to(device)
+ )[0]
+
+ dtype = self.text_encoder.dtype
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+ attention_mask = attention_mask.to(dtype=dtype, device=device)
+
+ _, seq_len, _ = prompt_embeds.shape
+
+ # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ attention_mask = attention_mask.repeat(1, num_images_per_prompt)
+ attention_mask = attention_mask.view(batch_size * num_images_per_prompt, seq_len)
+
+ return prompt_embeds, attention_mask
+
+ # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_inpaint.StableDiffusion3InpaintPipeline._encode_vae_image
+ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
+ if isinstance(generator, list):
+ image_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(image.shape[0])
+ ]
+ image_latents = torch.cat(image_latents, dim=0)
+ else:
+ image_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ image_latents = (image_latents - self.vae.config.shift_factor) * self.vae.config.scaling_factor
+
+ return image_latents
+
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Union[str, List[str]] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ do_classifier_free_guidance: bool = True,
+ max_sequence_length: int = 512,
+ lora_scale: Optional[float] = None,
+ ):
+ r"""
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`
+ instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, FluxLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None and USE_PEFT_BACKEND:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ prompt_embeds, prompt_attention_mask = self._get_t5_prompt_embeds(
+ prompt=prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+
+ dtype = self.text_encoder.dtype if self.text_encoder is not None else self.transformer.dtype
+ text_ids = torch.zeros(prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)
+ negative_text_ids = None
+
+ if do_classifier_free_guidance:
+ if negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt = (
+ batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ )
+
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+
+ negative_prompt_embeds, negative_prompt_attention_mask = self._get_t5_prompt_embeds(
+ prompt=negative_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+
+ negative_text_ids = torch.zeros(negative_prompt_embeds.shape[1], 3).to(device=device, dtype=dtype)
+
+ if self.text_encoder is not None:
+ if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ return (
+ prompt_embeds,
+ text_ids,
+ prompt_attention_mask,
+ negative_prompt_embeds,
+ negative_text_ids,
+ negative_prompt_attention_mask,
+ )
+
+ # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ return image_embeds
+
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt
+ ):
+ device = device or self._execution_device
+
+ image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != self.transformer.encoder_hid_proj.num_ip_adapters:
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {self.transformer.encoder_hid_proj.num_ip_adapters} IP Adapters."
+ )
+
+ for single_ip_adapter_image in ip_adapter_image:
+ single_image_embeds = self.encode_image(single_ip_adapter_image, device, 1)
+ image_embeds.append(single_image_embeds[None, :])
+ else:
+ if not isinstance(ip_adapter_image_embeds, list):
+ ip_adapter_image_embeds = [ip_adapter_image_embeds]
+
+ if len(ip_adapter_image_embeds) != self.transformer.encoder_hid_proj.num_ip_adapters:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` must have same length as the number of IP Adapters. Got {len(ip_adapter_image_embeds)} image embeds and {self.transformer.encoder_hid_proj.num_ip_adapters} IP Adapters."
+ )
+
+ for single_image_embeds in ip_adapter_image_embeds:
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for single_image_embeds in image_embeds:
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ strength,
+ negative_prompt=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ prompt_attention_mask=None,
+ negative_prompt_attention_mask=None,
+ callback_on_step_end_tensor_inputs=None,
+ max_sequence_length=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+
+ if height % (self.vae_scale_factor * 2) != 0 or width % (self.vae_scale_factor * 2) != 0:
+ logger.warning(
+ f"`height` and `width` have to be divisible by {self.vae_scale_factor * 2} but are {height} and {width}. Dimensions will be resized accordingly"
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and prompt_attention_mask is None:
+ raise ValueError("Cannot provide `prompt_embeds` without also providing `prompt_attention_mask")
+
+ if negative_prompt_embeds is not None and negative_prompt_attention_mask is None:
+ raise ValueError(
+ "Cannot provide `negative_prompt_embeds` without also providing `negative_prompt_attention_mask"
+ )
+
+ if max_sequence_length is not None and max_sequence_length > 512:
+ raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}")
+
+ @staticmethod
+ def _prepare_latent_image_ids(height, width, device, dtype):
+ latent_image_ids = torch.zeros(height, width, 3)
+ latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height)[:, None]
+ latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width)[None, :]
+
+ latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape
+
+ latent_image_ids = latent_image_ids.reshape(
+ latent_image_id_height * latent_image_id_width, latent_image_id_channels
+ )
+
+ return latent_image_ids.to(device=device, dtype=dtype)
+
+ @staticmethod
+ def _pack_latents(latents, batch_size, num_channels_latents, height, width):
+ latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2)
+ latents = latents.permute(0, 2, 4, 1, 3, 5)
+ latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4)
+
+ return latents
+
+ @staticmethod
+ def _unpack_latents(latents, height, width, vae_scale_factor):
+ batch_size, num_patches, channels = latents.shape
+
+ # VAE applies 8x compression on images but we must also account for packing which requires
+ # latent height and width to be divisible by 2.
+ height = 2 * (int(height) // (vae_scale_factor * 2))
+ width = 2 * (int(width) // (vae_scale_factor * 2))
+
+ latents = latents.view(batch_size, height // 2, width // 2, channels // 4, 2, 2)
+ latents = latents.permute(0, 3, 1, 4, 2, 5)
+
+ latents = latents.reshape(batch_size, channels // (2 * 2), height, width)
+
+ return latents
+
+ def enable_vae_slicing(self):
+ r"""
+ Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+ compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+ """
+ self.vae.enable_slicing()
+
+ def disable_vae_slicing(self):
+ r"""
+ Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+ computing decoding in one step.
+ """
+ self.vae.disable_slicing()
+
+ def enable_vae_tiling(self):
+ r"""
+ Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+ compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+ processing larger images.
+ """
+ self.vae.enable_tiling()
+
+ def disable_vae_tiling(self):
+ r"""
+ Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+ computing decoding in one step.
+ """
+ self.vae.disable_tiling()
+
+ # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3_img2img.StableDiffusion3Img2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(num_inference_steps * strength, num_inference_steps)
+
+ t_start = int(max(num_inference_steps - init_timestep, 0))
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ def prepare_latents(
+ self,
+ image,
+ timestep,
+ batch_size,
+ num_channels_latents,
+ height,
+ width,
+ dtype,
+ device,
+ generator,
+ latents=None,
+ ):
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ # VAE applies 8x compression on images but we must also account for packing which requires
+ # latent height and width to be divisible by 2.
+ height = 2 * (int(height) // (self.vae_scale_factor * 2))
+ width = 2 * (int(width) // (self.vae_scale_factor * 2))
+ shape = (batch_size, num_channels_latents, height, width)
+ latent_image_ids = self._prepare_latent_image_ids(height // 2, width // 2, device, dtype)
+
+ if latents is not None:
+ return latents.to(device=device, dtype=dtype), latent_image_ids
+
+ image = image.to(device=device, dtype=dtype)
+ if image.shape[1] != self.latent_channels:
+ image_latents = self._encode_vae_image(image=image, generator=generator)
+ else:
+ image_latents = image
+ if batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] == 0:
+ # expand init_latents for batch_size
+ additional_image_per_prompt = batch_size // image_latents.shape[0]
+ image_latents = torch.cat([image_latents] * additional_image_per_prompt, dim=0)
+ elif batch_size > image_latents.shape[0] and batch_size % image_latents.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {image_latents.shape[0]} to {batch_size} text prompts."
+ )
+ else:
+ image_latents = torch.cat([image_latents], dim=0)
+
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ latents = self.scheduler.scale_noise(image_latents, timestep, noise)
+ latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width)
+ return latents, latent_image_ids
+
+ def _prepare_attention_mask(
+ self,
+ batch_size,
+ sequence_length,
+ dtype,
+ attention_mask=None,
+ ):
+ if attention_mask is None:
+ return attention_mask
+
+ # Extend the prompt attention mask to account for image tokens in the final sequence
+ attention_mask = torch.cat(
+ [attention_mask, torch.ones(batch_size, sequence_length, device=attention_mask.device)],
+ dim=1,
+ )
+ attention_mask = attention_mask.to(dtype)
+
+ return attention_mask
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def joint_attention_kwargs(self):
+ return self._joint_attention_kwargs
+
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def current_timestep(self):
+ return self._current_timestep
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ negative_prompt: Union[str, List[str]] = None,
+ image: PipelineImageInput = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 35,
+ sigmas: Optional[List[float]] = None,
+ guidance_scale: float = 5.0,
+ strength: float = 0.9,
+ num_images_per_prompt: Optional[int] = 1,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ negative_ip_adapter_image: Optional[PipelineImageInput] = None,
+ negative_ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.tensor] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ joint_attention_kwargs: Optional[Dict[str, Any]] = None,
+ callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ max_sequence_length: int = 512,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ not greater than `1`).
+ height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The height in pixels of the generated image. This is set to 1024 by default for the best results.
+ width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The width in pixels of the generated image. This is set to 1024 by default for the best results.
+ num_inference_steps (`int`, *optional*, defaults to 35):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+ their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+ will be used.
+ guidance_scale (`float`, *optional*, defaults to 5.0):
+ Embedded guiddance scale is enabled by setting `guidance_scale` > 1. Higher `guidance_scale` encourages
+ a model to generate images more aligned with `prompt` at the expense of lower image quality.
+
+ Guidance-distilled models approximates true classifer-free guidance for `guidance_scale` > 1. Refer to
+ the [paper](https://huggingface.co/papers/2210.03142) to learn more.
+ strength (`float, *optional*, defaults to 0.9):
+ Conceptually, indicates how much to transform the reference image. Must be between 0 and 1. image will
+ be used as a starting point, adding more noise to it the larger the strength. The number of denoising
+ steps depends on the amount of noise initially added. When strength is 1, added noise will be maximum
+ and the denoising process will run for the full number of iterations specified in num_inference_steps.
+ A value of 1, therefore, essentially ignores image.
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ negative_ip_adapter_image:
+ (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ negative_ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ prompt_attention_mask (torch.Tensor, *optional*):
+ Attention mask for the prompt embeddings. Used to mask out padding tokens in the prompt sequence.
+ Chroma requires a single padding token remain unmasked. Please refer to
+ https://huggingface.co/lodestones/Chroma#tldr-masking-t5-padding-tokens-enhanced-fidelity-and-increased-stability-during-training
+ negative_prompt_attention_mask (torch.Tensor, *optional*):
+ Attention mask for the negative prompt embeddings. Used to mask out padding tokens in the negative
+ prompt sequence. Chroma requires a single padding token remain unmasked. PLease refer to
+ https://huggingface.co/lodestones/Chroma#tldr-masking-t5-padding-tokens-enhanced-fidelity-and-increased-stability-during-training
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.flux.ChromaPipelineOutput`] instead of a plain tuple.
+ joint_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ callback_on_step_end (`Callable`, *optional*):
+ A function that calls at the end of each denoising steps during the inference. The function is called
+ with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+ callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+ `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+ max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.chroma.ChromaPipelineOutput`] or `tuple`: [`~pipelines.chroma.ChromaPipelineOutput`] if
+ `return_dict` is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the
+ generated images.
+ """
+
+ height = height or self.default_sample_size * self.vae_scale_factor
+ width = width or self.default_sample_size * self.vae_scale_factor
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ height,
+ width,
+ strength,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ prompt_attention_mask=prompt_attention_mask,
+ negative_prompt_attention_mask=negative_prompt_attention_mask,
+ callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+ max_sequence_length=max_sequence_length,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._joint_attention_kwargs = joint_attention_kwargs
+ self._current_timestep = None
+ self._interrupt = False
+
+ # 2. Preprocess image
+ init_image = self.image_processor.preprocess(image, height=height, width=width)
+ init_image = init_image.to(dtype=torch.float32)
+
+ # 3. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+ lora_scale = (
+ self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None
+ )
+
+ (
+ prompt_embeds,
+ text_ids,
+ prompt_attention_mask,
+ negative_prompt_embeds,
+ negative_text_ids,
+ negative_prompt_attention_mask,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ prompt_attention_mask=prompt_attention_mask,
+ negative_prompt_attention_mask=negative_prompt_attention_mask,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ max_sequence_length=max_sequence_length,
+ lora_scale=lora_scale,
+ )
+
+ # 4. Prepare timesteps
+ sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) if sigmas is None else sigmas
+ image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
+ mu = calculate_shift(
+ image_seq_len,
+ self.scheduler.config.get("base_image_seq_len", 256),
+ self.scheduler.config.get("max_image_seq_len", 4096),
+ self.scheduler.config.get("base_shift", 0.5),
+ self.scheduler.config.get("max_shift", 1.15),
+ )
+ timesteps, num_inference_steps = retrieve_timesteps(
+ self.scheduler,
+ num_inference_steps,
+ device,
+ sigmas=sigmas,
+ mu=mu,
+ )
+ timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+ self._num_timesteps = len(timesteps)
+
+ if num_inference_steps < 1:
+ raise ValueError(
+ f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
+ f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
+ )
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+
+ # 5. Prepare latent variables
+ num_channels_latents = self.transformer.config.in_channels // 4
+ latents, latent_image_ids = self.prepare_latents(
+ init_image,
+ latent_timestep,
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ attention_mask = self._prepare_attention_mask(
+ batch_size=latents.shape[0],
+ sequence_length=image_seq_len,
+ dtype=latents.dtype,
+ attention_mask=prompt_attention_mask,
+ )
+ negative_attention_mask = self._prepare_attention_mask(
+ batch_size=latents.shape[0],
+ sequence_length=image_seq_len,
+ dtype=latents.dtype,
+ attention_mask=negative_prompt_attention_mask,
+ )
+
+ # 6. Prepare image embeddings
+ if (ip_adapter_image is not None or ip_adapter_image_embeds is not None) and (
+ negative_ip_adapter_image is None and negative_ip_adapter_image_embeds is None
+ ):
+ negative_ip_adapter_image = np.zeros((width, height, 3), dtype=np.uint8)
+ negative_ip_adapter_image = [negative_ip_adapter_image] * self.transformer.encoder_hid_proj.num_ip_adapters
+
+ elif (ip_adapter_image is None and ip_adapter_image_embeds is None) and (
+ negative_ip_adapter_image is not None or negative_ip_adapter_image_embeds is not None
+ ):
+ ip_adapter_image = np.zeros((width, height, 3), dtype=np.uint8)
+ ip_adapter_image = [ip_adapter_image] * self.transformer.encoder_hid_proj.num_ip_adapters
+
+ if self.joint_attention_kwargs is None:
+ self._joint_attention_kwargs = {}
+
+ image_embeds = None
+ negative_image_embeds = None
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ )
+ if negative_ip_adapter_image is not None or negative_ip_adapter_image_embeds is not None:
+ negative_image_embeds = self.prepare_ip_adapter_image_embeds(
+ negative_ip_adapter_image,
+ negative_ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ )
+
+ # 6. Denoising loop
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ self._current_timestep = t
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = t.expand(latents.shape[0])
+
+ if image_embeds is not None:
+ self._joint_attention_kwargs["ip_adapter_image_embeds"] = image_embeds
+
+ noise_pred = self.transformer(
+ hidden_states=latents,
+ timestep=timestep / 1000,
+ encoder_hidden_states=prompt_embeds,
+ txt_ids=text_ids,
+ img_ids=latent_image_ids,
+ attention_mask=attention_mask,
+ joint_attention_kwargs=self.joint_attention_kwargs,
+ return_dict=False,
+ )[0]
+
+ if self.do_classifier_free_guidance:
+ if negative_image_embeds is not None:
+ self._joint_attention_kwargs["ip_adapter_image_embeds"] = negative_image_embeds
+
+ noise_pred_uncond = self.transformer(
+ hidden_states=latents,
+ timestep=timestep / 1000,
+ encoder_hidden_states=negative_prompt_embeds,
+ txt_ids=negative_text_ids,
+ img_ids=latent_image_ids,
+ attention_mask=negative_attention_mask,
+ joint_attention_kwargs=self.joint_attention_kwargs,
+ return_dict=False,
+ )[0]
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents_dtype = latents.dtype
+ latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+ if latents.dtype != latents_dtype:
+ if torch.backends.mps.is_available():
+ # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+ latents = latents.to(latents_dtype)
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ self._current_timestep = None
+
+ if output_type == "latent":
+ image = latents
+ else:
+ latents = self._unpack_latents(latents, height, width, self.vae_scale_factor)
+ latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
+ image = self.vae.decode(latents, return_dict=False)[0]
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return ChromaPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_output.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_output.py
new file mode 100644
index 0000000000000000000000000000000000000000..951d132dba2eab63a08808f347baa4121ca4dc0b
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/chroma/pipeline_output.py
@@ -0,0 +1,21 @@
+from dataclasses import dataclass
+from typing import List, Union
+
+import numpy as np
+import PIL.Image
+
+from ...utils import BaseOutput
+
+
+@dataclass
+class ChromaPipelineOutput(BaseOutput):
+ """
+ Output class for Stable Diffusion pipelines.
+
+ Args:
+ images (`List[PIL.Image.Image]` or `np.ndarray`)
+ List of denoised PIL images of length `batch_size` or numpy array of shape `(batch_size, height, width,
+ num_channels)`. PIL images or numpy array present the denoised images of the diffusion pipeline.
+ """
+
+ images: Union[List[PIL.Image.Image], np.ndarray]
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e4fa1dda53d3f6500d408a2adc199a27abfeb8c3
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__init__.py
@@ -0,0 +1,54 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+ DIFFUSERS_SLOW_IMPORT,
+ OptionalDependencyNotAvailable,
+ _LazyModule,
+ get_objects_from_module,
+ is_torch_available,
+ is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+
+try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+ from ...utils import dummy_torch_and_transformers_objects # noqa F403
+
+ _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+ _import_structure["pipeline_cogvideox"] = ["CogVideoXPipeline"]
+ _import_structure["pipeline_cogvideox_fun_control"] = ["CogVideoXFunControlPipeline"]
+ _import_structure["pipeline_cogvideox_image2video"] = ["CogVideoXImageToVideoPipeline"]
+ _import_structure["pipeline_cogvideox_video2video"] = ["CogVideoXVideoToVideoPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+
+ except OptionalDependencyNotAvailable:
+ from ...utils.dummy_torch_and_transformers_objects import *
+ else:
+ from .pipeline_cogvideox import CogVideoXPipeline
+ from .pipeline_cogvideox_fun_control import CogVideoXFunControlPipeline
+ from .pipeline_cogvideox_image2video import CogVideoXImageToVideoPipeline
+ from .pipeline_cogvideox_video2video import CogVideoXVideoToVideoPipeline
+
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
+
+ for name, value in _dummy_objects.items():
+ setattr(sys.modules[__name__], name, value)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..daf40ec4f771b5328d7774d15bbfc2e049e23f86
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox_fun_control.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox_fun_control.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..3a4ce6c1262da57e8cc527133c6db3382e73f8df
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox_fun_control.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox_image2video.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox_image2video.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..9412b27bd5df5cdf591ae26cc265eade82ebf295
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/__pycache__/pipeline_cogvideox_image2video.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox.py
new file mode 100644
index 0000000000000000000000000000000000000000..4ac33b24bbe114a346354b41dfc756ae693f4de9
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox.py
@@ -0,0 +1,789 @@
+# Copyright 2025 The CogVideoX team, Tsinghua University & ZhipuAI and The HuggingFace Team.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import math
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import torch
+from transformers import T5EncoderModel, T5Tokenizer
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...loaders import CogVideoXLoraLoaderMixin
+from ...models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
+from ...models.embeddings import get_3d_rotary_pos_embed
+from ...pipelines.pipeline_utils import DiffusionPipeline
+from ...schedulers import CogVideoXDDIMScheduler, CogVideoXDPMScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from .pipeline_output import CogVideoXPipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```python
+ >>> import torch
+ >>> from diffusers import CogVideoXPipeline
+ >>> from diffusers.utils import export_to_video
+
+ >>> # Models: "THUDM/CogVideoX-2b" or "THUDM/CogVideoX-5b"
+ >>> pipe = CogVideoXPipeline.from_pretrained("THUDM/CogVideoX-2b", torch_dtype=torch.float16).to("cuda")
+ >>> prompt = (
+ ... "A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool in a serene bamboo forest. "
+ ... "The panda's fluffy paws strum a miniature acoustic guitar, producing soft, melodic tunes. Nearby, a few other "
+ ... "pandas gather, watching curiously and some clapping in rhythm. Sunlight filters through the tall bamboo, "
+ ... "casting a gentle glow on the scene. The panda's face is expressive, showing concentration and joy as it plays. "
+ ... "The background includes a small, flowing stream and vibrant green foliage, enhancing the peaceful and magical "
+ ... "atmosphere of this unique musical performance."
+ ... )
+ >>> video = pipe(prompt=prompt, guidance_scale=6, num_inference_steps=50).frames[0]
+ >>> export_to_video(video, "output.mp4", fps=8)
+ ```
+"""
+
+
+# Similar to diffusers.pipelines.hunyuandit.pipeline_hunyuandit.get_resize_crop_region_for_grid
+def get_resize_crop_region_for_grid(src, tgt_width, tgt_height):
+ tw = tgt_width
+ th = tgt_height
+ h, w = src
+ r = h / w
+ if r > (th / tw):
+ resize_height = th
+ resize_width = int(round(th / h * w))
+ else:
+ resize_width = tw
+ resize_height = int(round(tw / w * h))
+
+ crop_top = int(round((th - resize_height) / 2.0))
+ crop_left = int(round((tw - resize_width) / 2.0))
+
+ return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width)
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class CogVideoXPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
+ r"""
+ Pipeline for text-to-video generation using CogVideoX.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+ text_encoder ([`T5EncoderModel`]):
+ Frozen text-encoder. CogVideoX uses
+ [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel); specifically the
+ [t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
+ tokenizer (`T5Tokenizer`):
+ Tokenizer of class
+ [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
+ transformer ([`CogVideoXTransformer3DModel`]):
+ A text conditioned `CogVideoXTransformer3DModel` to denoise the encoded video latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded video latents.
+ """
+
+ _optional_components = []
+ model_cpu_offload_seq = "text_encoder->transformer->vae"
+
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ ]
+
+ def __init__(
+ self,
+ tokenizer: T5Tokenizer,
+ text_encoder: T5EncoderModel,
+ vae: AutoencoderKLCogVideoX,
+ transformer: CogVideoXTransformer3DModel,
+ scheduler: Union[CogVideoXDDIMScheduler, CogVideoXDPMScheduler],
+ ):
+ super().__init__()
+
+ self.register_modules(
+ tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, scheduler=scheduler
+ )
+ self.vae_scale_factor_spatial = (
+ 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ )
+ self.vae_scale_factor_temporal = (
+ self.vae.config.temporal_compression_ratio if getattr(self, "vae", None) else 4
+ )
+ self.vae_scaling_factor_image = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.7
+
+ self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+ def _get_t5_prompt_embeds(
+ self,
+ prompt: Union[str, List[str]] = None,
+ num_videos_per_prompt: int = 1,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ device = device or self._execution_device
+ dtype = dtype or self.text_encoder.dtype
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ batch_size = len(prompt)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_sequence_length,
+ truncation=True,
+ add_special_tokens=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+ removed_text = self.tokenizer.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because `max_sequence_length` is set to "
+ f" {max_sequence_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = self.text_encoder(text_input_ids.to(device))[0]
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ _, seq_len, _ = prompt_embeds.shape
+ prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+
+ return prompt_embeds
+
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ do_classifier_free_guidance: bool = True,
+ num_videos_per_prompt: int = 1,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
+ Whether to use classifier free guidance or not.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ Number of videos that should be generated per prompt. torch device to place the resulting embeddings on
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ device: (`torch.device`, *optional*):
+ torch device
+ dtype: (`torch.dtype`, *optional*):
+ torch dtype
+ """
+ device = device or self._execution_device
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+
+ negative_prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=negative_prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ return prompt_embeds, negative_prompt_embeds
+
+ def prepare_latents(
+ self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None
+ ):
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ shape = (
+ batch_size,
+ (num_frames - 1) // self.vae_scale_factor_temporal + 1,
+ num_channels_latents,
+ height // self.vae_scale_factor_spatial,
+ width // self.vae_scale_factor_spatial,
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ def decode_latents(self, latents: torch.Tensor) -> torch.Tensor:
+ latents = latents.permute(0, 2, 1, 3, 4) # [batch_size, num_channels, num_frames, height, width]
+ latents = 1 / self.vae_scaling_factor_image * latents
+
+ frames = self.vae.decode(latents).sample
+ return frames
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ # Copied from diffusers.pipelines.latte.pipeline_latte.LattePipeline.check_inputs
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ negative_prompt,
+ callback_on_step_end_tensor_inputs,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ):
+ if height % 8 != 0 or width % 8 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ def fuse_qkv_projections(self) -> None:
+ r"""Enables fused QKV projections."""
+ self.fusing_transformer = True
+ self.transformer.fuse_qkv_projections()
+
+ def unfuse_qkv_projections(self) -> None:
+ r"""Disable QKV projection fusion if enabled."""
+ if not self.fusing_transformer:
+ logger.warning("The Transformer was not initially fused for QKV projections. Doing nothing.")
+ else:
+ self.transformer.unfuse_qkv_projections()
+ self.fusing_transformer = False
+
+ def _prepare_rotary_positional_embeddings(
+ self,
+ height: int,
+ width: int,
+ num_frames: int,
+ device: torch.device,
+ ) -> Tuple[torch.Tensor, torch.Tensor]:
+ grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+ grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+
+ p = self.transformer.config.patch_size
+ p_t = self.transformer.config.patch_size_t
+
+ base_size_width = self.transformer.config.sample_width // p
+ base_size_height = self.transformer.config.sample_height // p
+
+ if p_t is None:
+ # CogVideoX 1.0
+ grid_crops_coords = get_resize_crop_region_for_grid(
+ (grid_height, grid_width), base_size_width, base_size_height
+ )
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=grid_crops_coords,
+ grid_size=(grid_height, grid_width),
+ temporal_size=num_frames,
+ device=device,
+ )
+ else:
+ # CogVideoX 1.5
+ base_num_frames = (num_frames + p_t - 1) // p_t
+
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=None,
+ grid_size=(grid_height, grid_width),
+ temporal_size=base_num_frames,
+ grid_type="slice",
+ max_size=(base_size_height, base_size_width),
+ device=device,
+ )
+
+ return freqs_cos, freqs_sin
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def attention_kwargs(self):
+ return self._attention_kwargs
+
+ @property
+ def current_timestep(self):
+ return self._current_timestep
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_frames: Optional[int] = None,
+ num_inference_steps: int = 50,
+ timesteps: Optional[List[int]] = None,
+ guidance_scale: float = 6,
+ use_dynamic_cfg: bool = False,
+ num_videos_per_prompt: int = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.FloatTensor] = None,
+ prompt_embeds: Optional[torch.FloatTensor] = None,
+ negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+ output_type: str = "pil",
+ return_dict: bool = True,
+ attention_kwargs: Optional[Dict[str, Any]] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ max_sequence_length: int = 226,
+ ) -> Union[CogVideoXPipelineOutput, Tuple]:
+ """
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ height (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The height in pixels of the generated image. This is set to 480 by default for the best results.
+ width (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The width in pixels of the generated image. This is set to 720 by default for the best results.
+ num_frames (`int`, defaults to `48`):
+ Number of frames to generate. Must be divisible by self.vae_scale_factor_temporal. Generated video will
+ contain 1 extra frame because CogVideoX is conditioned with (num_seconds * fps + 1) frames where
+ num_seconds is 6 and fps is 8. However, since videos can be saved at any fps, the only condition that
+ needs to be satisfied is that of divisibility mentioned above.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ guidance_scale (`float`, *optional*, defaults to 7.0):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ The number of videos to generate per prompt.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.FloatTensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
+ of a plain tuple.
+ attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ callback_on_step_end (`Callable`, *optional*):
+ A function that calls at the end of each denoising steps during the inference. The function is called
+ with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+ callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+ `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+ max_sequence_length (`int`, defaults to `226`):
+ Maximum sequence length in encoded prompt. Must be consistent with
+ `self.transformer.config.max_text_seq_length` otherwise may lead to poor results.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipelineOutput`] or `tuple`:
+ [`~pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple`. When returning a tuple, the first element is a list with the generated images.
+ """
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ height = height or self.transformer.config.sample_height * self.vae_scale_factor_spatial
+ width = width or self.transformer.config.sample_width * self.vae_scale_factor_spatial
+ num_frames = num_frames or self.transformer.config.sample_frames
+
+ num_videos_per_prompt = 1
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ height,
+ width,
+ negative_prompt,
+ callback_on_step_end_tensor_inputs,
+ prompt_embeds,
+ negative_prompt_embeds,
+ )
+ self._guidance_scale = guidance_scale
+ self._attention_kwargs = attention_kwargs
+ self._current_timestep = None
+ self._interrupt = False
+
+ # 2. Default call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ do_classifier_free_guidance = guidance_scale > 1.0
+
+ # 3. Encode input prompt
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt,
+ negative_prompt,
+ do_classifier_free_guidance,
+ num_videos_per_prompt=num_videos_per_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+ if do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+
+ # 4. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+ self._num_timesteps = len(timesteps)
+
+ # 5. Prepare latents
+ latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+
+ # For CogVideoX 1.5, the latent frames should be padded to make it divisible by patch_size_t
+ patch_size_t = self.transformer.config.patch_size_t
+ additional_frames = 0
+ if patch_size_t is not None and latent_frames % patch_size_t != 0:
+ additional_frames = patch_size_t - latent_frames % patch_size_t
+ num_frames += additional_frames * self.vae_scale_factor_temporal
+
+ latent_channels = self.transformer.config.in_channels
+ latents = self.prepare_latents(
+ batch_size * num_videos_per_prompt,
+ latent_channels,
+ num_frames,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7. Create rotary embeds if required
+ image_rotary_emb = (
+ self._prepare_rotary_positional_embeddings(height, width, latents.size(1), device)
+ if self.transformer.config.use_rotary_positional_embeddings
+ else None
+ )
+
+ # 8. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ # for DPM-solver++
+ old_pred_original_sample = None
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ self._current_timestep = t
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = t.expand(latent_model_input.shape[0])
+
+ # predict noise model_output
+ with self.transformer.cache_context("cond_uncond"):
+ noise_pred = self.transformer(
+ hidden_states=latent_model_input,
+ encoder_hidden_states=prompt_embeds,
+ timestep=timestep,
+ image_rotary_emb=image_rotary_emb,
+ attention_kwargs=attention_kwargs,
+ return_dict=False,
+ )[0]
+ noise_pred = noise_pred.float()
+
+ # perform guidance
+ if use_dynamic_cfg:
+ self._guidance_scale = 1 + guidance_scale * (
+ (1 - math.cos(math.pi * ((num_inference_steps - t.item()) / num_inference_steps) ** 5.0)) / 2
+ )
+ if do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ if not isinstance(self.scheduler, CogVideoXDPMScheduler):
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+ else:
+ latents, old_pred_original_sample = self.scheduler.step(
+ noise_pred,
+ old_pred_original_sample,
+ t,
+ timesteps[i - 1] if i > 0 else None,
+ latents,
+ **extra_step_kwargs,
+ return_dict=False,
+ )
+ latents = latents.to(prompt_embeds.dtype)
+
+ # call the callback, if provided
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ self._current_timestep = None
+
+ if not output_type == "latent":
+ # Discard any padding frames that were added for CogVideoX 1.5
+ latents = latents[:, additional_frames:]
+ video = self.decode_latents(latents)
+ video = self.video_processor.postprocess_video(video=video, output_type=output_type)
+ else:
+ video = latents
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (video,)
+
+ return CogVideoXPipelineOutput(frames=video)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py
new file mode 100644
index 0000000000000000000000000000000000000000..c1335839f8485ca950affa22c9824b6e08ee381d
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_fun_control.py
@@ -0,0 +1,842 @@
+# Copyright 2025 The CogVideoX team, Tsinghua University & ZhipuAI, Alibaba-PAI and The HuggingFace Team.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import math
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import torch
+from PIL import Image
+from transformers import T5EncoderModel, T5Tokenizer
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...loaders import CogVideoXLoraLoaderMixin
+from ...models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
+from ...models.embeddings import get_3d_rotary_pos_embed
+from ...pipelines.pipeline_utils import DiffusionPipeline
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from .pipeline_output import CogVideoXPipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```python
+ >>> import torch
+ >>> from diffusers import CogVideoXFunControlPipeline, DDIMScheduler
+ >>> from diffusers.utils import export_to_video, load_video
+
+ >>> pipe = CogVideoXFunControlPipeline.from_pretrained(
+ ... "alibaba-pai/CogVideoX-Fun-V1.1-5b-Pose", torch_dtype=torch.bfloat16
+ ... )
+ >>> pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+ >>> pipe.to("cuda")
+
+ >>> control_video = load_video(
+ ... "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/hiker.mp4"
+ ... )
+ >>> prompt = (
+ ... "An astronaut stands triumphantly at the peak of a towering mountain. Panorama of rugged peaks and "
+ ... "valleys. Very futuristic vibe and animated aesthetic. Highlights of purple and golden colors in "
+ ... "the scene. The sky is looks like an animated/cartoonish dream of galaxies, nebulae, stars, planets, "
+ ... "moons, but the remainder of the scene is mostly realistic."
+ ... )
+
+ >>> video = pipe(prompt=prompt, control_video=control_video).frames[0]
+ >>> export_to_video(video, "output.mp4", fps=8)
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.get_resize_crop_region_for_grid
+def get_resize_crop_region_for_grid(src, tgt_width, tgt_height):
+ tw = tgt_width
+ th = tgt_height
+ h, w = src
+ r = h / w
+ if r > (th / tw):
+ resize_height = th
+ resize_width = int(round(th / h * w))
+ else:
+ resize_width = tw
+ resize_height = int(round(tw / w * h))
+
+ crop_top = int(round((th - resize_height) / 2.0))
+ crop_left = int(round((tw - resize_width) / 2.0))
+
+ return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width)
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class CogVideoXFunControlPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
+ r"""
+ Pipeline for controlled text-to-video generation using CogVideoX Fun.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+ text_encoder ([`T5EncoderModel`]):
+ Frozen text-encoder. CogVideoX uses
+ [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel); specifically the
+ [t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
+ tokenizer (`T5Tokenizer`):
+ Tokenizer of class
+ [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
+ transformer ([`CogVideoXTransformer3DModel`]):
+ A text conditioned `CogVideoXTransformer3DModel` to denoise the encoded video latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded video latents.
+ """
+
+ _optional_components = []
+ model_cpu_offload_seq = "text_encoder->vae->transformer->vae"
+
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ ]
+
+ def __init__(
+ self,
+ tokenizer: T5Tokenizer,
+ text_encoder: T5EncoderModel,
+ vae: AutoencoderKLCogVideoX,
+ transformer: CogVideoXTransformer3DModel,
+ scheduler: KarrasDiffusionSchedulers,
+ ):
+ super().__init__()
+
+ self.register_modules(
+ tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, scheduler=scheduler
+ )
+ self.vae_scale_factor_spatial = (
+ 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ )
+ self.vae_scale_factor_temporal = (
+ self.vae.config.temporal_compression_ratio if getattr(self, "vae", None) else 4
+ )
+ self.vae_scaling_factor_image = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.7
+
+ self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._get_t5_prompt_embeds
+ def _get_t5_prompt_embeds(
+ self,
+ prompt: Union[str, List[str]] = None,
+ num_videos_per_prompt: int = 1,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ device = device or self._execution_device
+ dtype = dtype or self.text_encoder.dtype
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ batch_size = len(prompt)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_sequence_length,
+ truncation=True,
+ add_special_tokens=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+ removed_text = self.tokenizer.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because `max_sequence_length` is set to "
+ f" {max_sequence_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = self.text_encoder(text_input_ids.to(device))[0]
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ _, seq_len, _ = prompt_embeds.shape
+ prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+
+ return prompt_embeds
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ do_classifier_free_guidance: bool = True,
+ num_videos_per_prompt: int = 1,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
+ Whether to use classifier free guidance or not.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ Number of videos that should be generated per prompt. torch device to place the resulting embeddings on
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ device: (`torch.device`, *optional*):
+ torch device
+ dtype: (`torch.dtype`, *optional*):
+ torch dtype
+ """
+ device = device or self._execution_device
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+
+ negative_prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=negative_prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ return prompt_embeds, negative_prompt_embeds
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.prepare_latents
+ def prepare_latents(
+ self, batch_size, num_channels_latents, num_frames, height, width, dtype, device, generator, latents=None
+ ):
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ shape = (
+ batch_size,
+ (num_frames - 1) // self.vae_scale_factor_temporal + 1,
+ num_channels_latents,
+ height // self.vae_scale_factor_spatial,
+ width // self.vae_scale_factor_spatial,
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ # Adapted from https://github.com/aigc-apps/CogVideoX-Fun/blob/2a93e5c14e02b2b5921d533fd59fc8c0ed69fb24/cogvideox/pipeline/pipeline_cogvideox_control.py#L366
+ def prepare_control_latents(
+ self, mask: Optional[torch.Tensor] = None, masked_image: Optional[torch.Tensor] = None
+ ) -> Tuple[torch.Tensor, torch.Tensor]:
+ if mask is not None:
+ masks = []
+ for i in range(mask.size(0)):
+ current_mask = mask[i].unsqueeze(0)
+ current_mask = self.vae.encode(current_mask)[0]
+ current_mask = current_mask.mode()
+ masks.append(current_mask)
+ mask = torch.cat(masks, dim=0)
+ mask = mask * self.vae.config.scaling_factor
+
+ if masked_image is not None:
+ mask_pixel_values = []
+ for i in range(masked_image.size(0)):
+ mask_pixel_value = masked_image[i].unsqueeze(0)
+ mask_pixel_value = self.vae.encode(mask_pixel_value)[0]
+ mask_pixel_value = mask_pixel_value.mode()
+ mask_pixel_values.append(mask_pixel_value)
+ masked_image_latents = torch.cat(mask_pixel_values, dim=0)
+ masked_image_latents = masked_image_latents * self.vae.config.scaling_factor
+ else:
+ masked_image_latents = None
+
+ return mask, masked_image_latents
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.decode_latents
+ def decode_latents(self, latents: torch.Tensor) -> torch.Tensor:
+ latents = latents.permute(0, 2, 1, 3, 4) # [batch_size, num_channels, num_frames, height, width]
+ latents = 1 / self.vae_scaling_factor_image * latents
+
+ frames = self.vae.decode(latents).sample
+ return frames
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ negative_prompt,
+ callback_on_step_end_tensor_inputs,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ control_video=None,
+ control_video_latents=None,
+ ):
+ if height % 8 != 0 or width % 8 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if control_video is not None and control_video_latents is not None:
+ raise ValueError(
+ "Cannot pass both `control_video` and `control_video_latents`. Please make sure to pass only one of these parameters."
+ )
+
+ def fuse_qkv_projections(self) -> None:
+ r"""Enables fused QKV projections."""
+ self.fusing_transformer = True
+ self.transformer.fuse_qkv_projections()
+
+ def unfuse_qkv_projections(self) -> None:
+ r"""Disable QKV projection fusion if enabled."""
+ if not self.fusing_transformer:
+ logger.warning("The Transformer was not initially fused for QKV projections. Doing nothing.")
+ else:
+ self.transformer.unfuse_qkv_projections()
+ self.fusing_transformer = False
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._prepare_rotary_positional_embeddings
+ def _prepare_rotary_positional_embeddings(
+ self,
+ height: int,
+ width: int,
+ num_frames: int,
+ device: torch.device,
+ ) -> Tuple[torch.Tensor, torch.Tensor]:
+ grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+ grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+
+ p = self.transformer.config.patch_size
+ p_t = self.transformer.config.patch_size_t
+
+ base_size_width = self.transformer.config.sample_width // p
+ base_size_height = self.transformer.config.sample_height // p
+
+ if p_t is None:
+ # CogVideoX 1.0
+ grid_crops_coords = get_resize_crop_region_for_grid(
+ (grid_height, grid_width), base_size_width, base_size_height
+ )
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=grid_crops_coords,
+ grid_size=(grid_height, grid_width),
+ temporal_size=num_frames,
+ device=device,
+ )
+ else:
+ # CogVideoX 1.5
+ base_num_frames = (num_frames + p_t - 1) // p_t
+
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=None,
+ grid_size=(grid_height, grid_width),
+ temporal_size=base_num_frames,
+ grid_type="slice",
+ max_size=(base_size_height, base_size_width),
+ device=device,
+ )
+
+ return freqs_cos, freqs_sin
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def attention_kwargs(self):
+ return self._attention_kwargs
+
+ @property
+ def current_timestep(self):
+ return self._current_timestep
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ control_video: Optional[List[Image.Image]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 50,
+ timesteps: Optional[List[int]] = None,
+ guidance_scale: float = 6,
+ use_dynamic_cfg: bool = False,
+ num_videos_per_prompt: int = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ control_video_latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ output_type: str = "pil",
+ return_dict: bool = True,
+ attention_kwargs: Optional[Dict[str, Any]] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ max_sequence_length: int = 226,
+ ) -> Union[CogVideoXPipelineOutput, Tuple]:
+ """
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ control_video (`List[PIL.Image.Image]`):
+ The control video to condition the generation on. Must be a list of images/frames of the video. If not
+ provided, `control_video_latents` must be provided.
+ height (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The height in pixels of the generated image. This is set to 480 by default for the best results.
+ width (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The width in pixels of the generated image. This is set to 720 by default for the best results.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ guidance_scale (`float`, *optional*, defaults to 6.0):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ The number of videos to generate per prompt.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for video
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ control_video_latents (`torch.Tensor`, *optional*):
+ Pre-generated control latents, sampled from a Gaussian distribution, to be used as inputs for
+ controlled video generation. If not provided, `control_video` must be provided.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
+ of a plain tuple.
+ attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ callback_on_step_end (`Callable`, *optional*):
+ A function that calls at the end of each denoising steps during the inference. The function is called
+ with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+ callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+ `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+ max_sequence_length (`int`, defaults to `226`):
+ Maximum sequence length in encoded prompt. Must be consistent with
+ `self.transformer.config.max_text_seq_length` otherwise may lead to poor results.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipelineOutput`] or `tuple`:
+ [`~pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple`. When returning a tuple, the first element is a list with the generated images.
+ """
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ if control_video is not None and isinstance(control_video[0], Image.Image):
+ control_video = [control_video]
+
+ height = height or self.transformer.config.sample_height * self.vae_scale_factor_spatial
+ width = width or self.transformer.config.sample_width * self.vae_scale_factor_spatial
+ num_frames = len(control_video[0]) if control_video is not None else control_video_latents.size(2)
+
+ num_videos_per_prompt = 1
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ height,
+ width,
+ negative_prompt,
+ callback_on_step_end_tensor_inputs,
+ prompt_embeds,
+ negative_prompt_embeds,
+ control_video,
+ control_video_latents,
+ )
+ self._guidance_scale = guidance_scale
+ self._attention_kwargs = attention_kwargs
+ self._current_timestep = None
+ self._interrupt = False
+
+ # 2. Default call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ do_classifier_free_guidance = guidance_scale > 1.0
+
+ # 3. Encode input prompt
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt,
+ negative_prompt,
+ do_classifier_free_guidance,
+ num_videos_per_prompt=num_videos_per_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+ if do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+
+ # 4. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+ self._num_timesteps = len(timesteps)
+
+ # 5. Prepare latents
+ latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+
+ # For CogVideoX 1.5, the latent frames should be padded to make it divisible by patch_size_t
+ patch_size_t = self.transformer.config.patch_size_t
+ if patch_size_t is not None and latent_frames % patch_size_t != 0:
+ raise ValueError(
+ f"The number of latent frames must be divisible by `{patch_size_t=}` but the given video "
+ f"contains {latent_frames=}, which is not divisible."
+ )
+
+ latent_channels = self.transformer.config.in_channels // 2
+ latents = self.prepare_latents(
+ batch_size * num_videos_per_prompt,
+ latent_channels,
+ num_frames,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ if control_video_latents is None:
+ control_video = self.video_processor.preprocess_video(control_video, height=height, width=width)
+ control_video = control_video.to(device=device, dtype=prompt_embeds.dtype)
+
+ _, control_video_latents = self.prepare_control_latents(None, control_video)
+ control_video_latents = control_video_latents.permute(0, 2, 1, 3, 4)
+
+ # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7. Create rotary embeds if required
+ image_rotary_emb = (
+ self._prepare_rotary_positional_embeddings(height, width, latents.size(1), device)
+ if self.transformer.config.use_rotary_positional_embeddings
+ else None
+ )
+
+ # 8. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ # for DPM-solver++
+ old_pred_original_sample = None
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ self._current_timestep = t
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ latent_control_input = (
+ torch.cat([control_video_latents] * 2) if do_classifier_free_guidance else control_video_latents
+ )
+ latent_model_input = torch.cat([latent_model_input, latent_control_input], dim=2)
+
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = t.expand(latent_model_input.shape[0])
+
+ # predict noise model_output
+ with self.transformer.cache_context("cond_uncond"):
+ noise_pred = self.transformer(
+ hidden_states=latent_model_input,
+ encoder_hidden_states=prompt_embeds,
+ timestep=timestep,
+ image_rotary_emb=image_rotary_emb,
+ attention_kwargs=attention_kwargs,
+ return_dict=False,
+ )[0]
+ noise_pred = noise_pred.float()
+
+ # perform guidance
+ if use_dynamic_cfg:
+ self._guidance_scale = 1 + guidance_scale * (
+ (1 - math.cos(math.pi * ((num_inference_steps - t.item()) / num_inference_steps) ** 5.0)) / 2
+ )
+ if do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+ latents = latents.to(prompt_embeds.dtype)
+
+ # call the callback, if provided
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ self._current_timestep = None
+
+ if not output_type == "latent":
+ video = self.decode_latents(latents)
+ video = self.video_processor.postprocess_video(video=video, output_type=output_type)
+ else:
+ video = latents
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (video,)
+
+ return CogVideoXPipelineOutput(frames=video)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py
new file mode 100644
index 0000000000000000000000000000000000000000..225240927fad770f321ae883a5fe9265122b3ca6
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_image2video.py
@@ -0,0 +1,903 @@
+# Copyright 2025 The CogVideoX team, Tsinghua University & ZhipuAI and The HuggingFace Team.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import math
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import PIL
+import torch
+from transformers import T5EncoderModel, T5Tokenizer
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput
+from ...loaders import CogVideoXLoraLoaderMixin
+from ...models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
+from ...models.embeddings import get_3d_rotary_pos_embed
+from ...pipelines.pipeline_utils import DiffusionPipeline
+from ...schedulers import CogVideoXDDIMScheduler, CogVideoXDPMScheduler
+from ...utils import (
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+)
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from .pipeline_output import CogVideoXPipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> import torch
+ >>> from diffusers import CogVideoXImageToVideoPipeline
+ >>> from diffusers.utils import export_to_video, load_image
+
+ >>> pipe = CogVideoXImageToVideoPipeline.from_pretrained("THUDM/CogVideoX-5b-I2V", torch_dtype=torch.bfloat16)
+ >>> pipe.to("cuda")
+
+ >>> prompt = "An astronaut hatching from an egg, on the surface of the moon, the darkness and depth of space realised in the background. High quality, ultrarealistic detail and breath-taking movie-like camera shot."
+ >>> image = load_image(
+ ... "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/astronaut.jpg"
+ ... )
+ >>> video = pipe(image, prompt, use_dynamic_cfg=True)
+ >>> export_to_video(video.frames[0], "output.mp4", fps=8)
+ ```
+"""
+
+
+# Similar to diffusers.pipelines.hunyuandit.pipeline_hunyuandit.get_resize_crop_region_for_grid
+def get_resize_crop_region_for_grid(src, tgt_width, tgt_height):
+ tw = tgt_width
+ th = tgt_height
+ h, w = src
+ r = h / w
+ if r > (th / tw):
+ resize_height = th
+ resize_width = int(round(th / h * w))
+ else:
+ resize_width = tw
+ resize_height = int(round(tw / w * h))
+
+ crop_top = int(round((th - resize_height) / 2.0))
+ crop_left = int(round((tw - resize_width) / 2.0))
+
+ return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width)
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class CogVideoXImageToVideoPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
+ r"""
+ Pipeline for image-to-video generation using CogVideoX.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+ text_encoder ([`T5EncoderModel`]):
+ Frozen text-encoder. CogVideoX uses
+ [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel); specifically the
+ [t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
+ tokenizer (`T5Tokenizer`):
+ Tokenizer of class
+ [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
+ transformer ([`CogVideoXTransformer3DModel`]):
+ A text conditioned `CogVideoXTransformer3DModel` to denoise the encoded video latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded video latents.
+ """
+
+ _optional_components = []
+ model_cpu_offload_seq = "text_encoder->transformer->vae"
+
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ ]
+
+ def __init__(
+ self,
+ tokenizer: T5Tokenizer,
+ text_encoder: T5EncoderModel,
+ vae: AutoencoderKLCogVideoX,
+ transformer: CogVideoXTransformer3DModel,
+ scheduler: Union[CogVideoXDDIMScheduler, CogVideoXDPMScheduler],
+ ):
+ super().__init__()
+
+ self.register_modules(
+ tokenizer=tokenizer,
+ text_encoder=text_encoder,
+ vae=vae,
+ transformer=transformer,
+ scheduler=scheduler,
+ )
+ self.vae_scale_factor_spatial = (
+ 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ )
+ self.vae_scale_factor_temporal = (
+ self.vae.config.temporal_compression_ratio if getattr(self, "vae", None) else 4
+ )
+ self.vae_scaling_factor_image = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.7
+
+ self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._get_t5_prompt_embeds
+ def _get_t5_prompt_embeds(
+ self,
+ prompt: Union[str, List[str]] = None,
+ num_videos_per_prompt: int = 1,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ device = device or self._execution_device
+ dtype = dtype or self.text_encoder.dtype
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ batch_size = len(prompt)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_sequence_length,
+ truncation=True,
+ add_special_tokens=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+ removed_text = self.tokenizer.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because `max_sequence_length` is set to "
+ f" {max_sequence_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = self.text_encoder(text_input_ids.to(device))[0]
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ _, seq_len, _ = prompt_embeds.shape
+ prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+
+ return prompt_embeds
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ do_classifier_free_guidance: bool = True,
+ num_videos_per_prompt: int = 1,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
+ Whether to use classifier free guidance or not.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ Number of videos that should be generated per prompt. torch device to place the resulting embeddings on
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ device: (`torch.device`, *optional*):
+ torch device
+ dtype: (`torch.dtype`, *optional*):
+ torch dtype
+ """
+ device = device or self._execution_device
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+
+ negative_prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=negative_prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ return prompt_embeds, negative_prompt_embeds
+
+ def prepare_latents(
+ self,
+ image: torch.Tensor,
+ batch_size: int = 1,
+ num_channels_latents: int = 16,
+ num_frames: int = 13,
+ height: int = 60,
+ width: int = 90,
+ dtype: Optional[torch.dtype] = None,
+ device: Optional[torch.device] = None,
+ generator: Optional[torch.Generator] = None,
+ latents: Optional[torch.Tensor] = None,
+ ):
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ num_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+ shape = (
+ batch_size,
+ num_frames,
+ num_channels_latents,
+ height // self.vae_scale_factor_spatial,
+ width // self.vae_scale_factor_spatial,
+ )
+
+ # For CogVideoX1.5, the latent should add 1 for padding (Not use)
+ if self.transformer.config.patch_size_t is not None:
+ shape = shape[:1] + (shape[1] + shape[1] % self.transformer.config.patch_size_t,) + shape[2:]
+
+ image = image.unsqueeze(2) # [B, C, F, H, W]
+
+ if isinstance(generator, list):
+ image_latents = [
+ retrieve_latents(self.vae.encode(image[i].unsqueeze(0)), generator[i]) for i in range(batch_size)
+ ]
+ else:
+ image_latents = [retrieve_latents(self.vae.encode(img.unsqueeze(0)), generator) for img in image]
+
+ image_latents = torch.cat(image_latents, dim=0).to(dtype).permute(0, 2, 1, 3, 4) # [B, F, C, H, W]
+
+ if not self.vae.config.invert_scale_latents:
+ image_latents = self.vae_scaling_factor_image * image_latents
+ else:
+ # This is awkward but required because the CogVideoX team forgot to multiply the
+ # scaling factor during training :)
+ image_latents = 1 / self.vae_scaling_factor_image * image_latents
+
+ padding_shape = (
+ batch_size,
+ num_frames - 1,
+ num_channels_latents,
+ height // self.vae_scale_factor_spatial,
+ width // self.vae_scale_factor_spatial,
+ )
+
+ latent_padding = torch.zeros(padding_shape, device=device, dtype=dtype)
+ image_latents = torch.cat([image_latents, latent_padding], dim=1)
+
+ # Select the first frame along the second dimension
+ if self.transformer.config.patch_size_t is not None:
+ first_frame = image_latents[:, : image_latents.size(1) % self.transformer.config.patch_size_t, ...]
+ image_latents = torch.cat([first_frame, image_latents], dim=1)
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents, image_latents
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.decode_latents
+ def decode_latents(self, latents: torch.Tensor) -> torch.Tensor:
+ latents = latents.permute(0, 2, 1, 3, 4) # [batch_size, num_channels, num_frames, height, width]
+ latents = 1 / self.vae_scaling_factor_image * latents
+
+ frames = self.vae.decode(latents).sample
+ return frames
+
+ # Copied from diffusers.pipelines.animatediff.pipeline_animatediff_video2video.AnimateDiffVideoToVideoPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, timesteps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+ t_start = max(num_inference_steps - init_timestep, 0)
+ timesteps = timesteps[t_start * self.scheduler.order :]
+
+ return timesteps, num_inference_steps - t_start
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ image,
+ prompt,
+ height,
+ width,
+ negative_prompt,
+ callback_on_step_end_tensor_inputs,
+ latents=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ):
+ if (
+ not isinstance(image, torch.Tensor)
+ and not isinstance(image, PIL.Image.Image)
+ and not isinstance(image, list)
+ ):
+ raise ValueError(
+ "`image` has to be of type `torch.Tensor` or `PIL.Image.Image` or `List[PIL.Image.Image]` but is"
+ f" {type(image)}"
+ )
+
+ if height % 8 != 0 or width % 8 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.fuse_qkv_projections
+ def fuse_qkv_projections(self) -> None:
+ r"""Enables fused QKV projections."""
+ self.fusing_transformer = True
+ self.transformer.fuse_qkv_projections()
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.unfuse_qkv_projections
+ def unfuse_qkv_projections(self) -> None:
+ r"""Disable QKV projection fusion if enabled."""
+ if not self.fusing_transformer:
+ logger.warning("The Transformer was not initially fused for QKV projections. Doing nothing.")
+ else:
+ self.transformer.unfuse_qkv_projections()
+ self.fusing_transformer = False
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._prepare_rotary_positional_embeddings
+ def _prepare_rotary_positional_embeddings(
+ self,
+ height: int,
+ width: int,
+ num_frames: int,
+ device: torch.device,
+ ) -> Tuple[torch.Tensor, torch.Tensor]:
+ grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+ grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+
+ p = self.transformer.config.patch_size
+ p_t = self.transformer.config.patch_size_t
+
+ base_size_width = self.transformer.config.sample_width // p
+ base_size_height = self.transformer.config.sample_height // p
+
+ if p_t is None:
+ # CogVideoX 1.0
+ grid_crops_coords = get_resize_crop_region_for_grid(
+ (grid_height, grid_width), base_size_width, base_size_height
+ )
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=grid_crops_coords,
+ grid_size=(grid_height, grid_width),
+ temporal_size=num_frames,
+ device=device,
+ )
+ else:
+ # CogVideoX 1.5
+ base_num_frames = (num_frames + p_t - 1) // p_t
+
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=None,
+ grid_size=(grid_height, grid_width),
+ temporal_size=base_num_frames,
+ grid_type="slice",
+ max_size=(base_size_height, base_size_width),
+ device=device,
+ )
+
+ return freqs_cos, freqs_sin
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def attention_kwargs(self):
+ return self._attention_kwargs
+
+ @property
+ def current_timestep(self):
+ return self._current_timestep
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ image: PipelineImageInput,
+ prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_frames: int = 49,
+ num_inference_steps: int = 50,
+ timesteps: Optional[List[int]] = None,
+ guidance_scale: float = 6,
+ use_dynamic_cfg: bool = False,
+ num_videos_per_prompt: int = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.FloatTensor] = None,
+ prompt_embeds: Optional[torch.FloatTensor] = None,
+ negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+ output_type: str = "pil",
+ return_dict: bool = True,
+ attention_kwargs: Optional[Dict[str, Any]] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ max_sequence_length: int = 226,
+ ) -> Union[CogVideoXPipelineOutput, Tuple]:
+ """
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ image (`PipelineImageInput`):
+ The input image to condition the generation on. Must be an image, a list of images or a `torch.Tensor`.
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ height (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The height in pixels of the generated image. This is set to 480 by default for the best results.
+ width (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The width in pixels of the generated image. This is set to 720 by default for the best results.
+ num_frames (`int`, defaults to `48`):
+ Number of frames to generate. Must be divisible by self.vae_scale_factor_temporal. Generated video will
+ contain 1 extra frame because CogVideoX is conditioned with (num_seconds * fps + 1) frames where
+ num_seconds is 6 and fps is 8. However, since videos can be saved at any fps, the only condition that
+ needs to be satisfied is that of divisibility mentioned above.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ guidance_scale (`float`, *optional*, defaults to 7.0):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ The number of videos to generate per prompt.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.FloatTensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
+ of a plain tuple.
+ attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ callback_on_step_end (`Callable`, *optional*):
+ A function that calls at the end of each denoising steps during the inference. The function is called
+ with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+ callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+ `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+ max_sequence_length (`int`, defaults to `226`):
+ Maximum sequence length in encoded prompt. Must be consistent with
+ `self.transformer.config.max_text_seq_length` otherwise may lead to poor results.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.cogvideo.pipeline_output.CogVideoXPipelineOutput`] or `tuple`:
+ [`~pipelines.cogvideo.pipeline_output.CogVideoXPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple`. When returning a tuple, the first element is a list with the generated images.
+ """
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ height = height or self.transformer.config.sample_height * self.vae_scale_factor_spatial
+ width = width or self.transformer.config.sample_width * self.vae_scale_factor_spatial
+ num_frames = num_frames or self.transformer.config.sample_frames
+
+ num_videos_per_prompt = 1
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ image=image,
+ prompt=prompt,
+ height=height,
+ width=width,
+ negative_prompt=negative_prompt,
+ callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+ latents=latents,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ )
+ self._guidance_scale = guidance_scale
+ self._current_timestep = None
+ self._attention_kwargs = attention_kwargs
+ self._interrupt = False
+
+ # 2. Default call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ do_classifier_free_guidance = guidance_scale > 1.0
+
+ # 3. Encode input prompt
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt=prompt,
+ negative_prompt=negative_prompt,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ num_videos_per_prompt=num_videos_per_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+ if do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+
+ # 4. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+ self._num_timesteps = len(timesteps)
+
+ # 5. Prepare latents
+ latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+
+ # For CogVideoX 1.5, the latent frames should be padded to make it divisible by patch_size_t
+ patch_size_t = self.transformer.config.patch_size_t
+ additional_frames = 0
+ if patch_size_t is not None and latent_frames % patch_size_t != 0:
+ additional_frames = patch_size_t - latent_frames % patch_size_t
+ num_frames += additional_frames * self.vae_scale_factor_temporal
+
+ image = self.video_processor.preprocess(image, height=height, width=width).to(
+ device, dtype=prompt_embeds.dtype
+ )
+
+ latent_channels = self.transformer.config.in_channels // 2
+ latents, image_latents = self.prepare_latents(
+ image,
+ batch_size * num_videos_per_prompt,
+ latent_channels,
+ num_frames,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7. Create rotary embeds if required
+ image_rotary_emb = (
+ self._prepare_rotary_positional_embeddings(height, width, latents.size(1), device)
+ if self.transformer.config.use_rotary_positional_embeddings
+ else None
+ )
+
+ # 8. Create ofs embeds if required
+ ofs_emb = None if self.transformer.config.ofs_embed_dim is None else latents.new_full((1,), fill_value=2.0)
+
+ # 8. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ # for DPM-solver++
+ old_pred_original_sample = None
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ self._current_timestep = t
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ latent_image_input = torch.cat([image_latents] * 2) if do_classifier_free_guidance else image_latents
+ latent_model_input = torch.cat([latent_model_input, latent_image_input], dim=2)
+
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = t.expand(latent_model_input.shape[0])
+
+ # predict noise model_output
+ with self.transformer.cache_context("cond_uncond"):
+ noise_pred = self.transformer(
+ hidden_states=latent_model_input,
+ encoder_hidden_states=prompt_embeds,
+ timestep=timestep,
+ ofs=ofs_emb,
+ image_rotary_emb=image_rotary_emb,
+ attention_kwargs=attention_kwargs,
+ return_dict=False,
+ )[0]
+ noise_pred = noise_pred.float()
+
+ # perform guidance
+ if use_dynamic_cfg:
+ self._guidance_scale = 1 + guidance_scale * (
+ (1 - math.cos(math.pi * ((num_inference_steps - t.item()) / num_inference_steps) ** 5.0)) / 2
+ )
+ if do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ if not isinstance(self.scheduler, CogVideoXDPMScheduler):
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+ else:
+ latents, old_pred_original_sample = self.scheduler.step(
+ noise_pred,
+ old_pred_original_sample,
+ t,
+ timesteps[i - 1] if i > 0 else None,
+ latents,
+ **extra_step_kwargs,
+ return_dict=False,
+ )
+ latents = latents.to(prompt_embeds.dtype)
+
+ # call the callback, if provided
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ self._current_timestep = None
+
+ if not output_type == "latent":
+ # Discard any padding frames that were added for CogVideoX 1.5
+ latents = latents[:, additional_frames:]
+ video = self.decode_latents(latents)
+ video = self.video_processor.postprocess_video(video=video, output_type=output_type)
+ else:
+ video = latents
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (video,)
+
+ return CogVideoXPipelineOutput(frames=video)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py
new file mode 100644
index 0000000000000000000000000000000000000000..897dc6d1b70aab26c2fcbca378addc28cb068ae7
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_cogvideox_video2video.py
@@ -0,0 +1,868 @@
+# Copyright 2025 The CogVideoX team, Tsinghua University & ZhipuAI and The HuggingFace Team.
+# All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+import math
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import torch
+from PIL import Image
+from transformers import T5EncoderModel, T5Tokenizer
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...loaders import CogVideoXLoraLoaderMixin
+from ...models import AutoencoderKLCogVideoX, CogVideoXTransformer3DModel
+from ...models.embeddings import get_3d_rotary_pos_embed
+from ...pipelines.pipeline_utils import DiffusionPipeline
+from ...schedulers import CogVideoXDDIMScheduler, CogVideoXDPMScheduler
+from ...utils import is_torch_xla_available, logging, replace_example_docstring
+from ...utils.torch_utils import randn_tensor
+from ...video_processor import VideoProcessor
+from .pipeline_output import CogVideoXPipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```python
+ >>> import torch
+ >>> from diffusers import CogVideoXDPMScheduler, CogVideoXVideoToVideoPipeline
+ >>> from diffusers.utils import export_to_video, load_video
+
+ >>> # Models: "THUDM/CogVideoX-2b" or "THUDM/CogVideoX-5b"
+ >>> pipe = CogVideoXVideoToVideoPipeline.from_pretrained("THUDM/CogVideoX-5b", torch_dtype=torch.bfloat16)
+ >>> pipe.to("cuda")
+ >>> pipe.scheduler = CogVideoXDPMScheduler.from_config(pipe.scheduler.config)
+
+ >>> input_video = load_video(
+ ... "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/hiker.mp4"
+ ... )
+ >>> prompt = (
+ ... "An astronaut stands triumphantly at the peak of a towering mountain. Panorama of rugged peaks and "
+ ... "valleys. Very futuristic vibe and animated aesthetic. Highlights of purple and golden colors in "
+ ... "the scene. The sky is looks like an animated/cartoonish dream of galaxies, nebulae, stars, planets, "
+ ... "moons, but the remainder of the scene is mostly realistic."
+ ... )
+
+ >>> video = pipe(
+ ... video=input_video, prompt=prompt, strength=0.8, guidance_scale=6, num_inference_steps=50
+ ... ).frames[0]
+ >>> export_to_video(video, "output.mp4", fps=8)
+ ```
+"""
+
+
+# Similar to diffusers.pipelines.hunyuandit.pipeline_hunyuandit.get_resize_crop_region_for_grid
+def get_resize_crop_region_for_grid(src, tgt_width, tgt_height):
+ tw = tgt_width
+ th = tgt_height
+ h, w = src
+ r = h / w
+ if r > (th / tw):
+ resize_height = th
+ resize_width = int(round(th / h * w))
+ else:
+ resize_width = tw
+ resize_height = int(round(tw / w * h))
+
+ crop_top = int(round((th - resize_height) / 2.0))
+ crop_left = int(round((tw - resize_width) / 2.0))
+
+ return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width)
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class CogVideoXVideoToVideoPipeline(DiffusionPipeline, CogVideoXLoraLoaderMixin):
+ r"""
+ Pipeline for video-to-video generation using CogVideoX.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode videos to and from latent representations.
+ text_encoder ([`T5EncoderModel`]):
+ Frozen text-encoder. CogVideoX uses
+ [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel); specifically the
+ [t5-v1_1-xxl](https://huggingface.co/PixArt-alpha/PixArt-alpha/tree/main/t5-v1_1-xxl) variant.
+ tokenizer (`T5Tokenizer`):
+ Tokenizer of class
+ [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
+ transformer ([`CogVideoXTransformer3DModel`]):
+ A text conditioned `CogVideoXTransformer3DModel` to denoise the encoded video latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded video latents.
+ """
+
+ _optional_components = []
+ model_cpu_offload_seq = "text_encoder->transformer->vae"
+
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ ]
+
+ def __init__(
+ self,
+ tokenizer: T5Tokenizer,
+ text_encoder: T5EncoderModel,
+ vae: AutoencoderKLCogVideoX,
+ transformer: CogVideoXTransformer3DModel,
+ scheduler: Union[CogVideoXDDIMScheduler, CogVideoXDPMScheduler],
+ ):
+ super().__init__()
+
+ self.register_modules(
+ tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, scheduler=scheduler
+ )
+
+ self.vae_scale_factor_spatial = (
+ 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ )
+ self.vae_scale_factor_temporal = (
+ self.vae.config.temporal_compression_ratio if getattr(self, "vae", None) else 4
+ )
+ self.vae_scaling_factor_image = self.vae.config.scaling_factor if getattr(self, "vae", None) else 0.7
+
+ self.video_processor = VideoProcessor(vae_scale_factor=self.vae_scale_factor_spatial)
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._get_t5_prompt_embeds
+ def _get_t5_prompt_embeds(
+ self,
+ prompt: Union[str, List[str]] = None,
+ num_videos_per_prompt: int = 1,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ device = device or self._execution_device
+ dtype = dtype or self.text_encoder.dtype
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ batch_size = len(prompt)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_sequence_length,
+ truncation=True,
+ add_special_tokens=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids):
+ removed_text = self.tokenizer.batch_decode(untruncated_ids[:, max_sequence_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because `max_sequence_length` is set to "
+ f" {max_sequence_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = self.text_encoder(text_input_ids.to(device))[0]
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ _, seq_len, _ = prompt_embeds.shape
+ prompt_embeds = prompt_embeds.repeat(1, num_videos_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(batch_size * num_videos_per_prompt, seq_len, -1)
+
+ return prompt_embeds
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ do_classifier_free_guidance: bool = True,
+ num_videos_per_prompt: int = 1,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ max_sequence_length: int = 226,
+ device: Optional[torch.device] = None,
+ dtype: Optional[torch.dtype] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
+ Whether to use classifier free guidance or not.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ Number of videos that should be generated per prompt. torch device to place the resulting embeddings on
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ device: (`torch.device`, *optional*):
+ torch device
+ dtype: (`torch.dtype`, *optional*):
+ torch dtype
+ """
+ device = device or self._execution_device
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+
+ negative_prompt_embeds = self._get_t5_prompt_embeds(
+ prompt=negative_prompt,
+ num_videos_per_prompt=num_videos_per_prompt,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ dtype=dtype,
+ )
+
+ return prompt_embeds, negative_prompt_embeds
+
+ def prepare_latents(
+ self,
+ video: Optional[torch.Tensor] = None,
+ batch_size: int = 1,
+ num_channels_latents: int = 16,
+ height: int = 60,
+ width: int = 90,
+ dtype: Optional[torch.dtype] = None,
+ device: Optional[torch.device] = None,
+ generator: Optional[torch.Generator] = None,
+ latents: Optional[torch.Tensor] = None,
+ timestep: Optional[torch.Tensor] = None,
+ ):
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ num_frames = (video.size(2) - 1) // self.vae_scale_factor_temporal + 1 if latents is None else latents.size(1)
+
+ shape = (
+ batch_size,
+ num_frames,
+ num_channels_latents,
+ height // self.vae_scale_factor_spatial,
+ width // self.vae_scale_factor_spatial,
+ )
+
+ if latents is None:
+ if isinstance(generator, list):
+ init_latents = [
+ retrieve_latents(self.vae.encode(video[i].unsqueeze(0)), generator[i]) for i in range(batch_size)
+ ]
+ else:
+ init_latents = [retrieve_latents(self.vae.encode(vid.unsqueeze(0)), generator) for vid in video]
+
+ init_latents = torch.cat(init_latents, dim=0).to(dtype).permute(0, 2, 1, 3, 4) # [B, F, C, H, W]
+ init_latents = self.vae_scaling_factor_image * init_latents
+
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ latents = self.scheduler.add_noise(init_latents, noise, timestep)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.decode_latents
+ def decode_latents(self, latents: torch.Tensor) -> torch.Tensor:
+ latents = latents.permute(0, 2, 1, 3, 4) # [batch_size, num_channels, num_frames, height, width]
+ latents = 1 / self.vae_scaling_factor_image * latents
+
+ frames = self.vae.decode(latents).sample
+ return frames
+
+ # Copied from diffusers.pipelines.animatediff.pipeline_animatediff_video2video.AnimateDiffVideoToVideoPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, timesteps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+ t_start = max(num_inference_steps - init_timestep, 0)
+ timesteps = timesteps[t_start * self.scheduler.order :]
+
+ return timesteps, num_inference_steps - t_start
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ strength,
+ negative_prompt,
+ callback_on_step_end_tensor_inputs,
+ video=None,
+ latents=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ):
+ if height % 8 != 0 or width % 8 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if video is not None and latents is not None:
+ raise ValueError("Only one of `video` or `latents` should be provided")
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.fuse_qkv_projections
+ def fuse_qkv_projections(self) -> None:
+ r"""Enables fused QKV projections."""
+ self.fusing_transformer = True
+ self.transformer.fuse_qkv_projections()
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline.unfuse_qkv_projections
+ def unfuse_qkv_projections(self) -> None:
+ r"""Disable QKV projection fusion if enabled."""
+ if not self.fusing_transformer:
+ logger.warning("The Transformer was not initially fused for QKV projections. Doing nothing.")
+ else:
+ self.transformer.unfuse_qkv_projections()
+ self.fusing_transformer = False
+
+ # Copied from diffusers.pipelines.cogvideo.pipeline_cogvideox.CogVideoXPipeline._prepare_rotary_positional_embeddings
+ def _prepare_rotary_positional_embeddings(
+ self,
+ height: int,
+ width: int,
+ num_frames: int,
+ device: torch.device,
+ ) -> Tuple[torch.Tensor, torch.Tensor]:
+ grid_height = height // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+ grid_width = width // (self.vae_scale_factor_spatial * self.transformer.config.patch_size)
+
+ p = self.transformer.config.patch_size
+ p_t = self.transformer.config.patch_size_t
+
+ base_size_width = self.transformer.config.sample_width // p
+ base_size_height = self.transformer.config.sample_height // p
+
+ if p_t is None:
+ # CogVideoX 1.0
+ grid_crops_coords = get_resize_crop_region_for_grid(
+ (grid_height, grid_width), base_size_width, base_size_height
+ )
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=grid_crops_coords,
+ grid_size=(grid_height, grid_width),
+ temporal_size=num_frames,
+ device=device,
+ )
+ else:
+ # CogVideoX 1.5
+ base_num_frames = (num_frames + p_t - 1) // p_t
+
+ freqs_cos, freqs_sin = get_3d_rotary_pos_embed(
+ embed_dim=self.transformer.config.attention_head_dim,
+ crops_coords=None,
+ grid_size=(grid_height, grid_width),
+ temporal_size=base_num_frames,
+ grid_type="slice",
+ max_size=(base_size_height, base_size_width),
+ device=device,
+ )
+
+ return freqs_cos, freqs_sin
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def attention_kwargs(self):
+ return self._attention_kwargs
+
+ @property
+ def current_timestep(self):
+ return self._current_timestep
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ video: List[Image.Image] = None,
+ prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 50,
+ timesteps: Optional[List[int]] = None,
+ strength: float = 0.8,
+ guidance_scale: float = 6,
+ use_dynamic_cfg: bool = False,
+ num_videos_per_prompt: int = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.FloatTensor] = None,
+ prompt_embeds: Optional[torch.FloatTensor] = None,
+ negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+ output_type: str = "pil",
+ return_dict: bool = True,
+ attention_kwargs: Optional[Dict[str, Any]] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ max_sequence_length: int = 226,
+ ) -> Union[CogVideoXPipelineOutput, Tuple]:
+ """
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ video (`List[PIL.Image.Image]`):
+ The input video to condition the generation on. Must be a list of images/frames of the video.
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ height (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The height in pixels of the generated image. This is set to 480 by default for the best results.
+ width (`int`, *optional*, defaults to self.transformer.config.sample_height * self.vae_scale_factor_spatial):
+ The width in pixels of the generated image. This is set to 720 by default for the best results.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ strength (`float`, *optional*, defaults to 0.8):
+ Higher strength leads to more differences between original video and generated video.
+ guidance_scale (`float`, *optional*, defaults to 7.0):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ num_videos_per_prompt (`int`, *optional*, defaults to 1):
+ The number of videos to generate per prompt.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.FloatTensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
+ of a plain tuple.
+ attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ callback_on_step_end (`Callable`, *optional*):
+ A function that calls at the end of each denoising steps during the inference. The function is called
+ with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+ callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+ `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+ max_sequence_length (`int`, defaults to `226`):
+ Maximum sequence length in encoded prompt. Must be consistent with
+ `self.transformer.config.max_text_seq_length` otherwise may lead to poor results.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.cogvideo.pipeline_output.CogVideoXPipelineOutput`] or `tuple`:
+ [`~pipelines.cogvideo.pipeline_output.CogVideoXPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple`. When returning a tuple, the first element is a list with the generated images.
+ """
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ height = height or self.transformer.config.sample_height * self.vae_scale_factor_spatial
+ width = width or self.transformer.config.sample_width * self.vae_scale_factor_spatial
+ num_frames = len(video) if latents is None else latents.size(1)
+
+ num_videos_per_prompt = 1
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt=prompt,
+ height=height,
+ width=width,
+ strength=strength,
+ negative_prompt=negative_prompt,
+ callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+ video=video,
+ latents=latents,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ )
+ self._guidance_scale = guidance_scale
+ self._attention_kwargs = attention_kwargs
+ self._current_timestep = None
+ self._interrupt = False
+
+ # 2. Default call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ do_classifier_free_guidance = guidance_scale > 1.0
+
+ # 3. Encode input prompt
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt,
+ negative_prompt,
+ do_classifier_free_guidance,
+ num_videos_per_prompt=num_videos_per_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ max_sequence_length=max_sequence_length,
+ device=device,
+ )
+ if do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+
+ # 4. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+ timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, timesteps, strength, device)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_videos_per_prompt)
+ self._num_timesteps = len(timesteps)
+
+ # 5. Prepare latents
+ latent_frames = (num_frames - 1) // self.vae_scale_factor_temporal + 1
+
+ # For CogVideoX 1.5, the latent frames should be padded to make it divisible by patch_size_t
+ patch_size_t = self.transformer.config.patch_size_t
+ if patch_size_t is not None and latent_frames % patch_size_t != 0:
+ raise ValueError(
+ f"The number of latent frames must be divisible by `{patch_size_t=}` but the given video "
+ f"contains {latent_frames=}, which is not divisible."
+ )
+
+ if latents is None:
+ video = self.video_processor.preprocess_video(video, height=height, width=width)
+ video = video.to(device=device, dtype=prompt_embeds.dtype)
+
+ latent_channels = self.transformer.config.in_channels
+ latents = self.prepare_latents(
+ video,
+ batch_size * num_videos_per_prompt,
+ latent_channels,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ latent_timestep,
+ )
+
+ # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7. Create rotary embeds if required
+ image_rotary_emb = (
+ self._prepare_rotary_positional_embeddings(height, width, latents.size(1), device)
+ if self.transformer.config.use_rotary_positional_embeddings
+ else None
+ )
+
+ # 8. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ # for DPM-solver++
+ old_pred_original_sample = None
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ self._current_timestep = t
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = t.expand(latent_model_input.shape[0])
+
+ # predict noise model_output
+ with self.transformer.cache_context("cond_uncond"):
+ noise_pred = self.transformer(
+ hidden_states=latent_model_input,
+ encoder_hidden_states=prompt_embeds,
+ timestep=timestep,
+ image_rotary_emb=image_rotary_emb,
+ attention_kwargs=attention_kwargs,
+ return_dict=False,
+ )[0]
+ noise_pred = noise_pred.float()
+
+ # perform guidance
+ if use_dynamic_cfg:
+ self._guidance_scale = 1 + guidance_scale * (
+ (1 - math.cos(math.pi * ((num_inference_steps - t.item()) / num_inference_steps) ** 5.0)) / 2
+ )
+ if do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ if not isinstance(self.scheduler, CogVideoXDPMScheduler):
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+ else:
+ latents, old_pred_original_sample = self.scheduler.step(
+ noise_pred,
+ old_pred_original_sample,
+ t,
+ timesteps[i - 1] if i > 0 else None,
+ latents,
+ **extra_step_kwargs,
+ return_dict=False,
+ )
+ latents = latents.to(prompt_embeds.dtype)
+
+ # call the callback, if provided
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ self._current_timestep = None
+
+ if not output_type == "latent":
+ video = self.decode_latents(latents)
+ video = self.video_processor.postprocess_video(video=video, output_type=output_type)
+ else:
+ video = latents
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (video,)
+
+ return CogVideoXPipelineOutput(frames=video)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_output.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_output.py
new file mode 100644
index 0000000000000000000000000000000000000000..3de030dd6928db49ab0bc4d11868a93ac98dea50
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/cogvideo/pipeline_output.py
@@ -0,0 +1,20 @@
+from dataclasses import dataclass
+
+import torch
+
+from diffusers.utils import BaseOutput
+
+
+@dataclass
+class CogVideoXPipelineOutput(BaseOutput):
+ r"""
+ Output class for CogVideo pipelines.
+
+ Args:
+ frames (`torch.Tensor`, `np.ndarray`, or List[List[PIL.Image.Image]]):
+ List of video outputs - It can be a nested list of length `batch_size,` with each sub-list containing
+ denoised PIL image sequences of length `num_frames.` It can also be a NumPy array or Torch tensor of shape
+ `(batch_size, num_frames, channels, height, width)`.
+ """
+
+ frames: torch.Tensor
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..162d91c010acf95aa2daf87c51ab1e0c68361fd5
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__init__.py
@@ -0,0 +1,24 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+ DIFFUSERS_SLOW_IMPORT,
+ _LazyModule,
+)
+
+
+_import_structure = {
+ "pipeline_consistency_models": ["ConsistencyModelPipeline"],
+}
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ from .pipeline_consistency_models import ConsistencyModelPipeline
+
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__pycache__/__init__.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..096ba589a9e2193a663356882a272734bdd8e2f5
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__pycache__/__init__.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__pycache__/pipeline_consistency_models.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__pycache__/pipeline_consistency_models.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..ffdcfcb7dd22414c75f7ecf32169ed6b22dc282e
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/consistency_models/__pycache__/pipeline_consistency_models.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..a49dccf235a3b84b2695bba6724ba913d3f60155
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__init__.py
@@ -0,0 +1,86 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+ DIFFUSERS_SLOW_IMPORT,
+ OptionalDependencyNotAvailable,
+ _LazyModule,
+ get_objects_from_module,
+ is_flax_available,
+ is_torch_available,
+ is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+ from ...utils import dummy_torch_and_transformers_objects # noqa F403
+
+ _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+ _import_structure["multicontrolnet"] = ["MultiControlNetModel"]
+ _import_structure["pipeline_controlnet"] = ["StableDiffusionControlNetPipeline"]
+ _import_structure["pipeline_controlnet_blip_diffusion"] = ["BlipDiffusionControlNetPipeline"]
+ _import_structure["pipeline_controlnet_img2img"] = ["StableDiffusionControlNetImg2ImgPipeline"]
+ _import_structure["pipeline_controlnet_inpaint"] = ["StableDiffusionControlNetInpaintPipeline"]
+ _import_structure["pipeline_controlnet_inpaint_sd_xl"] = ["StableDiffusionXLControlNetInpaintPipeline"]
+ _import_structure["pipeline_controlnet_sd_xl"] = ["StableDiffusionXLControlNetPipeline"]
+ _import_structure["pipeline_controlnet_sd_xl_img2img"] = ["StableDiffusionXLControlNetImg2ImgPipeline"]
+ _import_structure["pipeline_controlnet_union_inpaint_sd_xl"] = ["StableDiffusionXLControlNetUnionInpaintPipeline"]
+ _import_structure["pipeline_controlnet_union_sd_xl"] = ["StableDiffusionXLControlNetUnionPipeline"]
+ _import_structure["pipeline_controlnet_union_sd_xl_img2img"] = ["StableDiffusionXLControlNetUnionImg2ImgPipeline"]
+try:
+ if not (is_transformers_available() and is_flax_available()):
+ raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+ from ...utils import dummy_flax_and_transformers_objects # noqa F403
+
+ _dummy_objects.update(get_objects_from_module(dummy_flax_and_transformers_objects))
+else:
+ _import_structure["pipeline_flax_controlnet"] = ["FlaxStableDiffusionControlNetPipeline"]
+
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+
+ except OptionalDependencyNotAvailable:
+ from ...utils.dummy_torch_and_transformers_objects import *
+ else:
+ from .multicontrolnet import MultiControlNetModel
+ from .pipeline_controlnet import StableDiffusionControlNetPipeline
+ from .pipeline_controlnet_blip_diffusion import BlipDiffusionControlNetPipeline
+ from .pipeline_controlnet_img2img import StableDiffusionControlNetImg2ImgPipeline
+ from .pipeline_controlnet_inpaint import StableDiffusionControlNetInpaintPipeline
+ from .pipeline_controlnet_inpaint_sd_xl import StableDiffusionXLControlNetInpaintPipeline
+ from .pipeline_controlnet_sd_xl import StableDiffusionXLControlNetPipeline
+ from .pipeline_controlnet_sd_xl_img2img import StableDiffusionXLControlNetImg2ImgPipeline
+ from .pipeline_controlnet_union_inpaint_sd_xl import StableDiffusionXLControlNetUnionInpaintPipeline
+ from .pipeline_controlnet_union_sd_xl import StableDiffusionXLControlNetUnionPipeline
+ from .pipeline_controlnet_union_sd_xl_img2img import StableDiffusionXLControlNetUnionImg2ImgPipeline
+
+ try:
+ if not (is_transformers_available() and is_flax_available()):
+ raise OptionalDependencyNotAvailable()
+ except OptionalDependencyNotAvailable:
+ from ...utils.dummy_flax_and_transformers_objects import * # noqa F403
+ else:
+ from .pipeline_flax_controlnet import FlaxStableDiffusionControlNetPipeline
+
+
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
+ for name, value in _dummy_objects.items():
+ setattr(sys.modules[__name__], name, value)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/__init__.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..db753fdfe5b58c3750da4e623ff310f62e3a100d
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/__init__.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/multicontrolnet.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/multicontrolnet.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..26c5fcbd4ddcbb52c72b6fbcba54ec15b550b3c3
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/multicontrolnet.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..30021a8dd1127b6e146d2307ef12876ce77e4c3f
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_blip_diffusion.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_blip_diffusion.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..f1e982c19d95d108ec423ac42fa3106219803c82
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_blip_diffusion.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_img2img.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_img2img.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..45b855b334317c5faeaee9155467e4e6782886dc
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_img2img.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_inpaint.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_inpaint.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..b4779c972213555276e6475ee4d3eaa88a1a6b0f
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_inpaint.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_inpaint_sd_xl.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_inpaint_sd_xl.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..bbc4284f269f82aada7486e9fb6c9163c16206f4
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_inpaint_sd_xl.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_sd_xl.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_sd_xl.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..5f6ae58aa0367028aa99dfda6a037ae119104bb5
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_sd_xl.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_sd_xl_img2img.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_sd_xl_img2img.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..dafc75f2b01d85eab37733969e4a79ab2c12bd29
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_sd_xl_img2img.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_inpaint_sd_xl.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_inpaint_sd_xl.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..6242708c8c8dde4e259f418d1fc11275f8eda9e3
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_inpaint_sd_xl.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_sd_xl.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_sd_xl.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..34f6899e86ff4d476fe461fe3f62b146a6da4d53
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_sd_xl.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_sd_xl_img2img.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_sd_xl_img2img.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..c610950c14f42e195840e7b58dc29fb7f395f260
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_controlnet_union_sd_xl_img2img.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_flax_controlnet.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_flax_controlnet.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..c7920e4bdc3663ba140221046c00b40c4f392d0c
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/__pycache__/pipeline_flax_controlnet.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/multicontrolnet.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/multicontrolnet.py
new file mode 100644
index 0000000000000000000000000000000000000000..6526dd8c9a576076db072659c81b457b66af7a35
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/multicontrolnet.py
@@ -0,0 +1,12 @@
+from ...models.controlnets.multicontrolnet import MultiControlNetModel
+from ...utils import deprecate, logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class MultiControlNetModel(MultiControlNetModel):
+ def __init__(self, *args, **kwargs):
+ deprecation_message = "Importing `MultiControlNetModel` from `diffusers.pipelines.controlnet.multicontrolnet` is deprecated and this will be removed in a future version. Please use `from diffusers.models.controlnets.multicontrolnet import MultiControlNetModel`, instead."
+ deprecate("diffusers.pipelines.controlnet.multicontrolnet.MultiControlNetModel", "0.34", deprecation_message)
+ super().__init__(*args, **kwargs)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet.py
new file mode 100644
index 0000000000000000000000000000000000000000..fe0e69314ccaae267e502c1da1c636f51a66eb5b
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet.py
@@ -0,0 +1,1366 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, is_torch_version, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion.pipeline_output import StableDiffusionPipelineOutput
+from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # !pip install opencv-python transformers accelerate
+ >>> from diffusers import StableDiffusionControlNetPipeline, ControlNetModel, UniPCMultistepScheduler
+ >>> from diffusers.utils import load_image
+ >>> import numpy as np
+ >>> import torch
+
+ >>> import cv2
+ >>> from PIL import Image
+
+ >>> # download an image
+ >>> image = load_image(
+ ... "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_image_vermeer.png"
+ ... )
+ >>> image = np.array(image)
+
+ >>> # get canny image
+ >>> image = cv2.Canny(image, 100, 200)
+ >>> image = image[:, :, None]
+ >>> image = np.concatenate([image, image, image], axis=2)
+ >>> canny_image = Image.fromarray(image)
+
+ >>> # load control net and stable diffusion v1-5
+ >>> controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
+ >>> pipe = StableDiffusionControlNetPipeline.from_pretrained(
+ ... "stable-diffusion-v1-5/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
+ ... )
+
+ >>> # speed up diffusion process with faster scheduler and memory optimization
+ >>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+ >>> # remove following line if xformers is not installed
+ >>> pipe.enable_xformers_memory_efficient_attention()
+
+ >>> pipe.enable_model_cpu_offload()
+
+ >>> # generate image
+ >>> generator = torch.manual_seed(0)
+ >>> image = pipe(
+ ... "futuristic-looking woman", num_inference_steps=20, generator=generator, image=canny_image
+ ... ).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class StableDiffusionControlNetPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionLoraLoaderMixin,
+ IPAdapterMixin,
+ FromSingleFileMixin,
+):
+ r"""
+ Pipeline for text-to-image generation using Stable Diffusion with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+ text_encoder ([`~transformers.CLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ tokenizer ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ unet ([`UNet2DConditionModel`]):
+ A `UNet2DConditionModel` to denoise the encoded image latents.
+ controlnet ([`ControlNetModel`] or `List[ControlNetModel]`):
+ Provides additional conditioning to the `unet` during the denoising process. If you set multiple
+ ControlNets as a list, the outputs from each ControlNet are added together to create one combined
+ additional conditioning.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ safety_checker ([`StableDiffusionSafetyChecker`]):
+ Classification module that estimates whether generated images could be considered offensive or harmful.
+ Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+ more details about a model's potential harms.
+ feature_extractor ([`~transformers.CLIPImageProcessor`]):
+ A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+ """
+
+ model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
+ _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
+ _exclude_from_cpu_offload = ["safety_checker"]
+ _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds", "image"]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ tokenizer: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
+ scheduler: KarrasDiffusionSchedulers,
+ safety_checker: StableDiffusionSafetyChecker,
+ feature_extractor: CLIPImageProcessor,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ requires_safety_checker: bool = True,
+ ):
+ super().__init__()
+
+ if safety_checker is None and requires_safety_checker:
+ logger.warning(
+ f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+ " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+ " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+ " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+ " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+ " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+ )
+
+ if safety_checker is not None and feature_extractor is None:
+ raise ValueError(
+ "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+ " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+ )
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ safety_checker=safety_checker,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ self.register_to_config(requires_safety_checker=requires_safety_checker)
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
+ def _encode_prompt(
+ self,
+ prompt,
+ device,
+ num_images_per_prompt,
+ do_classifier_free_guidance,
+ negative_prompt=None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ **kwargs,
+ ):
+ deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple."
+ deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False)
+
+ prompt_embeds_tuple = self.encode_prompt(
+ prompt=prompt,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ lora_scale=lora_scale,
+ **kwargs,
+ )
+
+ # concatenate for backwards comp
+ prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]])
+
+ return prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt,
+ device,
+ num_images_per_prompt,
+ do_classifier_free_guidance,
+ negative_prompt=None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ lora_scale (`float`, *optional*):
+ A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ # textual inversion: process multi-vector tokens if necessary
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=self.tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = self.tokenizer.batch_decode(
+ untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+ )
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+ attention_mask = text_inputs.attention_mask.to(device)
+ else:
+ attention_mask = None
+
+ if clip_skip is None:
+ prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask)
+ prompt_embeds = prompt_embeds[0]
+ else:
+ prompt_embeds = self.text_encoder(
+ text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True
+ )
+ # Access the `hidden_states` first, that contains a tuple of
+ # all the hidden states from the encoder layers. Then index into
+ # the tuple to access the hidden states from the desired layer.
+ prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
+ # We also need to apply the final LayerNorm here to not mess with the
+ # representations. The `last_hidden_states` that we typically use for
+ # obtaining the final prompt representations passes through the LayerNorm
+ # layer.
+ prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+
+ if self.text_encoder is not None:
+ prompt_embeds_dtype = self.text_encoder.dtype
+ elif self.unet is not None:
+ prompt_embeds_dtype = self.unet.dtype
+ else:
+ prompt_embeds_dtype = prompt_embeds.dtype
+
+ prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ # get unconditional embeddings for classifier free guidance
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ uncond_tokens: List[str]
+ if negative_prompt is None:
+ uncond_tokens = [""] * batch_size
+ elif prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif isinstance(negative_prompt, str):
+ uncond_tokens = [negative_prompt]
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = negative_prompt
+
+ # textual inversion: process multi-vector tokens if necessary
+ if isinstance(self, TextualInversionLoaderMixin):
+ uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = self.tokenizer(
+ uncond_tokens,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+ attention_mask = uncond_input.attention_mask.to(device)
+ else:
+ attention_mask = None
+
+ negative_prompt_embeds = self.text_encoder(
+ uncond_input.input_ids.to(device),
+ attention_mask=attention_mask,
+ )
+ negative_prompt_embeds = negative_prompt_embeds[0]
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+ def run_safety_checker(self, image, device, dtype):
+ if self.safety_checker is None:
+ has_nsfw_concept = None
+ else:
+ if torch.is_tensor(image):
+ feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
+ else:
+ feature_extractor_input = self.image_processor.numpy_to_pil(image)
+ safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
+ image, has_nsfw_concept = self.safety_checker(
+ images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+ )
+ return image, has_nsfw_concept
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+ def decode_latents(self, latents):
+ deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead"
+ deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False)
+
+ latents = 1 / self.vae.config.scaling_factor * latents
+ image = self.vae.decode(latents, return_dict=False)[0]
+ image = (image / 2 + 0.5).clamp(0, 1)
+ # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
+ image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ image,
+ callback_steps,
+ negative_prompt=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ # Check `image`
+ is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
+ self.controlnet, torch._dynamo.eval_frame.OptimizedModule
+ )
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ self.check_image(image, prompt, prompt_embeds)
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+ # When `image` is a nested list:
+ # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
+ elif any(isinstance(i, list) for i in image):
+ transposed_image = [list(t) for t in zip(*image)]
+ if len(transposed_image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: if you pass`image` as a list of list, each sublist must have the same length as the number of controlnets, but the sublists in `image` got {len(transposed_image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+ for image_ in transposed_image:
+ self.check_image(image_, prompt, prompt_embeds)
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+ else:
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ else:
+ assert False
+
+ # Check `controlnet_conditioning_scale`
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ if not isinstance(controlnet_conditioning_scale, float):
+ raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError(
+ "A single batch of varying conditioning scale settings (e.g. [[1.0, 0.5], [0.2, 0.8]]) is not supported at the moment. "
+ "The conditioning scale must be fixed across the batch."
+ )
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+ else:
+ assert False
+
+ if not isinstance(control_guidance_start, (tuple, list)):
+ control_guidance_start = [control_guidance_start]
+
+ if not isinstance(control_guidance_end, (tuple, list)):
+ control_guidance_end = [control_guidance_end]
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ def prepare_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
+ def get_guidance_scale_embedding(
+ self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+ ) -> torch.Tensor:
+ """
+ See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+
+ Args:
+ w (`torch.Tensor`):
+ Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
+ embedding_dim (`int`, *optional*, defaults to 512):
+ Dimension of the embeddings to generate.
+ dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
+ Data type of the generated embeddings.
+
+ Returns:
+ `torch.Tensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
+ """
+ assert len(w.shape) == 1
+ w = w * 1000.0
+
+ half_dim = embedding_dim // 2
+ emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
+ emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
+ emb = w.to(dtype)[:, None] * emb[None, :]
+ emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+ if embedding_dim % 2 == 1: # zero pad
+ emb = torch.nn.functional.pad(emb, (0, 1))
+ assert emb.shape == (w.shape[0], embedding_dim)
+ return emb
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ image: PipelineImageInput = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 50,
+ timesteps: List[int] = None,
+ sigmas: List[float] = None,
+ guidance_scale: float = 7.5,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+ image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet. When `prompt` is a list, and if a list of images is passed for a single
+ ControlNet, each will be paired with each prompt in the `prompt` list. This also applies to multiple
+ ControlNets, where a list of image lists can be passed to batch for each prompt and each ControlNet.
+ height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The height in pixels of the generated image.
+ width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The width in pixels of the generated image.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+ their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+ will be used.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+ pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+ applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor is generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ callback (`Callable`, *optional*):
+ A function that calls every `callback_steps` steps during inference. The function is called with the
+ following arguments: `callback(step: int, timestep: int, latents: torch.Tensor)`.
+ callback_steps (`int`, *optional*, defaults to 1):
+ The frequency at which the `callback` function is called. If not specified, the callback is called at
+ every step.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+ [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ The ControlNet encoder tries to recognize the content of the input image even if you remove all
+ prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+ otherwise a `tuple` is returned where the first element is a list with the generated images and the
+ second element is a list of `bool`s indicating whether the corresponding generated image contains
+ "not-safe-for-work" (nsfw) content.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ image,
+ callback_steps,
+ negative_prompt,
+ prompt_embeds,
+ negative_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ callback_on_step_end_tensor_inputs,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+ # For classifier free guidance, we need to do two forward passes.
+ # Here we concatenate the unconditional and text embeddings into a single batch
+ # to avoid doing two forward passes
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. Prepare image
+ if isinstance(controlnet, ControlNetModel):
+ image = self.prepare_image(
+ image=image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+ height, width = image.shape[-2:]
+ elif isinstance(controlnet, MultiControlNetModel):
+ images = []
+
+ # Nested lists as ControlNet condition
+ if isinstance(image[0], list):
+ # Transpose the nested image list
+ image = [list(t) for t in zip(*image)]
+
+ for image_ in image:
+ image_ = self.prepare_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ images.append(image_)
+
+ image = images
+ height, width = image[0].shape[-2:]
+ else:
+ assert False
+
+ # 5. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(
+ self.scheduler, num_inference_steps, device, timesteps, sigmas
+ )
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.unet.config.in_channels
+ latents = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 6.5 Optionally get Guidance Scale Embedding
+ timestep_cond = None
+ if self.unet.config.time_cond_proj_dim is not None:
+ guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+ timestep_cond = self.get_guidance_scale_embedding(
+ guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+ ).to(device=device, dtype=latents.dtype)
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Add image embeds for IP-Adapter
+ added_cond_kwargs = (
+ {"image_embeds": image_embeds}
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+ else None
+ )
+
+ # 7.2 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps)
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ is_unet_compiled = is_compiled_module(self.unet)
+ is_controlnet_compiled = is_compiled_module(self.controlnet)
+ is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1")
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # Relevant thread:
+ # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428
+ if (is_unet_compiled and is_controlnet_compiled) and is_torch_higher_equal_2_1:
+ torch._inductor.cudagraph_mark_step_begin()
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=image,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ timestep_cond=timestep_cond,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ image = callback_outputs.pop("image", image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
+ 0
+ ]
+ image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+ else:
+ image = latents
+ has_nsfw_concept = None
+
+ if has_nsfw_concept is None:
+ do_denormalize = [True] * image.shape[0]
+ else:
+ do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+ image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image, has_nsfw_concept)
+
+ return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_blip_diffusion.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_blip_diffusion.py
new file mode 100644
index 0000000000000000000000000000000000000000..c2ae408778b346c984803741ab4f70fb265a18c6
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_blip_diffusion.py
@@ -0,0 +1,427 @@
+# Copyright 2025 Salesforce.com, inc.
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import List, Optional, Union
+
+import PIL.Image
+import torch
+from transformers import CLIPTokenizer
+
+from ...models import AutoencoderKL, ControlNetModel, UNet2DConditionModel
+from ...schedulers import PNDMScheduler
+from ...utils import (
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+)
+from ...utils.torch_utils import randn_tensor
+from ..blip_diffusion.blip_image_processing import BlipImageProcessor
+from ..blip_diffusion.modeling_blip2 import Blip2QFormerModel
+from ..blip_diffusion.modeling_ctx_clip import ContextCLIPTextModel
+from ..pipeline_utils import DeprecatedPipelineMixin, DiffusionPipeline, ImagePipelineOutput
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> from diffusers.pipelines import BlipDiffusionControlNetPipeline
+ >>> from diffusers.utils import load_image
+ >>> from controlnet_aux import CannyDetector
+ >>> import torch
+
+ >>> blip_diffusion_pipe = BlipDiffusionControlNetPipeline.from_pretrained(
+ ... "Salesforce/blipdiffusion-controlnet", torch_dtype=torch.float16
+ ... ).to("cuda")
+
+ >>> style_subject = "flower"
+ >>> tgt_subject = "teapot"
+ >>> text_prompt = "on a marble table"
+
+ >>> cldm_cond_image = load_image(
+ ... "https://huggingface.co/datasets/ayushtues/blipdiffusion_images/resolve/main/kettle.jpg"
+ ... ).resize((512, 512))
+ >>> canny = CannyDetector()
+ >>> cldm_cond_image = canny(cldm_cond_image, 30, 70, output_type="pil")
+ >>> style_image = load_image(
+ ... "https://huggingface.co/datasets/ayushtues/blipdiffusion_images/resolve/main/flower.jpg"
+ ... )
+ >>> guidance_scale = 7.5
+ >>> num_inference_steps = 50
+ >>> negative_prompt = "over-exposure, under-exposure, saturated, duplicate, out of frame, lowres, cropped, worst quality, low quality, jpeg artifacts, morbid, mutilated, out of frame, ugly, bad anatomy, bad proportions, deformed, blurry, duplicate"
+
+
+ >>> output = blip_diffusion_pipe(
+ ... text_prompt,
+ ... style_image,
+ ... cldm_cond_image,
+ ... style_subject,
+ ... tgt_subject,
+ ... guidance_scale=guidance_scale,
+ ... num_inference_steps=num_inference_steps,
+ ... neg_prompt=negative_prompt,
+ ... height=512,
+ ... width=512,
+ ... ).images
+ >>> output[0].save("image.png")
+ ```
+"""
+
+
+class BlipDiffusionControlNetPipeline(DeprecatedPipelineMixin, DiffusionPipeline):
+ """
+ Pipeline for Canny Edge based Controlled subject-driven generation using Blip Diffusion.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ Args:
+ tokenizer ([`CLIPTokenizer`]):
+ Tokenizer for the text encoder
+ text_encoder ([`ContextCLIPTextModel`]):
+ Text encoder to encode the text prompt
+ vae ([`AutoencoderKL`]):
+ VAE model to map the latents to the image
+ unet ([`UNet2DConditionModel`]):
+ Conditional U-Net architecture to denoise the image embedding.
+ scheduler ([`PNDMScheduler`]):
+ A scheduler to be used in combination with `unet` to generate image latents.
+ qformer ([`Blip2QFormerModel`]):
+ QFormer model to get multi-modal embeddings from the text and image.
+ controlnet ([`ControlNetModel`]):
+ ControlNet model to get the conditioning image embedding.
+ image_processor ([`BlipImageProcessor`]):
+ Image Processor to preprocess and postprocess the image.
+ ctx_begin_pos (int, `optional`, defaults to 2):
+ Position of the context token in the text encoder.
+ """
+
+ _last_supported_version = "0.33.1"
+ model_cpu_offload_seq = "qformer->text_encoder->unet->vae"
+
+ def __init__(
+ self,
+ tokenizer: CLIPTokenizer,
+ text_encoder: ContextCLIPTextModel,
+ vae: AutoencoderKL,
+ unet: UNet2DConditionModel,
+ scheduler: PNDMScheduler,
+ qformer: Blip2QFormerModel,
+ controlnet: ControlNetModel,
+ image_processor: BlipImageProcessor,
+ ctx_begin_pos: int = 2,
+ mean: List[float] = None,
+ std: List[float] = None,
+ ):
+ super().__init__()
+
+ self.register_modules(
+ tokenizer=tokenizer,
+ text_encoder=text_encoder,
+ vae=vae,
+ unet=unet,
+ scheduler=scheduler,
+ qformer=qformer,
+ controlnet=controlnet,
+ image_processor=image_processor,
+ )
+ self.register_to_config(ctx_begin_pos=ctx_begin_pos, mean=mean, std=std)
+
+ def get_query_embeddings(self, input_image, src_subject):
+ return self.qformer(image_input=input_image, text_input=src_subject, return_dict=False)
+
+ # from the original Blip Diffusion code, specifies the target subject and augments the prompt by repeating it
+ def _build_prompt(self, prompts, tgt_subjects, prompt_strength=1.0, prompt_reps=20):
+ rv = []
+ for prompt, tgt_subject in zip(prompts, tgt_subjects):
+ prompt = f"a {tgt_subject} {prompt.strip()}"
+ # a trick to amplify the prompt
+ rv.append(", ".join([prompt] * int(prompt_strength * prompt_reps)))
+
+ return rv
+
+ # Copied from diffusers.pipelines.consistency_models.pipeline_consistency_models.ConsistencyModelPipeline.prepare_latents
+ def prepare_latents(self, batch_size, num_channels, height, width, dtype, device, generator, latents=None):
+ shape = (batch_size, num_channels, height, width)
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device=device, dtype=dtype)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ def encode_prompt(self, query_embeds, prompt, device=None):
+ device = device or self._execution_device
+
+ # embeddings for prompt, with query_embeds as context
+ max_len = self.text_encoder.text_model.config.max_position_embeddings
+ max_len -= self.qformer.config.num_query_tokens
+
+ tokenized_prompt = self.tokenizer(
+ prompt,
+ padding="max_length",
+ truncation=True,
+ max_length=max_len,
+ return_tensors="pt",
+ ).to(device)
+
+ batch_size = query_embeds.shape[0]
+ ctx_begin_pos = [self.config.ctx_begin_pos] * batch_size
+
+ text_embeddings = self.text_encoder(
+ input_ids=tokenized_prompt.input_ids,
+ ctx_embeddings=query_embeds,
+ ctx_begin_pos=ctx_begin_pos,
+ )[0]
+
+ return text_embeddings
+
+ # Adapted from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.prepare_image
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ ):
+ image = self.image_processor.preprocess(
+ image,
+ size={"width": width, "height": height},
+ do_rescale=True,
+ do_center_crop=False,
+ do_normalize=False,
+ return_tensors="pt",
+ )["pixel_values"].to(device)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: List[str],
+ reference_image: PIL.Image.Image,
+ condtioning_image: PIL.Image.Image,
+ source_subject_category: List[str],
+ target_subject_category: List[str],
+ latents: Optional[torch.Tensor] = None,
+ guidance_scale: float = 7.5,
+ height: int = 512,
+ width: int = 512,
+ num_inference_steps: int = 50,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ neg_prompt: Optional[str] = "",
+ prompt_strength: float = 1.0,
+ prompt_reps: int = 20,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ ):
+ """
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`List[str]`):
+ The prompt or prompts to guide the image generation.
+ reference_image (`PIL.Image.Image`):
+ The reference image to condition the generation on.
+ condtioning_image (`PIL.Image.Image`):
+ The conditioning canny edge image to condition the generation on.
+ source_subject_category (`List[str]`):
+ The source subject category.
+ target_subject_category (`List[str]`):
+ The target subject category.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by random sampling.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ height (`int`, *optional*, defaults to 512):
+ The height of the generated image.
+ width (`int`, *optional*, defaults to 512):
+ The width of the generated image.
+ seed (`int`, *optional*, defaults to 42):
+ The seed to use for random generation.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ neg_prompt (`str`, *optional*, defaults to ""):
+ The prompt or prompts not to guide the image generation. Ignored when not using guidance (i.e., ignored
+ if `guidance_scale` is less than `1`).
+ prompt_strength (`float`, *optional*, defaults to 1.0):
+ The strength of the prompt. Specifies the number of times the prompt is repeated along with prompt_reps
+ to amplify the prompt.
+ prompt_reps (`int`, *optional*, defaults to 20):
+ The number of times the prompt is repeated along with prompt_strength to amplify the prompt.
+ Examples:
+
+ Returns:
+ [`~pipelines.ImagePipelineOutput`] or `tuple`
+ """
+ device = self._execution_device
+
+ reference_image = self.image_processor.preprocess(
+ reference_image, image_mean=self.config.mean, image_std=self.config.std, return_tensors="pt"
+ )["pixel_values"]
+ reference_image = reference_image.to(device)
+
+ if isinstance(prompt, str):
+ prompt = [prompt]
+ if isinstance(source_subject_category, str):
+ source_subject_category = [source_subject_category]
+ if isinstance(target_subject_category, str):
+ target_subject_category = [target_subject_category]
+
+ batch_size = len(prompt)
+
+ prompt = self._build_prompt(
+ prompts=prompt,
+ tgt_subjects=target_subject_category,
+ prompt_strength=prompt_strength,
+ prompt_reps=prompt_reps,
+ )
+ query_embeds = self.get_query_embeddings(reference_image, source_subject_category)
+ text_embeddings = self.encode_prompt(query_embeds, prompt, device)
+ # 3. unconditional embedding
+ do_classifier_free_guidance = guidance_scale > 1.0
+ if do_classifier_free_guidance:
+ max_length = self.text_encoder.text_model.config.max_position_embeddings
+
+ uncond_input = self.tokenizer(
+ [neg_prompt] * batch_size,
+ padding="max_length",
+ max_length=max_length,
+ return_tensors="pt",
+ )
+ uncond_embeddings = self.text_encoder(
+ input_ids=uncond_input.input_ids.to(device),
+ ctx_embeddings=None,
+ )[0]
+ # For classifier free guidance, we need to do two forward passes.
+ # Here we concatenate the unconditional and text embeddings into a single batch
+ # to avoid doing two forward passes
+ text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
+ scale_down_factor = 2 ** (len(self.unet.config.block_out_channels) - 1)
+ latents = self.prepare_latents(
+ batch_size=batch_size,
+ num_channels=self.unet.config.in_channels,
+ height=height // scale_down_factor,
+ width=width // scale_down_factor,
+ generator=generator,
+ latents=latents,
+ dtype=self.unet.dtype,
+ device=device,
+ )
+ # set timesteps
+ extra_set_kwargs = {}
+ self.scheduler.set_timesteps(num_inference_steps, **extra_set_kwargs)
+
+ cond_image = self.prepare_control_image(
+ image=condtioning_image,
+ width=width,
+ height=height,
+ batch_size=batch_size,
+ num_images_per_prompt=1,
+ device=device,
+ dtype=self.controlnet.dtype,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ )
+
+ for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)):
+ # expand the latents if we are doing classifier free guidance
+ do_classifier_free_guidance = guidance_scale > 1.0
+
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=text_embeddings,
+ controlnet_cond=cond_image,
+ return_dict=False,
+ )
+
+ noise_pred = self.unet(
+ latent_model_input,
+ timestep=t,
+ encoder_hidden_states=text_embeddings,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ )["sample"]
+
+ # perform guidance
+ if do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ latents = self.scheduler.step(
+ noise_pred,
+ t,
+ latents,
+ )["prev_sample"]
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return ImagePipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py
new file mode 100644
index 0000000000000000000000000000000000000000..12cc6f630d805658b104dcefd2a40a0fb459c893
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_img2img.py
@@ -0,0 +1,1338 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion import StableDiffusionPipelineOutput
+from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # !pip install opencv-python transformers accelerate
+ >>> from diffusers import StableDiffusionControlNetImg2ImgPipeline, ControlNetModel, UniPCMultistepScheduler
+ >>> from diffusers.utils import load_image
+ >>> import numpy as np
+ >>> import torch
+
+ >>> import cv2
+ >>> from PIL import Image
+
+ >>> # download an image
+ >>> image = load_image(
+ ... "https://hf.co/datasets/huggingface/documentation-images/resolve/main/diffusers/input_image_vermeer.png"
+ ... )
+ >>> np_image = np.array(image)
+
+ >>> # get canny image
+ >>> np_image = cv2.Canny(np_image, 100, 200)
+ >>> np_image = np_image[:, :, None]
+ >>> np_image = np.concatenate([np_image, np_image, np_image], axis=2)
+ >>> canny_image = Image.fromarray(np_image)
+
+ >>> # load control net and stable diffusion v1-5
+ >>> controlnet = ControlNetModel.from_pretrained("lllyasviel/sd-controlnet-canny", torch_dtype=torch.float16)
+ >>> pipe = StableDiffusionControlNetImg2ImgPipeline.from_pretrained(
+ ... "stable-diffusion-v1-5/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
+ ... )
+
+ >>> # speed up diffusion process with faster scheduler and memory optimization
+ >>> pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
+ >>> pipe.enable_model_cpu_offload()
+
+ >>> # generate image
+ >>> generator = torch.manual_seed(0)
+ >>> image = pipe(
+ ... "futuristic-looking woman",
+ ... num_inference_steps=20,
+ ... generator=generator,
+ ... image=image,
+ ... control_image=canny_image,
+ ... ).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+def prepare_image(image):
+ if isinstance(image, torch.Tensor):
+ # Batch single image
+ if image.ndim == 3:
+ image = image.unsqueeze(0)
+
+ image = image.to(dtype=torch.float32)
+ else:
+ # preprocess image
+ if isinstance(image, (PIL.Image.Image, np.ndarray)):
+ image = [image]
+
+ if isinstance(image, list) and isinstance(image[0], PIL.Image.Image):
+ image = [np.array(i.convert("RGB"))[None, :] for i in image]
+ image = np.concatenate(image, axis=0)
+ elif isinstance(image, list) and isinstance(image[0], np.ndarray):
+ image = np.concatenate([i[None, :] for i in image], axis=0)
+
+ image = image.transpose(0, 3, 1, 2)
+ image = torch.from_numpy(image).to(dtype=torch.float32) / 127.5 - 1.0
+
+ return image
+
+
+class StableDiffusionControlNetImg2ImgPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionLoraLoaderMixin,
+ IPAdapterMixin,
+ FromSingleFileMixin,
+):
+ r"""
+ Pipeline for image-to-image generation using Stable Diffusion with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+ text_encoder ([`~transformers.CLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ tokenizer ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ unet ([`UNet2DConditionModel`]):
+ A `UNet2DConditionModel` to denoise the encoded image latents.
+ controlnet ([`ControlNetModel`] or `List[ControlNetModel]`):
+ Provides additional conditioning to the `unet` during the denoising process. If you set multiple
+ ControlNets as a list, the outputs from each ControlNet are added together to create one combined
+ additional conditioning.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ safety_checker ([`StableDiffusionSafetyChecker`]):
+ Classification module that estimates whether generated images could be considered offensive or harmful.
+ Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+ more details about a model's potential harms.
+ feature_extractor ([`~transformers.CLIPImageProcessor`]):
+ A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+ """
+
+ model_cpu_offload_seq = "text_encoder->unet->vae"
+ _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
+ _exclude_from_cpu_offload = ["safety_checker"]
+ _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds", "control_image"]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ tokenizer: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
+ scheduler: KarrasDiffusionSchedulers,
+ safety_checker: StableDiffusionSafetyChecker,
+ feature_extractor: CLIPImageProcessor,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ requires_safety_checker: bool = True,
+ ):
+ super().__init__()
+
+ if safety_checker is None and requires_safety_checker:
+ logger.warning(
+ f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+ " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+ " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+ " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+ " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+ " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+ )
+
+ if safety_checker is not None and feature_extractor is None:
+ raise ValueError(
+ "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+ " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+ )
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ safety_checker=safety_checker,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ self.register_to_config(requires_safety_checker=requires_safety_checker)
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
+ def _encode_prompt(
+ self,
+ prompt,
+ device,
+ num_images_per_prompt,
+ do_classifier_free_guidance,
+ negative_prompt=None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ **kwargs,
+ ):
+ deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple."
+ deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False)
+
+ prompt_embeds_tuple = self.encode_prompt(
+ prompt=prompt,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ lora_scale=lora_scale,
+ **kwargs,
+ )
+
+ # concatenate for backwards comp
+ prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]])
+
+ return prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt,
+ device,
+ num_images_per_prompt,
+ do_classifier_free_guidance,
+ negative_prompt=None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ lora_scale (`float`, *optional*):
+ A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ # textual inversion: process multi-vector tokens if necessary
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=self.tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = self.tokenizer.batch_decode(
+ untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+ )
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+ attention_mask = text_inputs.attention_mask.to(device)
+ else:
+ attention_mask = None
+
+ if clip_skip is None:
+ prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask)
+ prompt_embeds = prompt_embeds[0]
+ else:
+ prompt_embeds = self.text_encoder(
+ text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True
+ )
+ # Access the `hidden_states` first, that contains a tuple of
+ # all the hidden states from the encoder layers. Then index into
+ # the tuple to access the hidden states from the desired layer.
+ prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
+ # We also need to apply the final LayerNorm here to not mess with the
+ # representations. The `last_hidden_states` that we typically use for
+ # obtaining the final prompt representations passes through the LayerNorm
+ # layer.
+ prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+
+ if self.text_encoder is not None:
+ prompt_embeds_dtype = self.text_encoder.dtype
+ elif self.unet is not None:
+ prompt_embeds_dtype = self.unet.dtype
+ else:
+ prompt_embeds_dtype = prompt_embeds.dtype
+
+ prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ # get unconditional embeddings for classifier free guidance
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ uncond_tokens: List[str]
+ if negative_prompt is None:
+ uncond_tokens = [""] * batch_size
+ elif prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif isinstance(negative_prompt, str):
+ uncond_tokens = [negative_prompt]
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = negative_prompt
+
+ # textual inversion: process multi-vector tokens if necessary
+ if isinstance(self, TextualInversionLoaderMixin):
+ uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = self.tokenizer(
+ uncond_tokens,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+ attention_mask = uncond_input.attention_mask.to(device)
+ else:
+ attention_mask = None
+
+ negative_prompt_embeds = self.text_encoder(
+ uncond_input.input_ids.to(device),
+ attention_mask=attention_mask,
+ )
+ negative_prompt_embeds = negative_prompt_embeds[0]
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+ def run_safety_checker(self, image, device, dtype):
+ if self.safety_checker is None:
+ has_nsfw_concept = None
+ else:
+ if torch.is_tensor(image):
+ feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
+ else:
+ feature_extractor_input = self.image_processor.numpy_to_pil(image)
+ safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
+ image, has_nsfw_concept = self.safety_checker(
+ images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+ )
+ return image, has_nsfw_concept
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+ def decode_latents(self, latents):
+ deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead"
+ deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False)
+
+ latents = 1 / self.vae.config.scaling_factor * latents
+ image = self.vae.decode(latents, return_dict=False)[0]
+ image = (image / 2 + 0.5).clamp(0, 1)
+ # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
+ image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ image,
+ callback_steps,
+ negative_prompt=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
+ self.controlnet, torch._dynamo.eval_frame.OptimizedModule
+ )
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ self.check_image(image, prompt, prompt_embeds)
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+ # When `image` is a nested list:
+ # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
+ elif any(isinstance(i, list) for i in image):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ else:
+ assert False
+
+ # Check `controlnet_conditioning_scale`
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ if not isinstance(controlnet_conditioning_scale, float):
+ raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+ else:
+ assert False
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.prepare_image
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+ t_start = max(num_inference_steps - init_timestep, 0)
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.prepare_latents
+ def prepare_latents(self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None):
+ if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
+ raise ValueError(
+ f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+ )
+
+ image = image.to(device=device, dtype=dtype)
+
+ batch_size = batch_size * num_images_per_prompt
+
+ if image.shape[1] == 4:
+ init_latents = image
+
+ else:
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ elif isinstance(generator, list):
+ if image.shape[0] < batch_size and batch_size % image.shape[0] == 0:
+ image = torch.cat([image] * (batch_size // image.shape[0]), dim=0)
+ elif image.shape[0] < batch_size and batch_size % image.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {image.shape[0]} to effective batch_size {batch_size} "
+ )
+
+ init_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(batch_size)
+ ]
+ init_latents = torch.cat(init_latents, dim=0)
+ else:
+ init_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ init_latents = self.vae.config.scaling_factor * init_latents
+
+ if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
+ # expand init_latents for batch_size
+ deprecation_message = (
+ f"You have passed {batch_size} text prompts (`prompt`), but only {init_latents.shape[0]} initial"
+ " images (`image`). Initial images are now duplicating to match the number of text prompts. Note"
+ " that this behavior is deprecated and will be removed in a version 1.0.0. Please make sure to update"
+ " your script to pass as many initial images as text prompts to suppress this warning."
+ )
+ deprecate("len(prompt) != len(image)", "1.0.0", deprecation_message, standard_warn=False)
+ additional_image_per_prompt = batch_size // init_latents.shape[0]
+ init_latents = torch.cat([init_latents] * additional_image_per_prompt, dim=0)
+ elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+ )
+ else:
+ init_latents = torch.cat([init_latents], dim=0)
+
+ shape = init_latents.shape
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+
+ # get latents
+ init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+ latents = init_latents
+
+ return latents
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ image: PipelineImageInput = None,
+ control_image: PipelineImageInput = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ strength: float = 0.8,
+ num_inference_steps: int = 50,
+ guidance_scale: float = 7.5,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 0.8,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+ image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The initial image to be used as the starting point for the image generation process. Can also accept
+ image latents as `image`, and if passing latents directly they are not encoded again.
+ control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The height in pixels of the generated image.
+ width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The width in pixels of the generated image.
+ strength (`float`, *optional*, defaults to 0.8):
+ Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
+ starting point and more noise is added the higher the `strength`. The number of denoising steps depends
+ on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
+ process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
+ essentially ignores `image`.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+ pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+ applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor is generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+ [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ The ControlNet encoder tries to recognize the content of the input image even if you remove all
+ prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+ otherwise a `tuple` is returned where the first element is a list with the generated images and the
+ second element is a list of `bool`s indicating whether the corresponding generated image contains
+ "not-safe-for-work" (nsfw) content.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ control_image,
+ callback_steps,
+ negative_prompt,
+ prompt_embeds,
+ negative_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ callback_on_step_end_tensor_inputs,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+ # For classifier free guidance, we need to do two forward passes.
+ # Here we concatenate the unconditional and text embeddings into a single batch
+ # to avoid doing two forward passes
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. Prepare image
+ image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+
+ # 5. Prepare controlnet_conditioning_image
+ if isinstance(controlnet, ControlNetModel):
+ control_image = self.prepare_control_image(
+ image=control_image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+ elif isinstance(controlnet, MultiControlNetModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ control_image_ = self.prepare_control_image(
+ image=control_image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(control_image_)
+
+ control_image = control_images
+ else:
+ assert False
+
+ # 5. Prepare timesteps
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ if latents is None:
+ latents = self.prepare_latents(
+ image,
+ latent_timestep,
+ batch_size,
+ num_images_per_prompt,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ )
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Add image embeds for IP-Adapter
+ added_cond_kwargs = (
+ {"image_embeds": image_embeds}
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+ else None
+ )
+
+ # 7.2 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps)
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ control_image = callback_outputs.pop("control_image", control_image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
+ 0
+ ]
+ image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+ else:
+ image = latents
+ has_nsfw_concept = None
+
+ if has_nsfw_concept is None:
+ do_denormalize = [True] * image.shape[0]
+ else:
+ do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+ image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image, has_nsfw_concept)
+
+ return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py
new file mode 100644
index 0000000000000000000000000000000000000000..41303d9c5c5a38b1c7c4b2ed32622a5744753934
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint.py
@@ -0,0 +1,1530 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# This model implementation is heavily inspired by https://github.com/haofanwang/ControlNet-for-Diffusers/
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import FromSingleFileMixin, IPAdapterMixin, StableDiffusionLoraLoaderMixin, TextualInversionLoaderMixin
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion import StableDiffusionPipelineOutput
+from ..stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # !pip install transformers accelerate
+ >>> from diffusers import StableDiffusionControlNetInpaintPipeline, ControlNetModel, DDIMScheduler
+ >>> from diffusers.utils import load_image
+ >>> import numpy as np
+ >>> import torch
+
+ >>> init_image = load_image(
+ ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy.png"
+ ... )
+ >>> init_image = init_image.resize((512, 512))
+
+ >>> generator = torch.Generator(device="cpu").manual_seed(1)
+
+ >>> mask_image = load_image(
+ ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy_mask.png"
+ ... )
+ >>> mask_image = mask_image.resize((512, 512))
+
+
+ >>> def make_canny_condition(image):
+ ... image = np.array(image)
+ ... image = cv2.Canny(image, 100, 200)
+ ... image = image[:, :, None]
+ ... image = np.concatenate([image, image, image], axis=2)
+ ... image = Image.fromarray(image)
+ ... return image
+
+
+ >>> control_image = make_canny_condition(init_image)
+
+ >>> controlnet = ControlNetModel.from_pretrained(
+ ... "lllyasviel/control_v11p_sd15_inpaint", torch_dtype=torch.float16
+ ... )
+ >>> pipe = StableDiffusionControlNetInpaintPipeline.from_pretrained(
+ ... "stable-diffusion-v1-5/stable-diffusion-v1-5", controlnet=controlnet, torch_dtype=torch.float16
+ ... )
+
+ >>> pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+ >>> pipe.enable_model_cpu_offload()
+
+ >>> # generate image
+ >>> image = pipe(
+ ... "a handsome man with ray-ban sunglasses",
+ ... num_inference_steps=20,
+ ... generator=generator,
+ ... eta=1.0,
+ ... image=init_image,
+ ... mask_image=mask_image,
+ ... control_image=control_image,
+ ... ).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class StableDiffusionControlNetInpaintPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionLoraLoaderMixin,
+ IPAdapterMixin,
+ FromSingleFileMixin,
+):
+ r"""
+ Pipeline for image inpainting using Stable Diffusion with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+
+
+ This pipeline can be used with checkpoints that have been specifically fine-tuned for inpainting
+ ([stable-diffusion-v1-5/stable-diffusion-inpainting](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-inpainting))
+ as well as default text-to-image Stable Diffusion checkpoints
+ ([stable-diffusion-v1-5/stable-diffusion-v1-5](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5)).
+ Default text-to-image Stable Diffusion checkpoints might be preferable for ControlNets that have been fine-tuned on
+ those, such as [lllyasviel/control_v11p_sd15_inpaint](https://huggingface.co/lllyasviel/control_v11p_sd15_inpaint).
+
+
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+ text_encoder ([`~transformers.CLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ tokenizer ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ unet ([`UNet2DConditionModel`]):
+ A `UNet2DConditionModel` to denoise the encoded image latents.
+ controlnet ([`ControlNetModel`] or `List[ControlNetModel]`):
+ Provides additional conditioning to the `unet` during the denoising process. If you set multiple
+ ControlNets as a list, the outputs from each ControlNet are added together to create one combined
+ additional conditioning.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ safety_checker ([`StableDiffusionSafetyChecker`]):
+ Classification module that estimates whether generated images could be considered offensive or harmful.
+ Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+ more details about a model's potential harms.
+ feature_extractor ([`~transformers.CLIPImageProcessor`]):
+ A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+ """
+
+ model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
+ _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
+ _exclude_from_cpu_offload = ["safety_checker"]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ "control_image",
+ "mask",
+ "masked_image_latents",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ tokenizer: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
+ scheduler: KarrasDiffusionSchedulers,
+ safety_checker: StableDiffusionSafetyChecker,
+ feature_extractor: CLIPImageProcessor,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ requires_safety_checker: bool = True,
+ ):
+ super().__init__()
+
+ if safety_checker is None and requires_safety_checker:
+ logger.warning(
+ f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+ " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+ " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+ " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+ " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+ " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+ )
+
+ if safety_checker is not None and feature_extractor is None:
+ raise ValueError(
+ "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+ " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+ )
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ safety_checker=safety_checker,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+ self.mask_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True
+ )
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ self.register_to_config(requires_safety_checker=requires_safety_checker)
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline._encode_prompt
+ def _encode_prompt(
+ self,
+ prompt,
+ device,
+ num_images_per_prompt,
+ do_classifier_free_guidance,
+ negative_prompt=None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ **kwargs,
+ ):
+ deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple."
+ deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False)
+
+ prompt_embeds_tuple = self.encode_prompt(
+ prompt=prompt,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ lora_scale=lora_scale,
+ **kwargs,
+ )
+
+ # concatenate for backwards comp
+ prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]])
+
+ return prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt,
+ device,
+ num_images_per_prompt,
+ do_classifier_free_guidance,
+ negative_prompt=None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ lora_scale (`float`, *optional*):
+ A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ # textual inversion: process multi-vector tokens if necessary
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+ text_inputs = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=self.tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = self.tokenizer.batch_decode(
+ untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+ )
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+ attention_mask = text_inputs.attention_mask.to(device)
+ else:
+ attention_mask = None
+
+ if clip_skip is None:
+ prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask)
+ prompt_embeds = prompt_embeds[0]
+ else:
+ prompt_embeds = self.text_encoder(
+ text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True
+ )
+ # Access the `hidden_states` first, that contains a tuple of
+ # all the hidden states from the encoder layers. Then index into
+ # the tuple to access the hidden states from the desired layer.
+ prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
+ # We also need to apply the final LayerNorm here to not mess with the
+ # representations. The `last_hidden_states` that we typically use for
+ # obtaining the final prompt representations passes through the LayerNorm
+ # layer.
+ prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+
+ if self.text_encoder is not None:
+ prompt_embeds_dtype = self.text_encoder.dtype
+ elif self.unet is not None:
+ prompt_embeds_dtype = self.unet.dtype
+ else:
+ prompt_embeds_dtype = prompt_embeds.dtype
+
+ prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ # get unconditional embeddings for classifier free guidance
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ uncond_tokens: List[str]
+ if negative_prompt is None:
+ uncond_tokens = [""] * batch_size
+ elif prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif isinstance(negative_prompt, str):
+ uncond_tokens = [negative_prompt]
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = negative_prompt
+
+ # textual inversion: process multi-vector tokens if necessary
+ if isinstance(self, TextualInversionLoaderMixin):
+ uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = self.tokenizer(
+ uncond_tokens,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+ attention_mask = uncond_input.attention_mask.to(device)
+ else:
+ attention_mask = None
+
+ negative_prompt_embeds = self.text_encoder(
+ uncond_input.input_ids.to(device),
+ attention_mask=attention_mask,
+ )
+ negative_prompt_embeds = negative_prompt_embeds[0]
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+ def run_safety_checker(self, image, device, dtype):
+ if self.safety_checker is None:
+ has_nsfw_concept = None
+ else:
+ if torch.is_tensor(image):
+ feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
+ else:
+ feature_extractor_input = self.image_processor.numpy_to_pil(image)
+ safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
+ image, has_nsfw_concept = self.safety_checker(
+ images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+ )
+ return image, has_nsfw_concept
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.decode_latents
+ def decode_latents(self, latents):
+ deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead"
+ deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False)
+
+ latents = 1 / self.vae.config.scaling_factor * latents
+ image = self.vae.decode(latents, return_dict=False)[0]
+ image = (image / 2 + 0.5).clamp(0, 1)
+ # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
+ image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+ t_start = max(num_inference_steps - init_timestep, 0)
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ def check_inputs(
+ self,
+ prompt,
+ image,
+ mask_image,
+ height,
+ width,
+ callback_steps,
+ output_type,
+ negative_prompt=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ callback_on_step_end_tensor_inputs=None,
+ padding_mask_crop=None,
+ ):
+ if height is not None and height % 8 != 0 or width is not None and width % 8 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if padding_mask_crop is not None:
+ if not isinstance(image, PIL.Image.Image):
+ raise ValueError(
+ f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}."
+ )
+ if not isinstance(mask_image, PIL.Image.Image):
+ raise ValueError(
+ f"The mask image should be a PIL image when inpainting mask crop, but is of type"
+ f" {type(mask_image)}."
+ )
+ if output_type != "pil":
+ raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.")
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
+ self.controlnet, torch._dynamo.eval_frame.OptimizedModule
+ )
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ self.check_image(image, prompt, prompt_embeds)
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+ # When `image` is a nested list:
+ # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
+ elif any(isinstance(i, list) for i in image):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ else:
+ assert False
+
+ # Check `controlnet_conditioning_scale`
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ if not isinstance(controlnet_conditioning_scale, float):
+ raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+ else:
+ assert False
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ crops_coords,
+ resize_mode,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(
+ image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
+ ).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_inpaint.StableDiffusionInpaintPipeline.prepare_latents
+ def prepare_latents(
+ self,
+ batch_size,
+ num_channels_latents,
+ height,
+ width,
+ dtype,
+ device,
+ generator,
+ latents=None,
+ image=None,
+ timestep=None,
+ is_strength_max=True,
+ return_noise=False,
+ return_image_latents=False,
+ ):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if (image is None or timestep is None) and not is_strength_max:
+ raise ValueError(
+ "Since strength < 1. initial latents are to be initialised as a combination of Image + Noise."
+ "However, either the image or the noise timestep has not been provided."
+ )
+
+ if return_image_latents or (latents is None and not is_strength_max):
+ image = image.to(device=device, dtype=dtype)
+
+ if image.shape[1] == 4:
+ image_latents = image
+ else:
+ image_latents = self._encode_vae_image(image=image, generator=generator)
+ image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
+
+ if latents is None:
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ # if strength is 1. then initialise the latents to noise, else initial to image + noise
+ latents = noise if is_strength_max else self.scheduler.add_noise(image_latents, noise, timestep)
+ # if pure noise then scale the initial latents by the Scheduler's init sigma
+ latents = latents * self.scheduler.init_noise_sigma if is_strength_max else latents
+ else:
+ noise = latents.to(device)
+ latents = noise * self.scheduler.init_noise_sigma
+
+ outputs = (latents,)
+
+ if return_noise:
+ outputs += (noise,)
+
+ if return_image_latents:
+ outputs += (image_latents,)
+
+ return outputs
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_inpaint.StableDiffusionInpaintPipeline.prepare_mask_latents
+ def prepare_mask_latents(
+ self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance
+ ):
+ # resize the mask to latents shape as we concatenate the mask to the latents
+ # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
+ # and half precision
+ mask = torch.nn.functional.interpolate(
+ mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor)
+ )
+ mask = mask.to(device=device, dtype=dtype)
+
+ masked_image = masked_image.to(device=device, dtype=dtype)
+
+ if masked_image.shape[1] == 4:
+ masked_image_latents = masked_image
+ else:
+ masked_image_latents = self._encode_vae_image(masked_image, generator=generator)
+
+ # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
+ if mask.shape[0] < batch_size:
+ if not batch_size % mask.shape[0] == 0:
+ raise ValueError(
+ "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
+ f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
+ " of masks that you pass is divisible by the total requested batch size."
+ )
+ mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
+ if masked_image_latents.shape[0] < batch_size:
+ if not batch_size % masked_image_latents.shape[0] == 0:
+ raise ValueError(
+ "The passed images and the required batch size don't match. Images are supposed to be duplicated"
+ f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
+ " Make sure the number of images that you pass is divisible by the total requested batch size."
+ )
+ masked_image_latents = masked_image_latents.repeat(batch_size // masked_image_latents.shape[0], 1, 1, 1)
+
+ mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
+ masked_image_latents = (
+ torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
+ )
+
+ # aligning device to prevent device errors when concating it with the latent model input
+ masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
+ return mask, masked_image_latents
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_inpaint.StableDiffusionInpaintPipeline._encode_vae_image
+ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
+ if isinstance(generator, list):
+ image_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(image.shape[0])
+ ]
+ image_latents = torch.cat(image_latents, dim=0)
+ else:
+ image_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ image_latents = self.vae.config.scaling_factor * image_latents
+
+ return image_latents
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ image: PipelineImageInput = None,
+ mask_image: PipelineImageInput = None,
+ control_image: PipelineImageInput = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ padding_mask_crop: Optional[int] = None,
+ strength: float = 1.0,
+ num_inference_steps: int = 50,
+ guidance_scale: float = 7.5,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 0.5,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+ image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`,
+ `List[PIL.Image.Image]`, or `List[np.ndarray]`):
+ `Image`, NumPy array or tensor representing an image batch to be used as the starting point. For both
+ NumPy array and PyTorch tensor, the expected value range is between `[0, 1]`. If it's a tensor or a
+ list or tensors, the expected shape should be `(B, C, H, W)` or `(C, H, W)`. If it is a NumPy array or
+ a list of arrays, the expected shape should be `(B, H, W, C)` or `(H, W, C)`. It can also accept image
+ latents as `image`, but if passing latents directly it is not encoded again.
+ mask_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`,
+ `List[PIL.Image.Image]`, or `List[np.ndarray]`):
+ `Image`, NumPy array or tensor representing an image batch to mask `image`. White pixels in the mask
+ are repainted while black pixels are preserved. If `mask_image` is a PIL image, it is converted to a
+ single channel (luminance) before use. If it's a NumPy array or PyTorch tensor, it should contain one
+ color channel (L) instead of 3, so the expected shape for PyTorch tensor would be `(B, 1, H, W)`, `(B,
+ H, W)`, `(1, H, W)`, `(H, W)`. And for NumPy array, it would be for `(B, H, W, 1)`, `(B, H, W)`, `(H,
+ W, 1)`, or `(H, W)`.
+ control_image (`torch.Tensor`, `PIL.Image.Image`, `List[torch.Tensor]`, `List[PIL.Image.Image]`,
+ `List[List[torch.Tensor]]`, or `List[List[PIL.Image.Image]]`):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The height in pixels of the generated image.
+ width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The width in pixels of the generated image.
+ padding_mask_crop (`int`, *optional*, defaults to `None`):
+ The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to
+ image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region
+ with the same aspect ration of the image and contains all masked area, and then expand that area based
+ on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before
+ resizing to the original image size for inpainting. This is useful when the masked area is small while
+ the image is large and contain information irrelevant for inpainting, such as background.
+ strength (`float`, *optional*, defaults to 1.0):
+ Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
+ starting point and more noise is added the higher the `strength`. The number of denoising steps depends
+ on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
+ process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
+ essentially ignores `image`.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+ pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+ applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor is generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+ [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 0.5):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ The ControlNet encoder tries to recognize the content of the input image even if you remove all
+ prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+ otherwise a `tuple` is returned where the first element is a list with the generated images and the
+ second element is a list of `bool`s indicating whether the corresponding generated image contains
+ "not-safe-for-work" (nsfw) content.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ control_image,
+ mask_image,
+ height,
+ width,
+ callback_steps,
+ output_type,
+ negative_prompt,
+ prompt_embeds,
+ negative_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ callback_on_step_end_tensor_inputs,
+ padding_mask_crop,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if padding_mask_crop is not None:
+ height, width = self.image_processor.get_default_height_width(image, height, width)
+ crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop)
+ resize_mode = "fill"
+ else:
+ crops_coords = None
+ resize_mode = "default"
+
+ device = self._execution_device
+
+ if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+ prompt,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+ # For classifier free guidance, we need to do two forward passes.
+ # Here we concatenate the unconditional and text embeddings into a single batch
+ # to avoid doing two forward passes
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. Prepare image
+ if isinstance(controlnet, ControlNetModel):
+ control_image = self.prepare_control_image(
+ image=control_image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ crops_coords=crops_coords,
+ resize_mode=resize_mode,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+ elif isinstance(controlnet, MultiControlNetModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ control_image_ = self.prepare_control_image(
+ image=control_image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ crops_coords=crops_coords,
+ resize_mode=resize_mode,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(control_image_)
+
+ control_image = control_images
+ else:
+ assert False
+
+ # 4.1 Preprocess mask and image - resizes image and mask w.r.t height and width
+ original_image = image
+ init_image = self.image_processor.preprocess(
+ image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
+ )
+ init_image = init_image.to(dtype=torch.float32)
+
+ mask = self.mask_processor.preprocess(
+ mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
+ )
+
+ masked_image = init_image * (mask < 0.5)
+ _, _, height, width = init_image.shape
+
+ # 5. Prepare timesteps
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps, num_inference_steps = self.get_timesteps(
+ num_inference_steps=num_inference_steps, strength=strength, device=device
+ )
+ # at which timestep to set the initial noise (n.b. 50% if strength is 0.5)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+ # create a boolean to check if the strength is set to 1. if so then initialise the latents with pure noise
+ is_strength_max = strength == 1.0
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.vae.config.latent_channels
+ num_channels_unet = self.unet.config.in_channels
+ return_image_latents = num_channels_unet == 4
+ latents_outputs = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ image=init_image,
+ timestep=latent_timestep,
+ is_strength_max=is_strength_max,
+ return_noise=True,
+ return_image_latents=return_image_latents,
+ )
+
+ if return_image_latents:
+ latents, noise, image_latents = latents_outputs
+ else:
+ latents, noise = latents_outputs
+
+ # 7. Prepare mask latent variables
+ mask, masked_image_latents = self.prepare_mask_latents(
+ mask,
+ masked_image,
+ batch_size * num_images_per_prompt,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ self.do_classifier_free_guidance,
+ )
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Add image embeds for IP-Adapter
+ added_cond_kwargs = (
+ {"image_embeds": image_embeds}
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None
+ else None
+ )
+
+ # 7.2 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps)
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ # predict the noise residual
+ if num_channels_unet == 9:
+ latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1)
+
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if num_channels_unet == 4:
+ init_latents_proper = image_latents
+ if self.do_classifier_free_guidance:
+ init_mask, _ = mask.chunk(2)
+ else:
+ init_mask = mask
+
+ if i < len(timesteps) - 1:
+ noise_timestep = timesteps[i + 1]
+ init_latents_proper = self.scheduler.add_noise(
+ init_latents_proper, noise, torch.tensor([noise_timestep])
+ )
+
+ latents = (1 - init_mask) * init_latents_proper + init_mask * latents
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ control_image = callback_outputs.pop("control_image", control_image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
+ 0
+ ]
+ image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+ else:
+ image = latents
+ has_nsfw_concept = None
+
+ if has_nsfw_concept is None:
+ do_denormalize = [True] * image.shape[0]
+ else:
+ do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+ image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+ if padding_mask_crop is not None:
+ image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image, has_nsfw_concept)
+
+ return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py
new file mode 100644
index 0000000000000000000000000000000000000000..397ab15715c2d09445451a252441a8cbbea8254c
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_inpaint_sd_xl.py
@@ -0,0 +1,1883 @@
+# Copyright 2025 Harutatsu Akiyama, Jinbin Bai, and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import (
+ CLIPImageProcessor,
+ CLIPTextModel,
+ CLIPTextModelWithProjection,
+ CLIPTokenizer,
+ CLIPVisionModelWithProjection,
+)
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import (
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ TextualInversionLoaderMixin,
+)
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
+from ...models.attention_processor import (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+)
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ is_invisible_watermark_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+if is_invisible_watermark_available():
+ from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+from ...utils import is_torch_xla_available
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # !pip install transformers accelerate
+ >>> from diffusers import StableDiffusionXLControlNetInpaintPipeline, ControlNetModel, DDIMScheduler
+ >>> from diffusers.utils import load_image
+ >>> from PIL import Image
+ >>> import numpy as np
+ >>> import torch
+
+ >>> init_image = load_image(
+ ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy.png"
+ ... )
+ >>> init_image = init_image.resize((1024, 1024))
+
+ >>> generator = torch.Generator(device="cpu").manual_seed(1)
+
+ >>> mask_image = load_image(
+ ... "https://huggingface.co/datasets/diffusers/test-arrays/resolve/main/stable_diffusion_inpaint/boy_mask.png"
+ ... )
+ >>> mask_image = mask_image.resize((1024, 1024))
+
+
+ >>> def make_canny_condition(image):
+ ... image = np.array(image)
+ ... image = cv2.Canny(image, 100, 200)
+ ... image = image[:, :, None]
+ ... image = np.concatenate([image, image, image], axis=2)
+ ... image = Image.fromarray(image)
+ ... return image
+
+
+ >>> control_image = make_canny_condition(init_image)
+
+ >>> controlnet = ControlNetModel.from_pretrained(
+ ... "diffusers/controlnet-canny-sdxl-1.0", torch_dtype=torch.float16
+ ... )
+ >>> pipe = StableDiffusionXLControlNetInpaintPipeline.from_pretrained(
+ ... "stabilityai/stable-diffusion-xl-base-1.0", controlnet=controlnet, torch_dtype=torch.float16
+ ... )
+
+ >>> pipe.enable_model_cpu_offload()
+
+ >>> # generate image
+ >>> image = pipe(
+ ... "a handsome man with ray-ban sunglasses",
+ ... num_inference_steps=20,
+ ... generator=generator,
+ ... eta=1.0,
+ ... image=init_image,
+ ... mask_image=mask_image,
+ ... control_image=control_image,
+ ... ).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
+def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
+ r"""
+ Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+ Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+ Flawed](https://huggingface.co/papers/2305.08891).
+
+ Args:
+ noise_cfg (`torch.Tensor`):
+ The predicted noise tensor for the guided diffusion process.
+ noise_pred_text (`torch.Tensor`):
+ The predicted noise tensor for the text-guided diffusion process.
+ guidance_rescale (`float`, *optional*, defaults to 0.0):
+ A rescale factor applied to the noise predictions.
+
+ Returns:
+ noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
+ """
+ std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
+ std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
+ # rescale the results from guidance (fixes overexposure)
+ noise_pred_rescaled = noise_cfg * (std_text / std_cfg)
+ # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images
+ noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg
+ return noise_cfg
+
+
+class StableDiffusionXLControlNetInpaintPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ TextualInversionLoaderMixin,
+):
+ r"""
+ Pipeline for text-to-image generation using Stable Diffusion XL.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+ text_encoder ([`CLIPTextModel`]):
+ Frozen text-encoder. Stable Diffusion XL uses the text portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+ the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+ text_encoder_2 ([` CLIPTextModelWithProjection`]):
+ Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
+ specifically the
+ [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
+ variant.
+ tokenizer (`CLIPTokenizer`):
+ Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ tokenizer_2 (`CLIPTokenizer`):
+ Second Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ """
+
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->unet->vae"
+
+ _optional_components = [
+ "tokenizer",
+ "tokenizer_2",
+ "text_encoder",
+ "text_encoder_2",
+ "image_encoder",
+ "feature_extractor",
+ ]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ "add_text_embeds",
+ "add_time_ids",
+ "negative_pooled_prompt_embeds",
+ "add_neg_time_ids",
+ "mask",
+ "masked_image_latents",
+ "control_image",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
+ scheduler: KarrasDiffusionSchedulers,
+ requires_aesthetics_score: bool = False,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None,
+ feature_extractor: Optional[CLIPImageProcessor] = None,
+ image_encoder: Optional[CLIPVisionModelWithProjection] = None,
+ ):
+ super().__init__()
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+ self.register_to_config(requires_aesthetics_score=requires_aesthetics_score)
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+ self.mask_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True
+ )
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+
+ add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+ if add_watermarker:
+ self.watermark = StableDiffusionXLWatermarker()
+ else:
+ self.watermark = None
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder_2, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+ # textual inversion: process multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+ pooled_prompt_embeds = prompt_embeds[0]
+
+ if clip_skip is None:
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+ else:
+ # "2" because SDXL always indexes from the penultimate layer.
+ prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ # normalize str to list
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ negative_prompt_2 = (
+ batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
+ )
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ if self.text_encoder_2 is not None:
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ if self.text_encoder_2 is not None:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ image,
+ mask_image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ output_type,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ pooled_prompt_embeds=None,
+ negative_pooled_prompt_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ callback_on_step_end_tensor_inputs=None,
+ padding_mask_crop=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+ if num_inference_steps is None:
+ raise ValueError("`num_inference_steps` cannot be None.")
+ elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0:
+ raise ValueError(
+ f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type"
+ f" {type(num_inference_steps)}."
+ )
+
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if padding_mask_crop is not None:
+ if not isinstance(image, PIL.Image.Image):
+ raise ValueError(
+ f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}."
+ )
+ if not isinstance(mask_image, PIL.Image.Image):
+ raise ValueError(
+ f"The mask image should be a PIL image when inpainting mask crop, but is of type"
+ f" {type(mask_image)}."
+ )
+ if output_type != "pil":
+ raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.")
+
+ if prompt_embeds is not None and pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+ )
+
+ if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
+ self.controlnet, torch._dynamo.eval_frame.OptimizedModule
+ )
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ self.check_image(image, prompt, prompt_embeds)
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+ # When `image` is a nested list:
+ # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
+ elif any(isinstance(i, list) for i in image):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ else:
+ assert False
+
+ # Check `controlnet_conditioning_scale`
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ if not isinstance(controlnet_conditioning_scale, float):
+ raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+ else:
+ assert False
+
+ if not isinstance(control_guidance_start, (tuple, list)):
+ control_guidance_start = [control_guidance_start]
+
+ if not isinstance(control_guidance_end, (tuple, list)):
+ control_guidance_end = [control_guidance_end]
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ crops_coords,
+ resize_mode,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(
+ image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
+ ).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ def prepare_latents(
+ self,
+ batch_size,
+ num_channels_latents,
+ height,
+ width,
+ dtype,
+ device,
+ generator,
+ latents=None,
+ image=None,
+ timestep=None,
+ is_strength_max=True,
+ add_noise=True,
+ return_noise=False,
+ return_image_latents=False,
+ ):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if (image is None or timestep is None) and not is_strength_max:
+ raise ValueError(
+ "Since strength < 1. initial latents are to be initialised as a combination of Image + Noise."
+ "However, either the image or the noise timestep has not been provided."
+ )
+
+ if return_image_latents or (latents is None and not is_strength_max):
+ image = image.to(device=device, dtype=dtype)
+
+ if image.shape[1] == 4:
+ image_latents = image
+ else:
+ image_latents = self._encode_vae_image(image=image, generator=generator)
+ image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
+
+ if latents is None and add_noise:
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ # if strength is 1. then initialise the latents to noise, else initial to image + noise
+ latents = noise if is_strength_max else self.scheduler.add_noise(image_latents, noise, timestep)
+ # if pure noise then scale the initial latents by the Scheduler's init sigma
+ latents = latents * self.scheduler.init_noise_sigma if is_strength_max else latents
+ elif add_noise:
+ noise = latents.to(device)
+ latents = noise * self.scheduler.init_noise_sigma
+ else:
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ latents = image_latents.to(device)
+
+ outputs = (latents,)
+
+ if return_noise:
+ outputs += (noise,)
+
+ if return_image_latents:
+ outputs += (image_latents,)
+
+ return outputs
+
+ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
+ dtype = image.dtype
+ if self.vae.config.force_upcast:
+ image = image.float()
+ self.vae.to(dtype=torch.float32)
+
+ if isinstance(generator, list):
+ image_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(image.shape[0])
+ ]
+ image_latents = torch.cat(image_latents, dim=0)
+ else:
+ image_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ if self.vae.config.force_upcast:
+ self.vae.to(dtype)
+
+ image_latents = image_latents.to(dtype)
+ image_latents = self.vae.config.scaling_factor * image_latents
+
+ return image_latents
+
+ def prepare_mask_latents(
+ self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance
+ ):
+ # resize the mask to latents shape as we concatenate the mask to the latents
+ # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
+ # and half precision
+ mask = torch.nn.functional.interpolate(
+ mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor)
+ )
+ mask = mask.to(device=device, dtype=dtype)
+
+ # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
+ if mask.shape[0] < batch_size:
+ if not batch_size % mask.shape[0] == 0:
+ raise ValueError(
+ "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
+ f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
+ " of masks that you pass is divisible by the total requested batch size."
+ )
+ mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
+
+ mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
+
+ masked_image_latents = None
+ if masked_image is not None:
+ masked_image = masked_image.to(device=device, dtype=dtype)
+ masked_image_latents = self._encode_vae_image(masked_image, generator=generator)
+ if masked_image_latents.shape[0] < batch_size:
+ if not batch_size % masked_image_latents.shape[0] == 0:
+ raise ValueError(
+ "The passed images and the required batch size don't match. Images are supposed to be duplicated"
+ f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
+ " Make sure the number of images that you pass is divisible by the total requested batch size."
+ )
+ masked_image_latents = masked_image_latents.repeat(
+ batch_size // masked_image_latents.shape[0], 1, 1, 1
+ )
+
+ masked_image_latents = (
+ torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
+ )
+
+ # aligning device to prevent device errors when concating it with the latent model input
+ masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
+
+ return mask, masked_image_latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=None):
+ # get the original timestep using init_timestep
+ if denoising_start is None:
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+ t_start = max(num_inference_steps - init_timestep, 0)
+
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ else:
+ # Strength is irrelevant if we directly request a timestep to start at;
+ # that is, strength is determined by the denoising_start instead.
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (denoising_start * self.scheduler.config.num_train_timesteps)
+ )
+ )
+
+ num_inference_steps = (self.scheduler.timesteps < discrete_timestep_cutoff).sum().item()
+ if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
+ # if the scheduler is a 2nd order scheduler we might have to do +1
+ # because `num_inference_steps` might be even given that every timestep
+ # (except the highest one) is duplicated. If `num_inference_steps` is even it would
+ # mean that we cut the timesteps in the middle of the denoising step
+ # (between 1st and 2nd derivative) which leads to incorrect results. By adding 1
+ # we ensure that the denoising process always ends after the 2nd derivate step of the scheduler
+ num_inference_steps = num_inference_steps + 1
+
+ # because t_n+1 >= t_n, we slice the timesteps starting from the end
+ t_start = len(self.scheduler.timesteps) - num_inference_steps
+ timesteps = self.scheduler.timesteps[t_start:]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start)
+ return timesteps, num_inference_steps
+
+ def _get_add_time_ids(
+ self,
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ dtype,
+ text_encoder_projection_dim=None,
+ ):
+ if self.config.requires_aesthetics_score:
+ add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,))
+ add_neg_time_ids = list(original_size + crops_coords_top_left + (negative_aesthetic_score,))
+ else:
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+ add_neg_time_ids = list(original_size + crops_coords_top_left + target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if (
+ expected_add_embed_dim > passed_add_embed_dim
+ and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model."
+ )
+ elif (
+ expected_add_embed_dim < passed_add_embed_dim
+ and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model."
+ )
+ elif expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype)
+
+ return add_time_ids, add_neg_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ image: PipelineImageInput = None,
+ mask_image: PipelineImageInput = None,
+ control_image: Union[
+ PipelineImageInput,
+ List[PipelineImageInput],
+ ] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ padding_mask_crop: Optional[int] = None,
+ strength: float = 0.9999,
+ num_inference_steps: int = 50,
+ denoising_start: Optional[float] = None,
+ denoising_end: Optional[float] = None,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ guidance_rescale: float = 0.0,
+ original_size: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ aesthetic_score: float = 6.0,
+ negative_aesthetic_score: float = 2.5,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ image (`PIL.Image.Image`):
+ `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will
+ be masked out with `mask_image` and repainted according to `prompt`.
+ mask_image (`PIL.Image.Image`):
+ `Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be
+ repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted
+ to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L)
+ instead of 3, so the expected shape would be `(B, H, W, 1)`.
+ height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The height in pixels of the generated image.
+ width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The width in pixels of the generated image.
+ padding_mask_crop (`int`, *optional*, defaults to `None`):
+ The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to
+ image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region
+ with the same aspect ration of the image and contains all masked area, and then expand that area based
+ on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before
+ resizing to the original image size for inpainting. This is useful when the masked area is small while
+ the image is large and contain information irrelevant for inpainting, such as background.
+ strength (`float`, *optional*, defaults to 0.9999):
+ Conceptually, indicates how much to transform the masked portion of the reference `image`. Must be
+ between 0 and 1. `image` will be used as a starting point, adding more noise to it the larger the
+ `strength`. The number of denoising steps depends on the amount of noise initially added. When
+ `strength` is 1, added noise will be maximum and the denoising process will run for the full number of
+ iterations specified in `num_inference_steps`. A value of 1, therefore, essentially ignores the masked
+ portion of the reference `image`. Note that in the case of `denoising_start` being declared as an
+ integer, the value of `strength` will be ignored.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ denoising_start (`float`, *optional*):
+ When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be
+ bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and
+ it is assumed that the passed `image` is a partly denoised image. Note that when this is specified,
+ strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline
+ is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
+ denoising_end (`float`, *optional*):
+ When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
+ completed before it is intentionally prematurely terminated. As a result, the returned sample will
+ still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be
+ denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the
+ final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline
+ forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+ applies to [`schedulers.DDIMScheduler`], will be ignored for others.
+ generator (`torch.Generator`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(width, height)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ aesthetic_score (`float`, *optional*, defaults to 6.0):
+ Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_aesthetic_score (`float`, *optional*, defaults to 2.5):
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
+ simulate an aesthetic score of the generated image by influencing the negative text condition.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] or `tuple`:
+ [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple. `tuple. When returning a tuple, the first element is a list with the generated images.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # # 0.0 Default height and width to unet
+ # height = height or self.unet.config.sample_size * self.vae_scale_factor
+ # width = width or self.unet.config.sample_size * self.vae_scale_factor
+
+ # 0.1 align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # 1. Check inputs
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ control_image,
+ mask_image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ output_type,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ callback_on_step_end_tensor_inputs,
+ padding_mask_crop,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+ # 3. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ prompt_2=prompt_2,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ negative_prompt_2=negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+
+ # 3.1 Encode ip_adapter_image
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. set timesteps
+ def denoising_value_valid(dnv):
+ return isinstance(dnv, float) and 0 < dnv < 1
+
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps, num_inference_steps = self.get_timesteps(
+ num_inference_steps,
+ strength,
+ device,
+ denoising_start=denoising_start if denoising_value_valid(denoising_start) else None,
+ )
+ # check that number of inference steps is not < 1 - as this doesn't make sense
+ if num_inference_steps < 1:
+ raise ValueError(
+ f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
+ f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
+ )
+ # at which timestep to set the initial noise (n.b. 50% if strength is 0.5)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+ # create a boolean to check if the strength is set to 1. if so then initialise the latents with pure noise
+ is_strength_max = strength == 1.0
+ self._num_timesteps = len(timesteps)
+
+ # 5. Preprocess mask and image - resizes image and mask w.r.t height and width
+ # 5.1 Prepare init image
+ if padding_mask_crop is not None:
+ height, width = self.image_processor.get_default_height_width(image, height, width)
+ crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop)
+ resize_mode = "fill"
+ else:
+ crops_coords = None
+ resize_mode = "default"
+
+ original_image = image
+ init_image = self.image_processor.preprocess(
+ image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
+ )
+ init_image = init_image.to(dtype=torch.float32)
+
+ # 5.2 Prepare control images
+ if isinstance(controlnet, ControlNetModel):
+ control_image = self.prepare_control_image(
+ image=control_image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ crops_coords=crops_coords,
+ resize_mode=resize_mode,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+ elif isinstance(controlnet, MultiControlNetModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ control_image_ = self.prepare_control_image(
+ image=control_image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ crops_coords=crops_coords,
+ resize_mode=resize_mode,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(control_image_)
+
+ control_image = control_images
+ else:
+ raise ValueError(f"{controlnet.__class__} is not supported.")
+
+ # 5.3 Prepare mask
+ mask = self.mask_processor.preprocess(
+ mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
+ )
+
+ masked_image = init_image * (mask < 0.5)
+ _, _, height, width = init_image.shape
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.vae.config.latent_channels
+ num_channels_unet = self.unet.config.in_channels
+ return_image_latents = num_channels_unet == 4
+
+ add_noise = True if denoising_start is None else False
+ latents_outputs = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ image=init_image,
+ timestep=latent_timestep,
+ is_strength_max=is_strength_max,
+ add_noise=add_noise,
+ return_noise=True,
+ return_image_latents=return_image_latents,
+ )
+
+ if return_image_latents:
+ latents, noise, image_latents = latents_outputs
+ else:
+ latents, noise = latents_outputs
+
+ # 7. Prepare mask latent variables
+ mask, masked_image_latents = self.prepare_mask_latents(
+ mask,
+ masked_image,
+ batch_size * num_images_per_prompt,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ self.do_classifier_free_guidance,
+ )
+
+ # 8. Check that sizes of mask, masked image and latents match
+ if num_channels_unet == 9:
+ # default case for stable-diffusion-v1-5/stable-diffusion-inpainting
+ num_channels_mask = mask.shape[1]
+ num_channels_masked_image = masked_image_latents.shape[1]
+ if num_channels_latents + num_channels_mask + num_channels_masked_image != self.unet.config.in_channels:
+ raise ValueError(
+ f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects"
+ f" {self.unet.config.in_channels} but received `num_channels_latents`: {num_channels_latents} +"
+ f" `num_channels_mask`: {num_channels_mask} + `num_channels_masked_image`: {num_channels_masked_image}"
+ f" = {num_channels_latents + num_channels_masked_image + num_channels_mask}. Please verify the config of"
+ " `pipeline.unet` or your `mask_image` or `image` input."
+ )
+ elif num_channels_unet != 4:
+ raise ValueError(
+ f"The unet {self.unet.__class__} should have either 4 or 9 input channels, not {self.unet.config.in_channels}."
+ )
+ # 8.1 Prepare extra step kwargs.
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 8.2 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps if isinstance(controlnet, MultiControlNetModel) else keeps[0])
+
+ # 9. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ height, width = latents.shape[-2:]
+ height = height * self.vae_scale_factor
+ width = width * self.vae_scale_factor
+
+ original_size = original_size or (height, width)
+ target_size = target_size or (height, width)
+
+ # 10. Prepare added time ids & embeddings
+ add_text_embeds = pooled_prompt_embeds
+ if self.text_encoder_2 is None:
+ text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+ else:
+ text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+ add_time_ids, add_neg_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+ add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_neg_time_ids = add_neg_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+ add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device)
+
+ # 11. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+ if (
+ denoising_end is not None
+ and denoising_start is not None
+ and denoising_value_valid(denoising_end)
+ and denoising_value_valid(denoising_start)
+ and denoising_start >= denoising_end
+ ):
+ raise ValueError(
+ f"`denoising_start`: {denoising_start} cannot be larger than or equal to `denoising_end`: "
+ + f" {denoising_end} when using type float."
+ )
+ elif denoising_end is not None and denoising_value_valid(denoising_end):
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (denoising_end * self.scheduler.config.num_train_timesteps)
+ )
+ )
+ num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
+ timesteps = timesteps[:num_inference_steps]
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+
+ # concat latents, mask, masked_image_latents in the channel dimension
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ controlnet_added_cond_kwargs = {
+ "text_embeds": add_text_embeds.chunk(2)[1],
+ "time_ids": add_time_ids.chunk(2)[1],
+ }
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+ controlnet_added_cond_kwargs = added_cond_kwargs
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ # # Resize control_image to match the size of the input to the controlnet
+ # if control_image.shape[-2:] != control_model_input.shape[-2:]:
+ # control_image = F.interpolate(control_image, size=control_model_input.shape[-2:], mode="bilinear", align_corners=False)
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ added_cond_kwargs=controlnet_added_cond_kwargs,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ added_cond_kwargs["image_embeds"] = image_embeds
+
+ if num_channels_unet == 9:
+ latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1)
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ if self.do_classifier_free_guidance and guidance_rescale > 0.0:
+ # Based on 3.4. in https://huggingface.co/papers/2305.08891
+ noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if num_channels_unet == 4:
+ init_latents_proper = image_latents
+ if self.do_classifier_free_guidance:
+ init_mask, _ = mask.chunk(2)
+ else:
+ init_mask = mask
+
+ if i < len(timesteps) - 1:
+ noise_timestep = timesteps[i + 1]
+ init_latents_proper = self.scheduler.add_noise(
+ init_latents_proper, noise, torch.tensor([noise_timestep])
+ )
+
+ latents = (1 - init_mask) * init_latents_proper + init_mask * latents
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ control_image = callback_outputs.pop("control_image", control_image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+ else:
+ return StableDiffusionXLPipelineOutput(images=latents)
+
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ if padding_mask_crop is not None:
+ image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py
new file mode 100644
index 0000000000000000000000000000000000000000..8b37d38b909924d308ca343eb78a0cb98955262a
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl.py
@@ -0,0 +1,1616 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import (
+ CLIPImageProcessor,
+ CLIPTextModel,
+ CLIPTextModelWithProjection,
+ CLIPTokenizer,
+ CLIPVisionModelWithProjection,
+)
+
+from diffusers.utils.import_utils import is_invisible_watermark_available
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import (
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ TextualInversionLoaderMixin,
+)
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
+from ...models.attention_processor import (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+)
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+if is_invisible_watermark_available():
+ from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+from ...utils import is_torch_xla_available
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # !pip install opencv-python transformers accelerate
+ >>> from diffusers import StableDiffusionXLControlNetPipeline, ControlNetModel, AutoencoderKL
+ >>> from diffusers.utils import load_image
+ >>> import numpy as np
+ >>> import torch
+
+ >>> import cv2
+ >>> from PIL import Image
+
+ >>> prompt = "aerial view, a futuristic research complex in a bright foggy jungle, hard lighting"
+ >>> negative_prompt = "low quality, bad quality, sketches"
+
+ >>> # download an image
+ >>> image = load_image(
+ ... "https://hf.co/datasets/hf-internal-testing/diffusers-images/resolve/main/sd_controlnet/hf-logo.png"
+ ... )
+
+ >>> # initialize the models and pipeline
+ >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization
+ >>> controlnet = ControlNetModel.from_pretrained(
+ ... "diffusers/controlnet-canny-sdxl-1.0", torch_dtype=torch.float16
+ ... )
+ >>> vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+ >>> pipe = StableDiffusionXLControlNetPipeline.from_pretrained(
+ ... "stabilityai/stable-diffusion-xl-base-1.0", controlnet=controlnet, vae=vae, torch_dtype=torch.float16
+ ... )
+ >>> pipe.enable_model_cpu_offload()
+
+ >>> # get canny image
+ >>> image = np.array(image)
+ >>> image = cv2.Canny(image, 100, 200)
+ >>> image = image[:, :, None]
+ >>> image = np.concatenate([image, image, image], axis=2)
+ >>> canny_image = Image.fromarray(image)
+
+ >>> # generate image
+ >>> image = pipe(
+ ... prompt, controlnet_conditioning_scale=controlnet_conditioning_scale, image=canny_image
+ ... ).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class StableDiffusionXLControlNetPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ IPAdapterMixin,
+ FromSingleFileMixin,
+):
+ r"""
+ Pipeline for text-to-image generation using Stable Diffusion XL with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+ text_encoder ([`~transformers.CLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ text_encoder_2 ([`~transformers.CLIPTextModelWithProjection`]):
+ Second frozen text-encoder
+ ([laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)).
+ tokenizer ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ tokenizer_2 ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ unet ([`UNet2DConditionModel`]):
+ A `UNet2DConditionModel` to denoise the encoded image latents.
+ controlnet ([`ControlNetModel`] or `List[ControlNetModel]`):
+ Provides additional conditioning to the `unet` during the denoising process. If you set multiple
+ ControlNets as a list, the outputs from each ControlNet are added together to create one combined
+ additional conditioning.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`):
+ Whether the negative prompt embeddings should always be set to 0. Also see the config of
+ `stabilityai/stable-diffusion-xl-base-1-0`.
+ add_watermarker (`bool`, *optional*):
+ Whether to use the [invisible_watermark](https://github.com/ShieldMnt/invisible-watermark/) library to
+ watermark output images. If not defined, it defaults to `True` if the package is installed; otherwise no
+ watermarker is used.
+ """
+
+ # leave controlnet out on purpose because it iterates with unet
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->unet->vae"
+ _optional_components = [
+ "tokenizer",
+ "tokenizer_2",
+ "text_encoder",
+ "text_encoder_2",
+ "feature_extractor",
+ "image_encoder",
+ ]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ "add_text_embeds",
+ "add_time_ids",
+ "negative_pooled_prompt_embeds",
+ "negative_add_time_ids",
+ "image",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
+ scheduler: KarrasDiffusionSchedulers,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None,
+ feature_extractor: CLIPImageProcessor = None,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ ):
+ super().__init__()
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+ if add_watermarker:
+ self.watermark = StableDiffusionXLWatermarker()
+ else:
+ self.watermark = None
+
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder_2, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+ # textual inversion: process multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+ pooled_prompt_embeds = prompt_embeds[0]
+
+ if clip_skip is None:
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+ else:
+ # "2" because SDXL always indexes from the penultimate layer.
+ prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ # normalize str to list
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ negative_prompt_2 = (
+ batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
+ )
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ if self.text_encoder_2 is not None:
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ if self.text_encoder_2 is not None:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ image,
+ callback_steps,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ pooled_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ negative_pooled_prompt_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if prompt_embeds is not None and pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+ )
+
+ if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
+ self.controlnet, torch._dynamo.eval_frame.OptimizedModule
+ )
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ self.check_image(image, prompt, prompt_embeds)
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+ # When `image` is a nested list:
+ # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
+ elif any(isinstance(i, list) for i in image):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ else:
+ assert False
+
+ # Check `controlnet_conditioning_scale`
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ if not isinstance(controlnet_conditioning_scale, float):
+ raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+ else:
+ assert False
+
+ if not isinstance(control_guidance_start, (tuple, list)):
+ control_guidance_start = [control_guidance_start]
+
+ if not isinstance(control_guidance_end, (tuple, list)):
+ control_guidance_end = [control_guidance_end]
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.prepare_image
+ def prepare_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids
+ def _get_add_time_ids(
+ self, original_size, crops_coords_top_left, target_size, dtype, text_encoder_projection_dim=None
+ ):
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ return add_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
+ def get_guidance_scale_embedding(
+ self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+ ) -> torch.Tensor:
+ """
+ See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+
+ Args:
+ w (`torch.Tensor`):
+ Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
+ embedding_dim (`int`, *optional*, defaults to 512):
+ Dimension of the embeddings to generate.
+ dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
+ Data type of the generated embeddings.
+
+ Returns:
+ `torch.Tensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
+ """
+ assert len(w.shape) == 1
+ w = w * 1000.0
+
+ half_dim = embedding_dim // 2
+ emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
+ emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
+ emb = w.to(dtype)[:, None] * emb[None, :]
+ emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+ if embedding_dim % 2 == 1: # zero pad
+ emb = torch.nn.functional.pad(emb, (0, 1))
+ assert emb.shape == (w.shape[0], embedding_dim)
+ return emb
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def denoising_end(self):
+ return self._denoising_end
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ image: PipelineImageInput = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 50,
+ timesteps: List[int] = None,
+ sigmas: List[float] = None,
+ denoising_end: Optional[float] = None,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ original_size: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ negative_original_size: Optional[Tuple[int, int]] = None,
+ negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
+ negative_target_size: Optional[Tuple[int, int]] = None,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders.
+ image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The height in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The width in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+ their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+ will be used.
+ denoising_end (`float`, *optional*):
+ When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
+ completed before it is intentionally prematurely terminated. As a result, the returned sample will
+ still retain a substantial amount of noise as determined by the discrete timesteps selected by the
+ scheduler. The denoising_end parameter should ideally be utilized when this pipeline forms a part of a
+ "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output)
+ guidance_scale (`float`, *optional*, defaults to 5.0):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+ pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. This is sent to `tokenizer_2`
+ and `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders.
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+ applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor is generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, pooled text embeddings are generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs (prompt
+ weighting). If not provided, pooled `negative_prompt_embeds` are generated from `negative_prompt` input
+ argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+ [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ The ControlNet encoder tries to recognize the content of the input image even if you remove all
+ prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a specific image resolution. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a target image resolution. It should be as same
+ as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+ otherwise a `tuple` is returned containing the output images.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ image,
+ callback_steps,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ negative_pooled_prompt_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ callback_on_step_end_tensor_inputs,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._denoising_end = denoising_end
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3.1 Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt,
+ prompt_2,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+
+ # 3.2 Encode ip_adapter_image
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. Prepare image
+ if isinstance(controlnet, ControlNetModel):
+ image = self.prepare_image(
+ image=image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+ height, width = image.shape[-2:]
+ elif isinstance(controlnet, MultiControlNetModel):
+ images = []
+
+ for image_ in image:
+ image_ = self.prepare_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ images.append(image_)
+
+ image = images
+ height, width = image[0].shape[-2:]
+ else:
+ assert False
+
+ # 5. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(
+ self.scheduler, num_inference_steps, device, timesteps, sigmas
+ )
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.unet.config.in_channels
+ latents = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 6.5 Optionally get Guidance Scale Embedding
+ timestep_cond = None
+ if self.unet.config.time_cond_proj_dim is not None:
+ guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+ timestep_cond = self.get_guidance_scale_embedding(
+ guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+ ).to(device=device, dtype=latents.dtype)
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps)
+
+ # 7.2 Prepare added time ids & embeddings
+ if isinstance(image, list):
+ original_size = original_size or image[0].shape[-2:]
+ else:
+ original_size = original_size or image.shape[-2:]
+ target_size = target_size or (height, width)
+
+ add_text_embeds = pooled_prompt_embeds
+ if self.text_encoder_2 is None:
+ text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+ else:
+ text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+ add_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+
+ if negative_original_size is not None and negative_target_size is not None:
+ negative_add_time_ids = self._get_add_time_ids(
+ negative_original_size,
+ negative_crops_coords_top_left,
+ negative_target_size,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+ else:
+ negative_add_time_ids = add_time_ids
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+
+ # 8.1 Apply denoising_end
+ if (
+ self.denoising_end is not None
+ and isinstance(self.denoising_end, float)
+ and self.denoising_end > 0
+ and self.denoising_end < 1
+ ):
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (self.denoising_end * self.scheduler.config.num_train_timesteps)
+ )
+ )
+ num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
+ timesteps = timesteps[:num_inference_steps]
+
+ is_unet_compiled = is_compiled_module(self.unet)
+ is_controlnet_compiled = is_compiled_module(self.controlnet)
+ is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1")
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # Relevant thread:
+ # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428
+ if (
+ torch.cuda.is_available()
+ and (is_unet_compiled and is_controlnet_compiled)
+ and is_torch_higher_equal_2_1
+ ):
+ torch._inductor.cudagraph_mark_step_begin()
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ controlnet_added_cond_kwargs = {
+ "text_embeds": add_text_embeds.chunk(2)[1],
+ "time_ids": add_time_ids.chunk(2)[1],
+ }
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+ controlnet_added_cond_kwargs = added_cond_kwargs
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=image,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ added_cond_kwargs=controlnet_added_cond_kwargs,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ added_cond_kwargs["image_embeds"] = image_embeds
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ timestep_cond=timestep_cond,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
+ negative_pooled_prompt_embeds = callback_outputs.pop(
+ "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds
+ )
+ add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
+ negative_add_time_ids = callback_outputs.pop("negative_add_time_ids", negative_add_time_ids)
+ image = callback_outputs.pop("image", image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ if not output_type == "latent":
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+
+ if needs_upcasting:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # unscale/denormalize the latents
+ # denormalize with the mean and std if available and not None
+ has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None
+ has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None
+ if has_latents_mean and has_latents_std:
+ latents_mean = (
+ torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents_std = (
+ torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean
+ else:
+ latents = latents / self.vae.config.scaling_factor
+
+ image = self.vae.decode(latents, return_dict=False)[0]
+
+ # cast back to fp16 if needed
+ if needs_upcasting:
+ self.vae.to(dtype=torch.float16)
+ else:
+ image = latents
+
+ if not output_type == "latent":
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py
new file mode 100644
index 0000000000000000000000000000000000000000..4d4845c5a0a3042e7fbe495cdc4e4bf1691239c1
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_sd_xl_img2img.py
@@ -0,0 +1,1681 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from transformers import (
+ CLIPImageProcessor,
+ CLIPTextModel,
+ CLIPTextModelWithProjection,
+ CLIPTokenizer,
+ CLIPVisionModelWithProjection,
+)
+
+from diffusers.utils.import_utils import is_invisible_watermark_available
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import (
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ TextualInversionLoaderMixin,
+)
+from ...models import AutoencoderKL, ControlNetModel, ImageProjection, MultiControlNetModel, UNet2DConditionModel
+from ...models.attention_processor import (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+)
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+if is_invisible_watermark_available():
+ from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+from ...utils import is_torch_xla_available
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # pip install accelerate transformers safetensors diffusers
+
+ >>> import torch
+ >>> import numpy as np
+ >>> from PIL import Image
+
+ >>> from transformers import DPTImageProcessor, DPTForDepthEstimation
+ >>> from diffusers import ControlNetModel, StableDiffusionXLControlNetImg2ImgPipeline, AutoencoderKL
+ >>> from diffusers.utils import load_image
+
+
+ >>> depth_estimator = DPTForDepthEstimation.from_pretrained("Intel/dpt-hybrid-midas").to("cuda")
+ >>> feature_extractor = DPTImageProcessor.from_pretrained("Intel/dpt-hybrid-midas")
+ >>> controlnet = ControlNetModel.from_pretrained(
+ ... "diffusers/controlnet-depth-sdxl-1.0-small",
+ ... variant="fp16",
+ ... use_safetensors=True,
+ ... torch_dtype=torch.float16,
+ ... )
+ >>> vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+ >>> pipe = StableDiffusionXLControlNetImg2ImgPipeline.from_pretrained(
+ ... "stabilityai/stable-diffusion-xl-base-1.0",
+ ... controlnet=controlnet,
+ ... vae=vae,
+ ... variant="fp16",
+ ... use_safetensors=True,
+ ... torch_dtype=torch.float16,
+ ... )
+ >>> pipe.enable_model_cpu_offload()
+
+
+ >>> def get_depth_map(image):
+ ... image = feature_extractor(images=image, return_tensors="pt").pixel_values.to("cuda")
+ ... with torch.no_grad(), torch.autocast("cuda"):
+ ... depth_map = depth_estimator(image).predicted_depth
+
+ ... depth_map = torch.nn.functional.interpolate(
+ ... depth_map.unsqueeze(1),
+ ... size=(1024, 1024),
+ ... mode="bicubic",
+ ... align_corners=False,
+ ... )
+ ... depth_min = torch.amin(depth_map, dim=[1, 2, 3], keepdim=True)
+ ... depth_max = torch.amax(depth_map, dim=[1, 2, 3], keepdim=True)
+ ... depth_map = (depth_map - depth_min) / (depth_max - depth_min)
+ ... image = torch.cat([depth_map] * 3, dim=1)
+ ... image = image.permute(0, 2, 3, 1).cpu().numpy()[0]
+ ... image = Image.fromarray((image * 255.0).clip(0, 255).astype(np.uint8))
+ ... return image
+
+
+ >>> prompt = "A robot, 4k photo"
+ >>> image = load_image(
+ ... "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
+ ... "/kandinsky/cat.png"
+ ... ).resize((1024, 1024))
+ >>> controlnet_conditioning_scale = 0.5 # recommended for good generalization
+ >>> depth_image = get_depth_map(image)
+
+ >>> images = pipe(
+ ... prompt,
+ ... image=image,
+ ... control_image=depth_image,
+ ... strength=0.99,
+ ... num_inference_steps=50,
+ ... controlnet_conditioning_scale=controlnet_conditioning_scale,
+ ... ).images
+ >>> images[0].save(f"robot_cat.png")
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class StableDiffusionXLControlNetImg2ImgPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ FromSingleFileMixin,
+ IPAdapterMixin,
+):
+ r"""
+ Pipeline for image-to-image generation using Stable Diffusion XL with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+ text_encoder ([`CLIPTextModel`]):
+ Frozen text-encoder. Stable Diffusion uses the text portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+ the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+ text_encoder_2 ([` CLIPTextModelWithProjection`]):
+ Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
+ specifically the
+ [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
+ variant.
+ tokenizer (`CLIPTokenizer`):
+ Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ tokenizer_2 (`CLIPTokenizer`):
+ Second Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+ controlnet ([`ControlNetModel`] or `List[ControlNetModel]`):
+ Provides additional conditioning to the unet during the denoising process. If you set multiple ControlNets
+ as a list, the outputs from each ControlNet are added together to create one combined additional
+ conditioning.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ requires_aesthetics_score (`bool`, *optional*, defaults to `"False"`):
+ Whether the `unet` requires an `aesthetic_score` condition to be passed during inference. Also see the
+ config of `stabilityai/stable-diffusion-xl-refiner-1-0`.
+ force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`):
+ Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of
+ `stabilityai/stable-diffusion-xl-base-1-0`.
+ add_watermarker (`bool`, *optional*):
+ Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to
+ watermark output images. If not defined, it will default to True if the package is installed, otherwise no
+ watermarker will be used.
+ feature_extractor ([`~transformers.CLIPImageProcessor`]):
+ A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+ """
+
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->unet->vae"
+ _optional_components = [
+ "tokenizer",
+ "tokenizer_2",
+ "text_encoder",
+ "text_encoder_2",
+ "feature_extractor",
+ "image_encoder",
+ ]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ "add_text_embeds",
+ "add_time_ids",
+ "negative_pooled_prompt_embeds",
+ "add_neg_time_ids",
+ "control_image",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[ControlNetModel, List[ControlNetModel], Tuple[ControlNetModel], MultiControlNetModel],
+ scheduler: KarrasDiffusionSchedulers,
+ requires_aesthetics_score: bool = False,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None,
+ feature_extractor: CLIPImageProcessor = None,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ ):
+ super().__init__()
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+ if add_watermarker:
+ self.watermark = StableDiffusionXLWatermarker()
+ else:
+ self.watermark = None
+
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+ self.register_to_config(requires_aesthetics_score=requires_aesthetics_score)
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder_2, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+ # textual inversion: process multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+ pooled_prompt_embeds = prompt_embeds[0]
+
+ if clip_skip is None:
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+ else:
+ # "2" because SDXL always indexes from the penultimate layer.
+ prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ # normalize str to list
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ negative_prompt_2 = (
+ batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
+ )
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ if self.text_encoder_2 is not None:
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ if self.text_encoder_2 is not None:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ pooled_prompt_embeds=None,
+ negative_pooled_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+ if num_inference_steps is None:
+ raise ValueError("`num_inference_steps` cannot be None.")
+ elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0:
+ raise ValueError(
+ f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type"
+ f" {type(num_inference_steps)}."
+ )
+
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if prompt_embeds is not None and pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+ )
+
+ if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ is_compiled = hasattr(F, "scaled_dot_product_attention") and isinstance(
+ self.controlnet, torch._dynamo.eval_frame.OptimizedModule
+ )
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ self.check_image(image, prompt, prompt_embeds)
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+
+ # When `image` is a nested list:
+ # (e.g. [[canny_image_1, pose_image_1], [canny_image_2, pose_image_2]])
+ elif any(isinstance(i, list) for i in image):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ else:
+ assert False
+
+ # Check `controlnet_conditioning_scale`
+ if (
+ isinstance(self.controlnet, ControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, ControlNetModel)
+ ):
+ if not isinstance(controlnet_conditioning_scale, float):
+ raise TypeError("For single controlnet: `controlnet_conditioning_scale` must be type `float`.")
+ elif (
+ isinstance(self.controlnet, MultiControlNetModel)
+ or is_compiled
+ and isinstance(self.controlnet._orig_mod, MultiControlNetModel)
+ ):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError("A single batch of multiple conditionings are supported at the moment.")
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+ else:
+ assert False
+
+ if not isinstance(control_guidance_start, (tuple, list)):
+ control_guidance_start = [control_guidance_start]
+
+ if not isinstance(control_guidance_end, (tuple, list)):
+ control_guidance_end = [control_guidance_end]
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(self.controlnet, MultiControlNetModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+ t_start = max(num_inference_steps - init_timestep, 0)
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.prepare_latents
+ def prepare_latents(
+ self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True
+ ):
+ if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
+ raise ValueError(
+ f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+ )
+
+ latents_mean = latents_std = None
+ if hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None:
+ latents_mean = torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1)
+ if hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None:
+ latents_std = torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1)
+
+ # Offload text encoder if `enable_model_cpu_offload` was enabled
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.text_encoder_2.to("cpu")
+ empty_device_cache()
+
+ image = image.to(device=device, dtype=dtype)
+
+ batch_size = batch_size * num_images_per_prompt
+
+ if image.shape[1] == 4:
+ init_latents = image
+
+ else:
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ if self.vae.config.force_upcast:
+ image = image.float()
+ self.vae.to(dtype=torch.float32)
+
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ elif isinstance(generator, list):
+ if image.shape[0] < batch_size and batch_size % image.shape[0] == 0:
+ image = torch.cat([image] * (batch_size // image.shape[0]), dim=0)
+ elif image.shape[0] < batch_size and batch_size % image.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {image.shape[0]} to effective batch_size {batch_size} "
+ )
+
+ init_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(batch_size)
+ ]
+ init_latents = torch.cat(init_latents, dim=0)
+ else:
+ init_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ if self.vae.config.force_upcast:
+ self.vae.to(dtype)
+
+ init_latents = init_latents.to(dtype)
+ if latents_mean is not None and latents_std is not None:
+ latents_mean = latents_mean.to(device=device, dtype=dtype)
+ latents_std = latents_std.to(device=device, dtype=dtype)
+ init_latents = (init_latents - latents_mean) * self.vae.config.scaling_factor / latents_std
+ else:
+ init_latents = self.vae.config.scaling_factor * init_latents
+
+ if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
+ # expand init_latents for batch_size
+ additional_image_per_prompt = batch_size // init_latents.shape[0]
+ init_latents = torch.cat([init_latents] * additional_image_per_prompt, dim=0)
+ elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+ )
+ else:
+ init_latents = torch.cat([init_latents], dim=0)
+
+ if add_noise:
+ shape = init_latents.shape
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ # get latents
+ init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+
+ latents = init_latents
+
+ return latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids
+ def _get_add_time_ids(
+ self,
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ negative_original_size,
+ negative_crops_coords_top_left,
+ negative_target_size,
+ dtype,
+ text_encoder_projection_dim=None,
+ ):
+ if self.config.requires_aesthetics_score:
+ add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,))
+ add_neg_time_ids = list(
+ negative_original_size + negative_crops_coords_top_left + (negative_aesthetic_score,)
+ )
+ else:
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+ add_neg_time_ids = list(negative_original_size + crops_coords_top_left + negative_target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if (
+ expected_add_embed_dim > passed_add_embed_dim
+ and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model."
+ )
+ elif (
+ expected_add_embed_dim < passed_add_embed_dim
+ and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model."
+ )
+ elif expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype)
+
+ return add_time_ids, add_neg_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ image: PipelineImageInput = None,
+ control_image: PipelineImageInput = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ strength: float = 0.8,
+ num_inference_steps: int = 50,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 0.8,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ original_size: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ negative_original_size: Optional[Tuple[int, int]] = None,
+ negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
+ negative_target_size: Optional[Tuple[int, int]] = None,
+ aesthetic_score: float = 6.0,
+ negative_aesthetic_score: float = 2.5,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The initial image will be used as the starting point for the image generation process. Can also accept
+ image latents as `image`, if passing latents directly, it will not be encoded again.
+ control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The ControlNet input condition. ControlNet uses this input condition to generate guidance to Unet. If
+ the type is specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also
+ be accepted as an image. The dimensions of the output image defaults to `image`'s dimensions. If height
+ and/or width are passed, `image` is resized according to them. If multiple ControlNets are specified in
+ init, images must be passed as a list such that each element of the list can be correctly batched for
+ input to a single controlnet.
+ height (`int`, *optional*, defaults to the size of control_image):
+ The height in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ width (`int`, *optional*, defaults to the size of control_image):
+ The width in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ strength (`float`, *optional*, defaults to 0.8):
+ Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
+ starting point and more noise is added the higher the `strength`. The number of denoising steps depends
+ on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
+ process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
+ essentially ignores `image`.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+ applies to [`schedulers.DDIMScheduler`], will be ignored for others.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the controlnet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original unet. If multiple ControlNets are specified in init, you can set the
+ corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ In this mode, the ControlNet encoder will try best to recognize the content of the input image even if
+ you remove all prompts. The `guidance_scale` between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the controlnet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the controlnet stops applying.
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a specific image resolution. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a target image resolution. It should be as same
+ as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ aesthetic_score (`float`, *optional*, defaults to 6.0):
+ Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_aesthetic_score (`float`, *optional*, defaults to 2.5):
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
+ simulate an aesthetic score of the generated image by influencing the negative text condition.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple`
+ containing the output images.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetModel) else 1
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ control_image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ callback_on_step_end_tensor_inputs,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ if isinstance(controlnet, MultiControlNetModel) and isinstance(controlnet_conditioning_scale, float):
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * len(controlnet.nets)
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3.1. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt,
+ prompt_2,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+
+ # 3.2 Encode ip_adapter_image
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. Prepare image and controlnet_conditioning_image
+ image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+
+ if isinstance(controlnet, ControlNetModel):
+ control_image = self.prepare_control_image(
+ image=control_image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+ height, width = control_image.shape[-2:]
+ elif isinstance(controlnet, MultiControlNetModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ control_image_ = self.prepare_control_image(
+ image=control_image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(control_image_)
+
+ control_image = control_images
+ height, width = control_image[0].shape[-2:]
+ else:
+ assert False
+
+ # 5. Prepare timesteps
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ if latents is None:
+ latents = self.prepare_latents(
+ image,
+ latent_timestep,
+ batch_size,
+ num_images_per_prompt,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ True,
+ )
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps[0] if isinstance(controlnet, ControlNetModel) else keeps)
+
+ # 7.2 Prepare added time ids & embeddings
+ if isinstance(control_image, list):
+ original_size = original_size or control_image[0].shape[-2:]
+ else:
+ original_size = original_size or control_image.shape[-2:]
+ target_size = target_size or (height, width)
+
+ if negative_original_size is None:
+ negative_original_size = original_size
+ if negative_target_size is None:
+ negative_target_size = target_size
+ add_text_embeds = pooled_prompt_embeds
+
+ if self.text_encoder_2 is None:
+ text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+ else:
+ text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+ add_time_ids, add_neg_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ negative_original_size,
+ negative_crops_coords_top_left,
+ negative_target_size,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+ add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_neg_time_ids = add_neg_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+ add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device)
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ controlnet_added_cond_kwargs = {
+ "text_embeds": add_text_embeds.chunk(2)[1],
+ "time_ids": add_time_ids.chunk(2)[1],
+ }
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+ controlnet_added_cond_kwargs = added_cond_kwargs
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ added_cond_kwargs=controlnet_added_cond_kwargs,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ added_cond_kwargs["image_embeds"] = image_embeds
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
+ negative_pooled_prompt_embeds = callback_outputs.pop(
+ "negative_pooled_prompt_embeds", negative_pooled_prompt_embeds
+ )
+ add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
+ add_neg_time_ids = callback_outputs.pop("add_neg_time_ids", add_neg_time_ids)
+ control_image = callback_outputs.pop("control_image", control_image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+
+ if needs_upcasting:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # unscale/denormalize the latents
+ # denormalize with the mean and std if available and not None
+ has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None
+ has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None
+ if has_latents_mean and has_latents_std:
+ latents_mean = (
+ torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents_std = (
+ torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean
+ else:
+ latents = latents / self.vae.config.scaling_factor
+
+ image = self.vae.decode(latents, return_dict=False)[0]
+
+ # cast back to fp16 if needed
+ if needs_upcasting:
+ self.vae.to(dtype=torch.float16)
+ else:
+ image = latents
+ return StableDiffusionXLPipelineOutput(images=image)
+
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py
new file mode 100644
index 0000000000000000000000000000000000000000..fb58b222112a2fd46c280f6100ecf40282426919
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_inpaint_sd_xl.py
@@ -0,0 +1,1910 @@
+# Copyright 2025 Harutatsu Akiyama, Jinbin Bai, and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+from transformers import (
+ CLIPImageProcessor,
+ CLIPTextModel,
+ CLIPTextModelWithProjection,
+ CLIPTokenizer,
+ CLIPVisionModelWithProjection,
+)
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import (
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ TextualInversionLoaderMixin,
+)
+from ...models import (
+ AutoencoderKL,
+ ControlNetUnionModel,
+ ImageProjection,
+ MultiControlNetUnionModel,
+ UNet2DConditionModel,
+)
+from ...models.attention_processor import (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+)
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ is_invisible_watermark_available,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+if is_invisible_watermark_available():
+ from diffusers.pipelines.stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+from ...utils import is_torch_xla_available
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ from diffusers import StableDiffusionXLControlNetUnionInpaintPipeline, ControlNetUnionModel, AutoencoderKL
+ from diffusers.utils import load_image
+ import torch
+ import numpy as np
+ from PIL import Image
+
+ prompt = "A cat"
+ # download an image
+ image = load_image(
+ "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/in_paint/overture-creations-5sI6fQgYIuo.png"
+ ).resize((1024, 1024))
+ mask = load_image(
+ "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/in_paint/overture-creations-5sI6fQgYIuo_mask.png"
+ ).resize((1024, 1024))
+ # initialize the models and pipeline
+ controlnet = ControlNetUnionModel.from_pretrained(
+ "brad-twinkl/controlnet-union-sdxl-1.0-promax", torch_dtype=torch.float16
+ )
+ vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+ pipe = StableDiffusionXLControlNetUnionInpaintPipeline.from_pretrained(
+ "stabilityai/stable-diffusion-xl-base-1.0",
+ controlnet=controlnet,
+ vae=vae,
+ torch_dtype=torch.float16,
+ variant="fp16",
+ )
+ pipe.enable_model_cpu_offload()
+ controlnet_img = image.copy()
+ controlnet_img_np = np.array(controlnet_img)
+ mask_np = np.array(mask)
+ controlnet_img_np[mask_np > 0] = 0
+ controlnet_img = Image.fromarray(controlnet_img_np)
+ # generate image
+ image = pipe(prompt, image=image, mask_image=mask, control_image=[controlnet_img], control_mode=[7]).images[0]
+ image.save("inpaint.png")
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
+def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
+ r"""
+ Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+ Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+ Flawed](https://huggingface.co/papers/2305.08891).
+
+ Args:
+ noise_cfg (`torch.Tensor`):
+ The predicted noise tensor for the guided diffusion process.
+ noise_pred_text (`torch.Tensor`):
+ The predicted noise tensor for the text-guided diffusion process.
+ guidance_rescale (`float`, *optional*, defaults to 0.0):
+ A rescale factor applied to the noise predictions.
+
+ Returns:
+ noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
+ """
+ std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
+ std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
+ # rescale the results from guidance (fixes overexposure)
+ noise_pred_rescaled = noise_cfg * (std_text / std_cfg)
+ # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images
+ noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg
+ return noise_cfg
+
+
+class StableDiffusionXLControlNetUnionInpaintPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ TextualInversionLoaderMixin,
+):
+ r"""
+ Pipeline for text-to-image generation using Stable Diffusion XL.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+ text_encoder ([`CLIPTextModel`]):
+ Frozen text-encoder. Stable Diffusion XL uses the text portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+ the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+ text_encoder_2 ([` CLIPTextModelWithProjection`]):
+ Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
+ specifically the
+ [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
+ variant.
+ tokenizer (`CLIPTokenizer`):
+ Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ tokenizer_2 (`CLIPTokenizer`):
+ Second Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ """
+
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->unet->vae"
+
+ _optional_components = [
+ "tokenizer",
+ "tokenizer_2",
+ "text_encoder",
+ "text_encoder_2",
+ "image_encoder",
+ "feature_extractor",
+ ]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "add_text_embeds",
+ "add_time_ids",
+ "mask",
+ "masked_image_latents",
+ "control_image",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[
+ ControlNetUnionModel, List[ControlNetUnionModel], Tuple[ControlNetUnionModel], MultiControlNetUnionModel
+ ],
+ scheduler: KarrasDiffusionSchedulers,
+ requires_aesthetics_score: bool = False,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None,
+ feature_extractor: Optional[CLIPImageProcessor] = None,
+ image_encoder: Optional[CLIPVisionModelWithProjection] = None,
+ ):
+ super().__init__()
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetUnionModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+ self.register_to_config(requires_aesthetics_score=requires_aesthetics_score)
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+ self.mask_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True
+ )
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+
+ add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+ if add_watermarker:
+ self.watermark = StableDiffusionXLWatermarker()
+ else:
+ self.watermark = None
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder_2, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+ # textual inversion: process multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+ pooled_prompt_embeds = prompt_embeds[0]
+
+ if clip_skip is None:
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+ else:
+ # "2" because SDXL always indexes from the penultimate layer.
+ prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ # normalize str to list
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ negative_prompt_2 = (
+ batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
+ )
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ if self.text_encoder_2 is not None:
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ if self.text_encoder_2 is not None:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ image,
+ mask_image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ output_type,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ pooled_prompt_embeds=None,
+ negative_pooled_prompt_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ control_mode=None,
+ callback_on_step_end_tensor_inputs=None,
+ padding_mask_crop=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+ if num_inference_steps is None:
+ raise ValueError("`num_inference_steps` cannot be None.")
+ elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0:
+ raise ValueError(
+ f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type"
+ f" {type(num_inference_steps)}."
+ )
+
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if padding_mask_crop is not None:
+ if not isinstance(image, PIL.Image.Image):
+ raise ValueError(
+ f"The image should be a PIL image when inpainting mask crop, but is of type {type(image)}."
+ )
+ if not isinstance(mask_image, PIL.Image.Image):
+ raise ValueError(
+ f"The mask image should be a PIL image when inpainting mask crop, but is of type"
+ f" {type(mask_image)}."
+ )
+ if output_type != "pil":
+ raise ValueError(f"The output type should be PIL when inpainting mask crop, but is {output_type}.")
+
+ if prompt_embeds is not None and pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+ )
+
+ if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetUnionModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+ elif not all(isinstance(i, list) for i in image):
+ raise ValueError("For multiple controlnets: elements of `image` must be list of conditionings.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for images_ in image:
+ for image_ in images_:
+ self.check_image(image_, prompt, prompt_embeds)
+
+ if not isinstance(control_guidance_start, (tuple, list)):
+ control_guidance_start = [control_guidance_start]
+
+ if not isinstance(control_guidance_end, (tuple, list)):
+ control_guidance_end = [control_guidance_end]
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ # Check `control_mode`
+ if isinstance(controlnet, ControlNetUnionModel):
+ if max(control_mode) >= controlnet.config.num_control_type:
+ raise ValueError(f"control_mode: must be lower than {controlnet.config.num_control_type}.")
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ for _control_mode, _controlnet in zip(control_mode, self.controlnet.nets):
+ if max(_control_mode) >= _controlnet.config.num_control_type:
+ raise ValueError(f"control_mode: must be lower than {_controlnet.config.num_control_type}.")
+
+ # Equal number of `image` and `control_mode` elements
+ if isinstance(controlnet, ControlNetUnionModel):
+ if len(image) != len(control_mode):
+ raise ValueError("Expected len(control_image) == len(control_mode)")
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ if not all(isinstance(i, list) for i in control_mode):
+ raise ValueError(
+ "For multiple controlnets: elements of control_mode must be lists representing conditioning mode."
+ )
+
+ elif sum(len(x) for x in image) != sum(len(x) for x in control_mode):
+ raise ValueError("Expected len(control_image) == len(control_mode)")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_inpaint_sd_xl.StableDiffusionXLControlNetInpaintPipeline.prepare_control_image
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ crops_coords,
+ resize_mode,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(
+ image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
+ ).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_inpaint_sd_xl.StableDiffusionXLControlNetInpaintPipeline.prepare_latents
+ def prepare_latents(
+ self,
+ batch_size,
+ num_channels_latents,
+ height,
+ width,
+ dtype,
+ device,
+ generator,
+ latents=None,
+ image=None,
+ timestep=None,
+ is_strength_max=True,
+ add_noise=True,
+ return_noise=False,
+ return_image_latents=False,
+ ):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if (image is None or timestep is None) and not is_strength_max:
+ raise ValueError(
+ "Since strength < 1. initial latents are to be initialised as a combination of Image + Noise."
+ "However, either the image or the noise timestep has not been provided."
+ )
+
+ if return_image_latents or (latents is None and not is_strength_max):
+ image = image.to(device=device, dtype=dtype)
+
+ if image.shape[1] == 4:
+ image_latents = image
+ else:
+ image_latents = self._encode_vae_image(image=image, generator=generator)
+ image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
+
+ if latents is None and add_noise:
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ # if strength is 1. then initialise the latents to noise, else initial to image + noise
+ latents = noise if is_strength_max else self.scheduler.add_noise(image_latents, noise, timestep)
+ # if pure noise then scale the initial latents by the Scheduler's init sigma
+ latents = latents * self.scheduler.init_noise_sigma if is_strength_max else latents
+ elif add_noise:
+ noise = latents.to(device)
+ latents = noise * self.scheduler.init_noise_sigma
+ else:
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ latents = image_latents.to(device)
+
+ outputs = (latents,)
+
+ if return_noise:
+ outputs += (noise,)
+
+ if return_image_latents:
+ outputs += (image_latents,)
+
+ return outputs
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_inpaint_sd_xl.StableDiffusionXLControlNetInpaintPipeline._encode_vae_image
+ def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
+ dtype = image.dtype
+ if self.vae.config.force_upcast:
+ image = image.float()
+ self.vae.to(dtype=torch.float32)
+
+ if isinstance(generator, list):
+ image_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(image.shape[0])
+ ]
+ image_latents = torch.cat(image_latents, dim=0)
+ else:
+ image_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ if self.vae.config.force_upcast:
+ self.vae.to(dtype)
+
+ image_latents = image_latents.to(dtype)
+ image_latents = self.vae.config.scaling_factor * image_latents
+
+ return image_latents
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_inpaint_sd_xl.StableDiffusionXLControlNetInpaintPipeline.prepare_mask_latents
+ def prepare_mask_latents(
+ self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance
+ ):
+ # resize the mask to latents shape as we concatenate the mask to the latents
+ # we do that before converting to dtype to avoid breaking in case we're using cpu_offload
+ # and half precision
+ mask = torch.nn.functional.interpolate(
+ mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor)
+ )
+ mask = mask.to(device=device, dtype=dtype)
+
+ # duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
+ if mask.shape[0] < batch_size:
+ if not batch_size % mask.shape[0] == 0:
+ raise ValueError(
+ "The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
+ f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
+ " of masks that you pass is divisible by the total requested batch size."
+ )
+ mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
+
+ mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
+
+ masked_image_latents = None
+ if masked_image is not None:
+ masked_image = masked_image.to(device=device, dtype=dtype)
+ masked_image_latents = self._encode_vae_image(masked_image, generator=generator)
+ if masked_image_latents.shape[0] < batch_size:
+ if not batch_size % masked_image_latents.shape[0] == 0:
+ raise ValueError(
+ "The passed images and the required batch size don't match. Images are supposed to be duplicated"
+ f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
+ " Make sure the number of images that you pass is divisible by the total requested batch size."
+ )
+ masked_image_latents = masked_image_latents.repeat(
+ batch_size // masked_image_latents.shape[0], 1, 1, 1
+ )
+
+ masked_image_latents = (
+ torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
+ )
+
+ # aligning device to prevent device errors when concating it with the latent model input
+ masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
+
+ return mask, masked_image_latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=None):
+ # get the original timestep using init_timestep
+ if denoising_start is None:
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+ t_start = max(num_inference_steps - init_timestep, 0)
+
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ else:
+ # Strength is irrelevant if we directly request a timestep to start at;
+ # that is, strength is determined by the denoising_start instead.
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (denoising_start * self.scheduler.config.num_train_timesteps)
+ )
+ )
+
+ num_inference_steps = (self.scheduler.timesteps < discrete_timestep_cutoff).sum().item()
+ if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
+ # if the scheduler is a 2nd order scheduler we might have to do +1
+ # because `num_inference_steps` might be even given that every timestep
+ # (except the highest one) is duplicated. If `num_inference_steps` is even it would
+ # mean that we cut the timesteps in the middle of the denoising step
+ # (between 1st and 2nd derivative) which leads to incorrect results. By adding 1
+ # we ensure that the denoising process always ends after the 2nd derivate step of the scheduler
+ num_inference_steps = num_inference_steps + 1
+
+ # because t_n+1 >= t_n, we slice the timesteps starting from the end
+ t_start = len(self.scheduler.timesteps) - num_inference_steps
+ timesteps = self.scheduler.timesteps[t_start:]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start)
+ return timesteps, num_inference_steps
+
+ def _get_add_time_ids(
+ self,
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ dtype,
+ text_encoder_projection_dim=None,
+ ):
+ if self.config.requires_aesthetics_score:
+ add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,))
+ add_neg_time_ids = list(original_size + crops_coords_top_left + (negative_aesthetic_score,))
+ else:
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+ add_neg_time_ids = list(original_size + crops_coords_top_left + target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if (
+ expected_add_embed_dim > passed_add_embed_dim
+ and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model."
+ )
+ elif (
+ expected_add_embed_dim < passed_add_embed_dim
+ and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model."
+ )
+ elif expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype)
+
+ return add_time_ids, add_neg_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ image: PipelineImageInput = None,
+ mask_image: PipelineImageInput = None,
+ control_image: Union[PipelineImageInput, List[PipelineImageInput]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ padding_mask_crop: Optional[int] = None,
+ strength: float = 0.9999,
+ num_inference_steps: int = 50,
+ denoising_start: Optional[float] = None,
+ denoising_end: Optional[float] = None,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ control_mode: Optional[Union[int, List[int], List[List[int]]]] = None,
+ guidance_rescale: float = 0.0,
+ original_size: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ aesthetic_score: float = 6.0,
+ negative_aesthetic_score: float = 2.5,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ image (`PIL.Image.Image`):
+ `Image`, or tensor representing an image batch which will be inpainted, *i.e.* parts of the image will
+ be masked out with `mask_image` and repainted according to `prompt`.
+ mask_image (`PIL.Image.Image`):
+ `Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be
+ repainted, while black pixels will be preserved. If `mask_image` is a PIL image, it will be converted
+ to a single channel (luminance) before use. If it's a tensor, it should contain one color channel (L)
+ instead of 3, so the expected shape would be `(B, H, W, 1)`.
+ control_image (`PipelineImageInput` or `List[PipelineImageInput]`, *optional*):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The height in pixels of the generated image.
+ width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
+ The width in pixels of the generated image.
+ padding_mask_crop (`int`, *optional*, defaults to `None`):
+ The size of margin in the crop to be applied to the image and masking. If `None`, no crop is applied to
+ image and mask_image. If `padding_mask_crop` is not `None`, it will first find a rectangular region
+ with the same aspect ration of the image and contains all masked area, and then expand that area based
+ on `padding_mask_crop`. The image and mask_image will then be cropped based on the expanded area before
+ resizing to the original image size for inpainting. This is useful when the masked area is small while
+ the image is large and contain information irrelevant for inpainting, such as background.
+ strength (`float`, *optional*, defaults to 0.9999):
+ Conceptually, indicates how much to transform the masked portion of the reference `image`. Must be
+ between 0 and 1. `image` will be used as a starting point, adding more noise to it the larger the
+ `strength`. The number of denoising steps depends on the amount of noise initially added. When
+ `strength` is 1, added noise will be maximum and the denoising process will run for the full number of
+ iterations specified in `num_inference_steps`. A value of 1, therefore, essentially ignores the masked
+ portion of the reference `image`. Note that in the case of `denoising_start` being declared as an
+ integer, the value of `strength` will be ignored.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ denoising_start (`float`, *optional*):
+ When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be
+ bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and
+ it is assumed that the passed `image` is a partly denoised image. Note that when this is specified,
+ strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline
+ is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
+ denoising_end (`float`, *optional*):
+ When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
+ completed before it is intentionally prematurely terminated. As a result, the returned sample will
+ still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be
+ denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the
+ final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline
+ forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+ applies to [`schedulers.DDIMScheduler`], will be ignored for others.
+ generator (`torch.Generator`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ The ControlNet encoder tries to recognize the content of the input image even if you remove all
+ prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ control_mode (`int` or `List[int]` or `List[List[int]], *optional*):
+ The control condition types for the ControlNet. See the ControlNet's model card forinformation on the
+ available control modes. If multiple ControlNets are specified in `init`, control_mode should be a list
+ where each ControlNet should have its corresponding control mode list. Should reflect the order of
+ conditions in control_image.
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(width, height)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ aesthetic_score (`float`, *optional*, defaults to 6.0):
+ Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_aesthetic_score (`float`, *optional*, defaults to 2.5):
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
+ simulate an aesthetic score of the generated image by influencing the negative text condition.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] or `tuple`:
+ [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple. `tuple. When returning a tuple, the first element is a list with the generated images.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ if not isinstance(control_image, list):
+ control_image = [control_image]
+ else:
+ control_image = control_image.copy()
+
+ if not isinstance(control_mode, list):
+ control_mode = [control_mode]
+
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ control_image = [[item] for item in control_image]
+ control_mode = [[item] for item in control_mode]
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ if isinstance(controlnet_conditioning_scale, float):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * mult
+
+ # 1. Check inputs
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ control_image,
+ mask_image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ output_type,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ control_mode,
+ callback_on_step_end_tensor_inputs,
+ padding_mask_crop,
+ )
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_type = torch.zeros(controlnet.config.num_control_type).scatter_(0, torch.tensor(control_mode), 1)
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_type = [
+ torch.zeros(controlnet_.config.num_control_type).scatter_(0, torch.tensor(control_mode_), 1)
+ for control_mode_, controlnet_ in zip(control_mode, self.controlnet.nets)
+ ]
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # 3. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ prompt_2=prompt_2,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ negative_prompt_2=negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+
+ # 3.1 Encode ip_adapter_image
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. set timesteps
+ def denoising_value_valid(dnv):
+ return isinstance(dnv, float) and 0 < dnv < 1
+
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps, num_inference_steps = self.get_timesteps(
+ num_inference_steps,
+ strength,
+ device,
+ denoising_start=denoising_start if denoising_value_valid(denoising_start) else None,
+ )
+ # check that number of inference steps is not < 1 - as this doesn't make sense
+ if num_inference_steps < 1:
+ raise ValueError(
+ f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
+ f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
+ )
+ # at which timestep to set the initial noise (n.b. 50% if strength is 0.5)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+ # create a boolean to check if the strength is set to 1. if so then initialise the latents with pure noise
+ is_strength_max = strength == 1.0
+ self._num_timesteps = len(timesteps)
+
+ # 5. Preprocess mask and image - resizes image and mask w.r.t height and width
+ # 5.1 Prepare init image
+ if padding_mask_crop is not None:
+ height, width = self.image_processor.get_default_height_width(image, height, width)
+ crops_coords = self.mask_processor.get_crop_region(mask_image, width, height, pad=padding_mask_crop)
+ resize_mode = "fill"
+ else:
+ crops_coords = None
+ resize_mode = "default"
+
+ original_image = image
+ init_image = self.image_processor.preprocess(
+ image, height=height, width=width, crops_coords=crops_coords, resize_mode=resize_mode
+ )
+ init_image = init_image.to(dtype=torch.float32)
+
+ # 5.2 Prepare control images
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_images = []
+
+ for image_ in control_image:
+ image_ = self.prepare_control_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ crops_coords=crops_coords,
+ resize_mode=resize_mode,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(image_)
+
+ control_image = control_images
+ height, width = control_image[0].shape[-2:]
+
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ images = []
+
+ for image_ in control_image_:
+ image_ = self.prepare_control_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ crops_coords=crops_coords,
+ resize_mode=resize_mode,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ images.append(image_)
+ control_images.append(images)
+
+ control_image = control_images
+ height, width = control_image[0][0].shape[-2:]
+
+ # 5.3 Prepare mask
+ mask = self.mask_processor.preprocess(
+ mask_image, height=height, width=width, resize_mode=resize_mode, crops_coords=crops_coords
+ )
+
+ masked_image = init_image * (mask < 0.5)
+ _, _, height, width = init_image.shape
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.vae.config.latent_channels
+ num_channels_unet = self.unet.config.in_channels
+ return_image_latents = num_channels_unet == 4
+
+ add_noise = True if denoising_start is None else False
+ latents_outputs = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ image=init_image,
+ timestep=latent_timestep,
+ is_strength_max=is_strength_max,
+ add_noise=add_noise,
+ return_noise=True,
+ return_image_latents=return_image_latents,
+ )
+
+ if return_image_latents:
+ latents, noise, image_latents = latents_outputs
+ else:
+ latents, noise = latents_outputs
+
+ # 7. Prepare mask latent variables
+ mask, _ = self.prepare_mask_latents(
+ mask,
+ masked_image,
+ batch_size * num_images_per_prompt,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ self.do_classifier_free_guidance,
+ )
+
+ # 8. Check that sizes of mask, masked image and latents match
+ if num_channels_unet != 4:
+ raise ValueError(
+ f"The unet {self.unet.__class__} should have either 4 or 9 input channels, not {self.unet.config.in_channels}."
+ )
+ # 8.1 Prepare extra step kwargs.
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 8.2 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps)
+
+ # 9. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ height, width = latents.shape[-2:]
+ height = height * self.vae_scale_factor
+ width = width * self.vae_scale_factor
+
+ original_size = original_size or (height, width)
+ target_size = target_size or (height, width)
+ for _image in control_image:
+ if isinstance(_image, torch.Tensor):
+ original_size = original_size or _image.shape[-2:]
+
+ # 10. Prepare added time ids & embeddings
+ add_text_embeds = pooled_prompt_embeds
+ if self.text_encoder_2 is None:
+ text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+ else:
+ text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+ add_time_ids, add_neg_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+ add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_neg_time_ids = add_neg_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+ add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device)
+
+ # 11. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+ if (
+ denoising_end is not None
+ and denoising_start is not None
+ and denoising_value_valid(denoising_end)
+ and denoising_value_valid(denoising_start)
+ and denoising_start >= denoising_end
+ ):
+ raise ValueError(
+ f"`denoising_start`: {denoising_start} cannot be larger than or equal to `denoising_end`: "
+ + f" {denoising_end} when using type float."
+ )
+ elif denoising_end is not None and denoising_value_valid(denoising_end):
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (denoising_end * self.scheduler.config.num_train_timesteps)
+ )
+ )
+ num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
+ timesteps = timesteps[:num_inference_steps]
+
+ control_type_repeat_factor = (
+ batch_size * num_images_per_prompt * (2 if self.do_classifier_free_guidance else 1)
+ )
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_type = (
+ control_type.reshape(1, -1)
+ .to(self._execution_device, dtype=prompt_embeds.dtype)
+ .repeat(control_type_repeat_factor, 1)
+ )
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_type = [
+ _control_type.reshape(1, -1)
+ .to(self._execution_device, dtype=prompt_embeds.dtype)
+ .repeat(control_type_repeat_factor, 1)
+ for _control_type in control_type
+ ]
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ added_cond_kwargs = {
+ "text_embeds": add_text_embeds,
+ "time_ids": add_time_ids,
+ }
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ controlnet_added_cond_kwargs = {
+ "text_embeds": add_text_embeds.chunk(2)[1],
+ "time_ids": add_time_ids.chunk(2)[1],
+ }
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+ controlnet_added_cond_kwargs = added_cond_kwargs
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ # # Resize control_image to match the size of the input to the controlnet
+ # if control_image.shape[-2:] != control_model_input.shape[-2:]:
+ # control_image = F.interpolate(control_image, size=control_model_input.shape[-2:], mode="bilinear", align_corners=False)
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ control_type=control_type,
+ control_type_idx=control_mode,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ added_cond_kwargs=controlnet_added_cond_kwargs,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ added_cond_kwargs["image_embeds"] = image_embeds
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ if self.do_classifier_free_guidance and guidance_rescale > 0.0:
+ # Based on 3.4. in https://huggingface.co/papers/2305.08891
+ noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ init_latents_proper = image_latents
+ if self.do_classifier_free_guidance:
+ init_mask, _ = mask.chunk(2)
+ else:
+ init_mask = mask
+
+ if i < len(timesteps) - 1:
+ noise_timestep = timesteps[i + 1]
+ init_latents_proper = self.scheduler.add_noise(
+ init_latents_proper, noise, torch.tensor([noise_timestep])
+ )
+
+ latents = (1 - init_mask) * init_latents_proper + init_mask * latents
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ control_image = callback_outputs.pop("control_image", control_image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+ else:
+ return StableDiffusionXLPipelineOutput(images=latents)
+
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ if padding_mask_crop is not None:
+ image = [self.image_processor.apply_overlay(mask_image, original_image, i, crops_coords) for i in image]
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py
new file mode 100644
index 0000000000000000000000000000000000000000..5961d389effb68b880efc58de2eba9b525d9b127
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl.py
@@ -0,0 +1,1620 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+from transformers import (
+ CLIPImageProcessor,
+ CLIPTextModel,
+ CLIPTextModelWithProjection,
+ CLIPTokenizer,
+ CLIPVisionModelWithProjection,
+)
+
+from diffusers.utils.import_utils import is_invisible_watermark_available
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import (
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ TextualInversionLoaderMixin,
+)
+from ...models import (
+ AutoencoderKL,
+ ControlNetUnionModel,
+ ImageProjection,
+ MultiControlNetUnionModel,
+ UNet2DConditionModel,
+)
+from ...models.attention_processor import (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+)
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import is_compiled_module, is_torch_version, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+if is_invisible_watermark_available():
+ from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+from ...utils import is_torch_xla_available
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> # !pip install controlnet_aux
+ >>> from controlnet_aux import LineartAnimeDetector
+ >>> from diffusers import StableDiffusionXLControlNetUnionPipeline, ControlNetUnionModel, AutoencoderKL
+ >>> from diffusers.utils import load_image
+ >>> import torch
+
+ >>> prompt = "A cat"
+ >>> # download an image
+ >>> image = load_image(
+ ... "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png"
+ ... ).resize((1024, 1024))
+ >>> # initialize the models and pipeline
+ >>> controlnet = ControlNetUnionModel.from_pretrained(
+ ... "xinsir/controlnet-union-sdxl-1.0", torch_dtype=torch.float16
+ ... )
+ >>> vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+ >>> pipe = StableDiffusionXLControlNetUnionPipeline.from_pretrained(
+ ... "stabilityai/stable-diffusion-xl-base-1.0",
+ ... controlnet=controlnet,
+ ... vae=vae,
+ ... torch_dtype=torch.float16,
+ ... variant="fp16",
+ ... )
+ >>> pipe.enable_model_cpu_offload()
+ >>> # prepare image
+ >>> processor = LineartAnimeDetector.from_pretrained("lllyasviel/Annotators")
+ >>> controlnet_img = processor(image, output_type="pil")
+ >>> # generate image
+ >>> image = pipe(prompt, control_image=[controlnet_img], control_mode=[3], height=1024, width=1024).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps
+def retrieve_timesteps(
+ scheduler,
+ num_inference_steps: Optional[int] = None,
+ device: Optional[Union[str, torch.device]] = None,
+ timesteps: Optional[List[int]] = None,
+ sigmas: Optional[List[float]] = None,
+ **kwargs,
+):
+ r"""
+ Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+ custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+ Args:
+ scheduler (`SchedulerMixin`):
+ The scheduler to get timesteps from.
+ num_inference_steps (`int`):
+ The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+ must be `None`.
+ device (`str` or `torch.device`, *optional*):
+ The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
+ `num_inference_steps` and `sigmas` must be `None`.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
+ `num_inference_steps` and `timesteps` must be `None`.
+
+ Returns:
+ `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+ second element is the number of inference steps.
+ """
+ if timesteps is not None and sigmas is not None:
+ raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values")
+ if timesteps is not None:
+ accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accepts_timesteps:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" timestep schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ elif sigmas is not None:
+ accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+ if not accept_sigmas:
+ raise ValueError(
+ f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+ f" sigmas schedules. Please check whether you are using the correct scheduler."
+ )
+ scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ num_inference_steps = len(timesteps)
+ else:
+ scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+ timesteps = scheduler.timesteps
+ return timesteps, num_inference_steps
+
+
+class StableDiffusionXLControlNetUnionPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ IPAdapterMixin,
+ FromSingleFileMixin,
+):
+ r"""
+ Pipeline for text-to-image generation using Stable Diffusion XL with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+ text_encoder ([`~transformers.CLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ text_encoder_2 ([`~transformers.CLIPTextModelWithProjection`]):
+ Second frozen text-encoder
+ ([laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)).
+ tokenizer ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ tokenizer_2 ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ unet ([`UNet2DConditionModel`]):
+ A `UNet2DConditionModel` to denoise the encoded image latents.
+ controlnet ([`ControlNetUnionModel`]`):
+ Provides additional conditioning to the `unet` during the denoising process.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`):
+ Whether the negative prompt embeddings should always be set to 0. Also see the config of
+ `stabilityai/stable-diffusion-xl-base-1-0`.
+ add_watermarker (`bool`, *optional*):
+ Whether to use the [invisible_watermark](https://github.com/ShieldMnt/invisible-watermark/) library to
+ watermark output images. If not defined, it defaults to `True` if the package is installed; otherwise no
+ watermarker is used.
+ """
+
+ # leave controlnet out on purpose because it iterates with unet
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->unet->vae"
+ _optional_components = [
+ "tokenizer",
+ "tokenizer_2",
+ "text_encoder",
+ "text_encoder_2",
+ "feature_extractor",
+ "image_encoder",
+ ]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "add_text_embeds",
+ "add_time_ids",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[
+ ControlNetUnionModel, List[ControlNetUnionModel], Tuple[ControlNetUnionModel], MultiControlNetUnionModel
+ ],
+ scheduler: KarrasDiffusionSchedulers,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None,
+ feature_extractor: CLIPImageProcessor = None,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ ):
+ super().__init__()
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetUnionModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+ if add_watermarker:
+ self.watermark = StableDiffusionXLWatermarker()
+ else:
+ self.watermark = None
+
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder_2, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+ # textual inversion: process multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+ pooled_prompt_embeds = prompt_embeds[0]
+
+ if clip_skip is None:
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+ else:
+ # "2" because SDXL always indexes from the penultimate layer.
+ prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ # normalize str to list
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ negative_prompt_2 = (
+ batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
+ )
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ if self.text_encoder_2 is not None:
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ if self.text_encoder_2 is not None:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ image: PipelineImageInput,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ pooled_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ negative_pooled_prompt_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ control_mode=None,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if prompt_embeds is not None and pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+ )
+
+ if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetUnionModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+ elif not all(isinstance(i, list) for i in image):
+ raise ValueError("For multiple controlnets: elements of `image` must be list of conditionings.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for images_ in image:
+ for image_ in images_:
+ self.check_image(image_, prompt, prompt_embeds)
+
+ # Check `controlnet_conditioning_scale`
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ if isinstance(controlnet_conditioning_scale, list):
+ if any(isinstance(i, list) for i in controlnet_conditioning_scale):
+ raise ValueError("A single batch of multiple conditionings is not supported at the moment.")
+ elif isinstance(controlnet_conditioning_scale, list) and len(controlnet_conditioning_scale) != len(
+ self.controlnet.nets
+ ):
+ raise ValueError(
+ "For multiple controlnets: When `controlnet_conditioning_scale` is specified as `list`, it must have"
+ " the same length as the number of controlnets"
+ )
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control_guidance_start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control_guidance_start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control_guidance_end: {end} can't be larger than 1.0.")
+
+ # Check `control_mode`
+ if isinstance(controlnet, ControlNetUnionModel):
+ if max(control_mode) >= controlnet.config.num_control_type:
+ raise ValueError(f"control_mode: must be lower than {controlnet.config.num_control_type}.")
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ for _control_mode, _controlnet in zip(control_mode, self.controlnet.nets):
+ if max(_control_mode) >= _controlnet.config.num_control_type:
+ raise ValueError(f"control_mode: must be lower than {_controlnet.config.num_control_type}.")
+
+ # Equal number of `image` and `control_mode` elements
+ if isinstance(controlnet, ControlNetUnionModel):
+ if len(image) != len(control_mode):
+ raise ValueError("Expected len(control_image) == len(control_mode)")
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ if not all(isinstance(i, list) for i in control_mode):
+ raise ValueError(
+ "For multiple controlnets: elements of control_mode must be lists representing conditioning mode."
+ )
+
+ elif sum(len(x) for x in image) != sum(len(x) for x in control_mode):
+ raise ValueError("Expected len(control_image) == len(control_mode)")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet.StableDiffusionControlNetPipeline.prepare_image
+ def prepare_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids
+ def _get_add_time_ids(
+ self, original_size, crops_coords_top_left, target_size, dtype, text_encoder_projection_dim=None
+ ):
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ return add_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
+ def get_guidance_scale_embedding(
+ self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+ ) -> torch.Tensor:
+ """
+ See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+
+ Args:
+ w (`torch.Tensor`):
+ Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
+ embedding_dim (`int`, *optional*, defaults to 512):
+ Dimension of the embeddings to generate.
+ dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
+ Data type of the generated embeddings.
+
+ Returns:
+ `torch.Tensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
+ """
+ assert len(w.shape) == 1
+ w = w * 1000.0
+
+ half_dim = embedding_dim // 2
+ emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
+ emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
+ emb = w.to(dtype)[:, None] * emb[None, :]
+ emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+ if embedding_dim % 2 == 1: # zero pad
+ emb = torch.nn.functional.pad(emb, (0, 1))
+ assert emb.shape == (w.shape[0], embedding_dim)
+ return emb
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def denoising_end(self):
+ return self._denoising_end
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ control_image: Union[PipelineImageInput, List[PipelineImageInput]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: int = 50,
+ timesteps: List[int] = None,
+ sigmas: List[float] = None,
+ denoising_end: Optional[float] = None,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ control_mode: Optional[Union[int, List[int], List[List[int]]]] = None,
+ original_size: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ negative_original_size: Optional[Tuple[int, int]] = None,
+ negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
+ negative_target_size: Optional[Tuple[int, int]] = None,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ ):
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders.
+ control_image (`PipelineImageInput` or `List[PipelineImageInput]`, *optional*):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The height in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+ The width in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ timesteps (`List[int]`, *optional*):
+ Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+ in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+ passed will be used. Must be in descending order.
+ sigmas (`List[float]`, *optional*):
+ Custom sigmas to use for the denoising process with schedulers which support a `sigmas` argument in
+ their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is passed
+ will be used.
+ denoising_end (`float`, *optional*):
+ When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
+ completed before it is intentionally prematurely terminated. As a result, the returned sample will
+ still retain a substantial amount of noise as determined by the discrete timesteps selected by the
+ scheduler. The denoising_end parameter should ideally be utilized when this pipeline forms a part of a
+ "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output)
+ guidance_scale (`float`, *optional*, defaults to 5.0):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+ pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. This is sent to `tokenizer_2`
+ and `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders.
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+ applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor is generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, pooled text embeddings are generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs (prompt
+ weighting). If not provided, pooled `negative_prompt_embeds` are generated from `negative_prompt` input
+ argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+ [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ The ControlNet encoder tries to recognize the content of the input image even if you remove all
+ prompts. A `guidance_scale` value between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ control_mode (`int` or `List[int]` or `List[List[int]], *optional*):
+ The control condition types for the ControlNet. See the ControlNet's model card forinformation on the
+ available control modes. If multiple ControlNets are specified in `init`, control_mode should be a list
+ where each ControlNet should have its corresponding control mode list. Should reflect the order of
+ conditions in control_image.
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a specific image resolution. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a target image resolution. It should be as same
+ as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+ otherwise a `tuple` is returned containing the output images.
+ """
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ if not isinstance(control_image, list):
+ control_image = [control_image]
+ else:
+ control_image = control_image.copy()
+
+ if not isinstance(control_mode, list):
+ control_mode = [control_mode]
+
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ control_image = [[item] for item in control_image]
+ control_mode = [[item] for item in control_mode]
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ if isinstance(controlnet_conditioning_scale, float):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * mult
+
+ # 1. Check inputs
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ control_image,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ negative_pooled_prompt_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ control_mode,
+ callback_on_step_end_tensor_inputs,
+ )
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_type = torch.zeros(controlnet.config.num_control_type).scatter_(0, torch.tensor(control_mode), 1)
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_type = [
+ torch.zeros(controlnet_.config.num_control_type).scatter_(0, torch.tensor(control_mode_), 1)
+ for control_mode_, controlnet_ in zip(control_mode, self.controlnet.nets)
+ ]
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._denoising_end = denoising_end
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetUnionModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3.1 Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt,
+ prompt_2,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+
+ # 3.2 Encode ip_adapter_image
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4. Prepare image
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_images = []
+
+ for image_ in control_image:
+ image_ = self.prepare_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(image_)
+
+ control_image = control_images
+ height, width = control_image[0].shape[-2:]
+
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ images = []
+
+ for image_ in control_image_:
+ image_ = self.prepare_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ images.append(image_)
+ control_images.append(images)
+
+ control_image = control_images
+ height, width = control_image[0][0].shape[-2:]
+
+ # 5. Prepare timesteps
+ timesteps, num_inference_steps = retrieve_timesteps(
+ self.scheduler, num_inference_steps, device, timesteps, sigmas
+ )
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.unet.config.in_channels
+ latents = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 6.5 Optionally get Guidance Scale Embedding
+ timestep_cond = None
+ if self.unet.config.time_cond_proj_dim is not None:
+ guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+ timestep_cond = self.get_guidance_scale_embedding(
+ guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+ ).to(device=device, dtype=latents.dtype)
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps)
+
+ # 7.2 Prepare added time ids & embeddings
+ original_size = original_size or (height, width)
+ target_size = target_size or (height, width)
+ for _image in control_image:
+ if isinstance(_image, torch.Tensor):
+ original_size = original_size or _image.shape[-2:]
+ add_text_embeds = pooled_prompt_embeds
+ if self.text_encoder_2 is None:
+ text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+ else:
+ text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+ add_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+
+ if negative_original_size is not None and negative_target_size is not None:
+ negative_add_time_ids = self._get_add_time_ids(
+ negative_original_size,
+ negative_crops_coords_top_left,
+ negative_target_size,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+ else:
+ negative_add_time_ids = add_time_ids
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_time_ids = torch.cat([negative_add_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+
+ # 8.1 Apply denoising_end
+ if (
+ self.denoising_end is not None
+ and isinstance(self.denoising_end, float)
+ and self.denoising_end > 0
+ and self.denoising_end < 1
+ ):
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (self.denoising_end * self.scheduler.config.num_train_timesteps)
+ )
+ )
+ num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
+ timesteps = timesteps[:num_inference_steps]
+
+ is_unet_compiled = is_compiled_module(self.unet)
+ is_controlnet_compiled = is_compiled_module(self.controlnet)
+ is_torch_higher_equal_2_1 = is_torch_version(">=", "2.1")
+
+ control_type_repeat_factor = (
+ batch_size * num_images_per_prompt * (2 if self.do_classifier_free_guidance else 1)
+ )
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_type = (
+ control_type.reshape(1, -1)
+ .to(self._execution_device, dtype=prompt_embeds.dtype)
+ .repeat(control_type_repeat_factor, 1)
+ )
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_type = [
+ _control_type.reshape(1, -1)
+ .to(self._execution_device, dtype=prompt_embeds.dtype)
+ .repeat(control_type_repeat_factor, 1)
+ for _control_type in control_type
+ ]
+
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # Relevant thread:
+ # https://dev-discuss.pytorch.org/t/cudagraphs-in-pytorch-2-0/1428
+ if (is_unet_compiled and is_controlnet_compiled) and is_torch_higher_equal_2_1:
+ torch._inductor.cudagraph_mark_step_begin()
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ added_cond_kwargs = {
+ "text_embeds": add_text_embeds,
+ "time_ids": add_time_ids,
+ }
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ controlnet_added_cond_kwargs = {
+ "text_embeds": add_text_embeds.chunk(2)[1],
+ "time_ids": add_time_ids.chunk(2)[1],
+ }
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+ controlnet_added_cond_kwargs = added_cond_kwargs
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ control_type=control_type,
+ control_type_idx=control_mode,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ added_cond_kwargs=controlnet_added_cond_kwargs,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ added_cond_kwargs["image_embeds"] = image_embeds
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ timestep_cond=timestep_cond,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
+ add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ if not output_type == "latent":
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+
+ if needs_upcasting:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # unscale/denormalize the latents
+ # denormalize with the mean and std if available and not None
+ has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None
+ has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None
+ if has_latents_mean and has_latents_std:
+ latents_mean = (
+ torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents_std = (
+ torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean
+ else:
+ latents = latents / self.vae.config.scaling_factor
+
+ image = self.vae.decode(latents, return_dict=False)[0]
+
+ # cast back to fp16 if needed
+ if needs_upcasting:
+ self.vae.to(dtype=torch.float16)
+ else:
+ image = latents
+
+ if not output_type == "latent":
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py
new file mode 100644
index 0000000000000000000000000000000000000000..8fedb6d8609ad9c5163b6efcd4fc60f9eb916e39
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_controlnet_union_sd_xl_img2img.py
@@ -0,0 +1,1725 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+from transformers import (
+ CLIPImageProcessor,
+ CLIPTextModel,
+ CLIPTextModelWithProjection,
+ CLIPTokenizer,
+ CLIPVisionModelWithProjection,
+)
+
+from diffusers.utils.import_utils import is_invisible_watermark_available
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...loaders import (
+ FromSingleFileMixin,
+ IPAdapterMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ TextualInversionLoaderMixin,
+)
+from ...models import (
+ AutoencoderKL,
+ ControlNetUnionModel,
+ ImageProjection,
+ MultiControlNetUnionModel,
+ UNet2DConditionModel,
+)
+from ...models.attention_processor import (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+)
+from ...models.lora import adjust_lora_scale_text_encoder
+from ...schedulers import KarrasDiffusionSchedulers
+from ...utils import (
+ USE_PEFT_BACKEND,
+ deprecate,
+ logging,
+ replace_example_docstring,
+ scale_lora_layers,
+ unscale_lora_layers,
+)
+from ...utils.torch_utils import empty_device_cache, is_compiled_module, randn_tensor
+from ..pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from ..stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+if is_invisible_watermark_available():
+ from ..stable_diffusion_xl.watermark import StableDiffusionXLWatermarker
+
+
+from ...utils import is_torch_xla_available
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ # !pip install controlnet_aux
+ from diffusers import (
+ StableDiffusionXLControlNetUnionImg2ImgPipeline,
+ ControlNetUnionModel,
+ AutoencoderKL,
+ )
+ from diffusers.utils import load_image
+ import torch
+ from PIL import Image
+ import numpy as np
+
+ prompt = "A cat"
+ # download an image
+ image = load_image(
+ "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/kandinsky/cat.png"
+ )
+ # initialize the models and pipeline
+ controlnet = ControlNetUnionModel.from_pretrained(
+ "brad-twinkl/controlnet-union-sdxl-1.0-promax", torch_dtype=torch.float16
+ )
+ vae = AutoencoderKL.from_pretrained("madebyollin/sdxl-vae-fp16-fix", torch_dtype=torch.float16)
+ pipe = StableDiffusionXLControlNetUnionImg2ImgPipeline.from_pretrained(
+ "stabilityai/stable-diffusion-xl-base-1.0",
+ controlnet=controlnet,
+ vae=vae,
+ torch_dtype=torch.float16,
+ variant="fp16",
+ ).to("cuda")
+ # `enable_model_cpu_offload` is not recommended due to multiple generations
+ height = image.height
+ width = image.width
+ ratio = np.sqrt(1024.0 * 1024.0 / (width * height))
+ # 3 * 3 upscale correspond to 16 * 3 multiply, 2 * 2 correspond to 16 * 2 multiply and so on.
+ scale_image_factor = 3
+ base_factor = 16
+ factor = scale_image_factor * base_factor
+ W, H = int(width * ratio) // factor * factor, int(height * ratio) // factor * factor
+ image = image.resize((W, H))
+ target_width = W // scale_image_factor
+ target_height = H // scale_image_factor
+ images = []
+ crops_coords_list = [
+ (0, 0),
+ (0, width // 2),
+ (height // 2, 0),
+ (width // 2, height // 2),
+ 0,
+ 0,
+ 0,
+ 0,
+ 0,
+ ]
+ for i in range(scale_image_factor):
+ for j in range(scale_image_factor):
+ left = j * target_width
+ top = i * target_height
+ right = left + target_width
+ bottom = top + target_height
+ cropped_image = image.crop((left, top, right, bottom))
+ cropped_image = cropped_image.resize((W, H))
+ images.append(cropped_image)
+ # set ControlNetUnion input
+ result_images = []
+ for sub_img, crops_coords in zip(images, crops_coords_list):
+ new_width, new_height = W, H
+ out = pipe(
+ prompt=[prompt] * 1,
+ image=sub_img,
+ control_image=[sub_img],
+ control_mode=[6],
+ width=new_width,
+ height=new_height,
+ num_inference_steps=30,
+ crops_coords_top_left=(W, H),
+ target_size=(W, H),
+ original_size=(W * 2, H * 2),
+ )
+ result_images.append(out.images[0])
+ new_im = Image.new("RGB", (new_width * scale_image_factor, new_height * scale_image_factor))
+ new_im.paste(result_images[0], (0, 0))
+ new_im.paste(result_images[1], (new_width, 0))
+ new_im.paste(result_images[2], (new_width * 2, 0))
+ new_im.paste(result_images[3], (0, new_height))
+ new_im.paste(result_images[4], (new_width, new_height))
+ new_im.paste(result_images[5], (new_width * 2, new_height))
+ new_im.paste(result_images[6], (0, new_height * 2))
+ new_im.paste(result_images[7], (new_width, new_height * 2))
+ new_im.paste(result_images[8], (new_width * 2, new_height * 2))
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
+def retrieve_latents(
+ encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
+):
+ if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
+ return encoder_output.latent_dist.sample(generator)
+ elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
+ return encoder_output.latent_dist.mode()
+ elif hasattr(encoder_output, "latents"):
+ return encoder_output.latents
+ else:
+ raise AttributeError("Could not access latents of provided encoder_output")
+
+
+class StableDiffusionXLControlNetUnionImg2ImgPipeline(
+ DiffusionPipeline,
+ StableDiffusionMixin,
+ TextualInversionLoaderMixin,
+ StableDiffusionXLLoraLoaderMixin,
+ FromSingleFileMixin,
+ IPAdapterMixin,
+):
+ r"""
+ Pipeline for image-to-image generation using Stable Diffusion XL with ControlNet guidance.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ The pipeline also inherits the following loading methods:
+ - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+ - [`~loaders.StableDiffusionXLLoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+ - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+ text_encoder ([`CLIPTextModel`]):
+ Frozen text-encoder. Stable Diffusion uses the text portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+ the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+ text_encoder_2 ([` CLIPTextModelWithProjection`]):
+ Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
+ specifically the
+ [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
+ variant.
+ tokenizer (`CLIPTokenizer`):
+ Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ tokenizer_2 (`CLIPTokenizer`):
+ Second Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+ controlnet ([`ControlNetUnionModel`]):
+ Provides additional conditioning to the unet during the denoising process.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ requires_aesthetics_score (`bool`, *optional*, defaults to `"False"`):
+ Whether the `unet` requires an `aesthetic_score` condition to be passed during inference. Also see the
+ config of `stabilityai/stable-diffusion-xl-refiner-1-0`.
+ force_zeros_for_empty_prompt (`bool`, *optional*, defaults to `"True"`):
+ Whether the negative prompt embeddings shall be forced to always be set to 0. Also see the config of
+ `stabilityai/stable-diffusion-xl-base-1-0`.
+ add_watermarker (`bool`, *optional*):
+ Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to
+ watermark output images. If not defined, it will default to True if the package is installed, otherwise no
+ watermarker will be used.
+ feature_extractor ([`~transformers.CLIPImageProcessor`]):
+ A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+ """
+
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->image_encoder->unet->vae"
+ _optional_components = [
+ "tokenizer",
+ "tokenizer_2",
+ "text_encoder",
+ "text_encoder_2",
+ "feature_extractor",
+ "image_encoder",
+ ]
+ _callback_tensor_inputs = ["latents", "prompt_embeds", "add_text_embeds", "add_time_ids", "control_image"]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ controlnet: Union[
+ ControlNetUnionModel, List[ControlNetUnionModel], Tuple[ControlNetUnionModel], MultiControlNetUnionModel
+ ],
+ scheduler: KarrasDiffusionSchedulers,
+ requires_aesthetics_score: bool = False,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None,
+ feature_extractor: CLIPImageProcessor = None,
+ image_encoder: CLIPVisionModelWithProjection = None,
+ ):
+ super().__init__()
+
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = MultiControlNetUnionModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ feature_extractor=feature_extractor,
+ image_encoder=image_encoder,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True)
+ self.control_image_processor = VaeImageProcessor(
+ vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True, do_normalize=False
+ )
+ add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
+
+ if add_watermarker:
+ self.watermark = StableDiffusionXLWatermarker()
+ else:
+ self.watermark = None
+
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+ self.register_to_config(requires_aesthetics_score=requires_aesthetics_score)
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ lora_scale: Optional[float] = None,
+ clip_skip: Optional[int] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, StableDiffusionXLLoraLoaderMixin):
+ self._lora_scale = lora_scale
+
+ # dynamically adjust the LoRA scale
+ if self.text_encoder is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if not USE_PEFT_BACKEND:
+ adjust_lora_scale_text_encoder(self.text_encoder_2, lora_scale)
+ else:
+ scale_lora_layers(self.text_encoder_2, lora_scale)
+
+ prompt = [prompt] if isinstance(prompt, str) else prompt
+
+ if prompt is not None:
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2
+
+ # textual inversion: process multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(text_input_ids.to(device), output_hidden_states=True)
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if pooled_prompt_embeds is None and prompt_embeds[0].ndim == 2:
+ pooled_prompt_embeds = prompt_embeds[0]
+
+ if clip_skip is None:
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+ else:
+ # "2" because SDXL always indexes from the penultimate layer.
+ prompt_embeds = prompt_embeds.hidden_states[-(clip_skip + 2)]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ # normalize str to list
+ negative_prompt = batch_size * [negative_prompt] if isinstance(negative_prompt, str) else negative_prompt
+ negative_prompt_2 = (
+ batch_size * [negative_prompt_2] if isinstance(negative_prompt_2, str) else negative_prompt_2
+ )
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ if negative_pooled_prompt_embeds is None and negative_prompt_embeds[0].ndim == 2:
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ if self.text_encoder_2 is not None:
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ prompt_embeds = prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ if self.text_encoder_2 is not None:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ else:
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=self.unet.dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ if self.text_encoder is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder, lora_scale)
+
+ if self.text_encoder_2 is not None:
+ if isinstance(self, StableDiffusionXLLoraLoaderMixin) and USE_PEFT_BACKEND:
+ # Retrieve the original scale by scaling back the LoRA layers
+ unscale_lora_layers(self.text_encoder_2, lora_scale)
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.encode_image
+ def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+ dtype = next(self.image_encoder.parameters()).dtype
+
+ if not isinstance(image, torch.Tensor):
+ image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+ image = image.to(device=device, dtype=dtype)
+ if output_hidden_states:
+ image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+ image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_enc_hidden_states = self.image_encoder(
+ torch.zeros_like(image), output_hidden_states=True
+ ).hidden_states[-2]
+ uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+ num_images_per_prompt, dim=0
+ )
+ return image_enc_hidden_states, uncond_image_enc_hidden_states
+ else:
+ image_embeds = self.image_encoder(image).image_embeds
+ image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+ uncond_image_embeds = torch.zeros_like(image_embeds)
+
+ return image_embeds, uncond_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_ip_adapter_image_embeds
+ def prepare_ip_adapter_image_embeds(
+ self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+ ):
+ image_embeds = []
+ if do_classifier_free_guidance:
+ negative_image_embeds = []
+ if ip_adapter_image_embeds is None:
+ if not isinstance(ip_adapter_image, list):
+ ip_adapter_image = [ip_adapter_image]
+
+ if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+ raise ValueError(
+ f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+ )
+
+ for single_ip_adapter_image, image_proj_layer in zip(
+ ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+ ):
+ output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+ single_image_embeds, single_negative_image_embeds = self.encode_image(
+ single_ip_adapter_image, device, 1, output_hidden_state
+ )
+
+ image_embeds.append(single_image_embeds[None, :])
+ if do_classifier_free_guidance:
+ negative_image_embeds.append(single_negative_image_embeds[None, :])
+ else:
+ for single_image_embeds in ip_adapter_image_embeds:
+ if do_classifier_free_guidance:
+ single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+ negative_image_embeds.append(single_negative_image_embeds)
+ image_embeds.append(single_image_embeds)
+
+ ip_adapter_image_embeds = []
+ for i, single_image_embeds in enumerate(image_embeds):
+ single_image_embeds = torch.cat([single_image_embeds] * num_images_per_prompt, dim=0)
+ if do_classifier_free_guidance:
+ single_negative_image_embeds = torch.cat([negative_image_embeds[i]] * num_images_per_prompt, dim=0)
+ single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds], dim=0)
+
+ single_image_embeds = single_image_embeds.to(device=device)
+ ip_adapter_image_embeds.append(single_image_embeds)
+
+ return ip_adapter_image_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ pooled_prompt_embeds=None,
+ negative_pooled_prompt_embeds=None,
+ ip_adapter_image=None,
+ ip_adapter_image_embeds=None,
+ controlnet_conditioning_scale=1.0,
+ control_guidance_start=0.0,
+ control_guidance_end=1.0,
+ control_mode=None,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+ if num_inference_steps is None:
+ raise ValueError("`num_inference_steps` cannot be None.")
+ elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0:
+ raise ValueError(
+ f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type"
+ f" {type(num_inference_steps)}."
+ )
+
+ if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+
+ if prompt_embeds is not None and pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`."
+ )
+
+ if negative_prompt_embeds is not None and negative_pooled_prompt_embeds is None:
+ raise ValueError(
+ "If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
+ )
+
+ # `prompt` needs more sophisticated handling when there are multiple
+ # conditionings.
+ if isinstance(self.controlnet, MultiControlNetUnionModel):
+ if isinstance(prompt, list):
+ logger.warning(
+ f"You have {len(self.controlnet.nets)} ControlNets and you have passed {len(prompt)}"
+ " prompts. The conditionings will be fixed across the prompts."
+ )
+
+ # Check `image`
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ for image_ in image:
+ self.check_image(image_, prompt, prompt_embeds)
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ if not isinstance(image, list):
+ raise TypeError("For multiple controlnets: `image` must be type `list`")
+ elif not all(isinstance(i, list) for i in image):
+ raise ValueError("For multiple controlnets: elements of `image` must be list of conditionings.")
+ elif len(image) != len(self.controlnet.nets):
+ raise ValueError(
+ f"For multiple controlnets: `image` must have the same length as the number of controlnets, but got {len(image)} images and {len(self.controlnet.nets)} ControlNets."
+ )
+
+ for images_ in image:
+ for image_ in images_:
+ self.check_image(image_, prompt, prompt_embeds)
+
+ if not isinstance(control_guidance_start, (tuple, list)):
+ control_guidance_start = [control_guidance_start]
+
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ if len(control_guidance_start) != len(self.controlnet.nets):
+ raise ValueError(
+ f"`control_guidance_start`: {control_guidance_start} has {len(control_guidance_start)} elements but there are {len(self.controlnet.nets)} controlnets available. Make sure to provide {len(self.controlnet.nets)}."
+ )
+
+ if not isinstance(control_guidance_end, (tuple, list)):
+ control_guidance_end = [control_guidance_end]
+
+ if len(control_guidance_start) != len(control_guidance_end):
+ raise ValueError(
+ f"`control_guidance_start` has {len(control_guidance_start)} elements, but `control_guidance_end` has {len(control_guidance_end)} elements. Make sure to provide the same number of elements to each list."
+ )
+
+ for start, end in zip(control_guidance_start, control_guidance_end):
+ if start >= end:
+ raise ValueError(
+ f"control guidance start: {start} cannot be larger or equal to control guidance end: {end}."
+ )
+ if start < 0.0:
+ raise ValueError(f"control guidance start: {start} can't be smaller than 0.")
+ if end > 1.0:
+ raise ValueError(f"control guidance end: {end} can't be larger than 1.0.")
+
+ # Check `control_mode`
+ if isinstance(controlnet, ControlNetUnionModel):
+ if max(control_mode) >= controlnet.config.num_control_type:
+ raise ValueError(f"control_mode: must be lower than {controlnet.config.num_control_type}.")
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ for _control_mode, _controlnet in zip(control_mode, self.controlnet.nets):
+ if max(_control_mode) >= _controlnet.config.num_control_type:
+ raise ValueError(f"control_mode: must be lower than {_controlnet.config.num_control_type}.")
+
+ if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+ raise ValueError(
+ "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+ )
+
+ if ip_adapter_image_embeds is not None:
+ if not isinstance(ip_adapter_image_embeds, list):
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+ )
+ elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+ raise ValueError(
+ f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.check_image
+ def check_image(self, image, prompt, prompt_embeds):
+ image_is_pil = isinstance(image, PIL.Image.Image)
+ image_is_tensor = isinstance(image, torch.Tensor)
+ image_is_np = isinstance(image, np.ndarray)
+ image_is_pil_list = isinstance(image, list) and isinstance(image[0], PIL.Image.Image)
+ image_is_tensor_list = isinstance(image, list) and isinstance(image[0], torch.Tensor)
+ image_is_np_list = isinstance(image, list) and isinstance(image[0], np.ndarray)
+
+ if (
+ not image_is_pil
+ and not image_is_tensor
+ and not image_is_np
+ and not image_is_pil_list
+ and not image_is_tensor_list
+ and not image_is_np_list
+ ):
+ raise TypeError(
+ f"image must be passed and be one of PIL image, numpy array, torch tensor, list of PIL images, list of numpy arrays or list of torch tensors, but is {type(image)}"
+ )
+
+ if image_is_pil:
+ image_batch_size = 1
+ else:
+ image_batch_size = len(image)
+
+ if prompt is not None and isinstance(prompt, str):
+ prompt_batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ prompt_batch_size = len(prompt)
+ elif prompt_embeds is not None:
+ prompt_batch_size = prompt_embeds.shape[0]
+
+ if image_batch_size != 1 and image_batch_size != prompt_batch_size:
+ raise ValueError(
+ f"If image batch size is not 1, image batch size must be same as prompt batch size. image batch size: {image_batch_size}, prompt batch size: {prompt_batch_size}"
+ )
+
+ # Copied from diffusers.pipelines.controlnet.pipeline_controlnet_sd_xl.StableDiffusionXLControlNetPipeline.prepare_image
+ def prepare_control_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ image = self.control_image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.StableDiffusionImg2ImgPipeline.get_timesteps
+ def get_timesteps(self, num_inference_steps, strength, device):
+ # get the original timestep using init_timestep
+ init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
+
+ t_start = max(num_inference_steps - init_timestep, 0)
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
+
+ return timesteps, num_inference_steps - t_start
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline.prepare_latents
+ def prepare_latents(
+ self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True
+ ):
+ if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
+ raise ValueError(
+ f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+ )
+
+ latents_mean = latents_std = None
+ if hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None:
+ latents_mean = torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1)
+ if hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None:
+ latents_std = torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1)
+
+ # Offload text encoder if `enable_model_cpu_offload` was enabled
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.text_encoder_2.to("cpu")
+ empty_device_cache()
+
+ image = image.to(device=device, dtype=dtype)
+
+ batch_size = batch_size * num_images_per_prompt
+
+ if image.shape[1] == 4:
+ init_latents = image
+
+ else:
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ if self.vae.config.force_upcast:
+ image = image.float()
+ self.vae.to(dtype=torch.float32)
+
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ elif isinstance(generator, list):
+ if image.shape[0] < batch_size and batch_size % image.shape[0] == 0:
+ image = torch.cat([image] * (batch_size // image.shape[0]), dim=0)
+ elif image.shape[0] < batch_size and batch_size % image.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {image.shape[0]} to effective batch_size {batch_size} "
+ )
+
+ init_latents = [
+ retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
+ for i in range(batch_size)
+ ]
+ init_latents = torch.cat(init_latents, dim=0)
+ else:
+ init_latents = retrieve_latents(self.vae.encode(image), generator=generator)
+
+ if self.vae.config.force_upcast:
+ self.vae.to(dtype)
+
+ init_latents = init_latents.to(dtype)
+ if latents_mean is not None and latents_std is not None:
+ latents_mean = latents_mean.to(device=device, dtype=dtype)
+ latents_std = latents_std.to(device=device, dtype=dtype)
+ init_latents = (init_latents - latents_mean) * self.vae.config.scaling_factor / latents_std
+ else:
+ init_latents = self.vae.config.scaling_factor * init_latents
+
+ if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
+ # expand init_latents for batch_size
+ additional_image_per_prompt = batch_size // init_latents.shape[0]
+ init_latents = torch.cat([init_latents] * additional_image_per_prompt, dim=0)
+ elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
+ )
+ else:
+ init_latents = torch.cat([init_latents], dim=0)
+
+ if add_noise:
+ shape = init_latents.shape
+ noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ # get latents
+ init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
+
+ latents = init_latents
+
+ return latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl_img2img.StableDiffusionXLImg2ImgPipeline._get_add_time_ids
+ def _get_add_time_ids(
+ self,
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ negative_original_size,
+ negative_crops_coords_top_left,
+ negative_target_size,
+ dtype,
+ text_encoder_projection_dim=None,
+ ):
+ if self.config.requires_aesthetics_score:
+ add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,))
+ add_neg_time_ids = list(
+ negative_original_size + negative_crops_coords_top_left + (negative_aesthetic_score,)
+ )
+ else:
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+ add_neg_time_ids = list(negative_original_size + crops_coords_top_left + negative_target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + text_encoder_projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if (
+ expected_add_embed_dim > passed_add_embed_dim
+ and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model."
+ )
+ elif (
+ expected_add_embed_dim < passed_add_embed_dim
+ and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model."
+ )
+ elif expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype)
+
+ return add_time_ids, add_neg_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def clip_skip(self):
+ return self._clip_skip
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def cross_attention_kwargs(self):
+ return self._cross_attention_kwargs
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ image: PipelineImageInput = None,
+ control_image: Union[PipelineImageInput, List[PipelineImageInput]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ strength: float = 0.8,
+ num_inference_steps: int = 50,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: float = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_pooled_prompt_embeds: Optional[torch.Tensor] = None,
+ ip_adapter_image: Optional[PipelineImageInput] = None,
+ ip_adapter_image_embeds: Optional[List[torch.Tensor]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 0.8,
+ guess_mode: bool = False,
+ control_guidance_start: Union[float, List[float]] = 0.0,
+ control_guidance_end: Union[float, List[float]] = 1.0,
+ control_mode: Optional[Union[int, List[int], List[List[int]]]] = None,
+ original_size: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ negative_original_size: Optional[Tuple[int, int]] = None,
+ negative_crops_coords_top_left: Tuple[int, int] = (0, 0),
+ negative_target_size: Optional[Tuple[int, int]] = None,
+ aesthetic_score: float = 6.0,
+ negative_aesthetic_score: float = 2.5,
+ clip_skip: Optional[int] = None,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ **kwargs,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The initial image will be used as the starting point for the image generation process. Can also accept
+ image latents as `image`, if passing latents directly, it will not be encoded again.
+ control_image (`PipelineImageInput` or `List[PipelineImageInput]`, *optional*):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ height (`int`, *optional*, defaults to the size of control_image):
+ The height in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ width (`int`, *optional*, defaults to the size of control_image):
+ The width in pixels of the generated image. Anything below 512 pixels won't work well for
+ [stabilityai/stable-diffusion-xl-base-1.0](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
+ and checkpoints that are not specifically fine-tuned on low resolutions.
+ strength (`float`, *optional*, defaults to 0.8):
+ Indicates extent to transform the reference `image`. Must be between 0 and 1. `image` is used as a
+ starting point and more noise is added the higher the `strength`. The number of denoising steps depends
+ on the amount of noise initially added. When `strength` is 1, added noise is maximum and the denoising
+ process runs for the full number of iterations specified in `num_inference_steps`. A value of 1
+ essentially ignores `image`.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ Guidance scale as defined in [Classifier-Free Diffusion
+ Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
+ of [Imagen Paper](https://huggingface.co/papers/2205.11487). Guidance scale is enabled by setting
+ `guidance_scale > 1`. Higher guidance scale encourages to generate images that are closely linked to
+ the text `prompt`, usually at the expense of lower image quality.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) in the DDIM paper: https://huggingface.co/papers/2010.02502. Only
+ applies to [`schedulers.DDIMScheduler`], will be ignored for others.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html)
+ to make generation deterministic.
+ latents (`torch.Tensor`, *optional*):
+ Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ tensor will be generated by sampling using the supplied random `generator`.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+ ip_adapter_image_embeds (`List[torch.Tensor]`, *optional*):
+ Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+ IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+ contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+ provided, embeddings are computed from the `ip_adapter_image` input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ guess_mode (`bool`, *optional*, defaults to `False`):
+ In this mode, the ControlNet encoder will try best to recognize the content of the input image even if
+ you remove all prompts. The `guidance_scale` between 3.0 and 5.0 is recommended.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ control_mode (`int` or `List[int]` or `List[List[int]], *optional*):
+ The control condition types for the ControlNet. See the ControlNet's model card forinformation on the
+ available control modes. If multiple ControlNets are specified in `init`, control_mode should be a list
+ where each ControlNet should have its corresponding control mode list. Should reflect the order of
+ conditions in control_image
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(height, width)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(height, width)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a specific image resolution. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ To negatively condition the generation process based on a specific crop coordinates. Part of SDXL's
+ micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ negative_target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ To negatively condition the generation process based on a target image resolution. It should be as same
+ as the `target_size` for most cases. Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). For more
+ information, refer to this issue thread: https://github.com/huggingface/diffusers/issues/4208.
+ aesthetic_score (`float`, *optional*, defaults to 6.0):
+ Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_aesthetic_score (`float`, *optional*, defaults to 2.5):
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
+ simulate an aesthetic score of the generated image by influencing the negative text condition.
+ clip_skip (`int`, *optional*):
+ Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+ the output of the pre-final layer will be used for computing the prompt embeddings.
+ callback_on_step_end (`Callable`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A function or a subclass of `PipelineCallback` or `MultiPipelineCallbacks` that is called at the end of
+ each denoising step during the inference. with the following arguments: `callback_on_step_end(self:
+ DiffusionPipeline, step: int, timestep: int, callback_kwargs: Dict)`. `callback_kwargs` will include a
+ list of all tensors as specified by `callback_on_step_end_tensor_inputs`.
+ callback_on_step_end_tensor_inputs (`List`, *optional*):
+ The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+ will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+ `._callback_tensor_inputs` attribute of your pipeline class.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] if `return_dict` is True, otherwise a `tuple`
+ containing the output images.
+ """
+
+ callback = kwargs.pop("callback", None)
+ callback_steps = kwargs.pop("callback_steps", None)
+
+ if callback is not None:
+ deprecate(
+ "callback",
+ "1.0.0",
+ "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+ if callback_steps is not None:
+ deprecate(
+ "callback_steps",
+ "1.0.0",
+ "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+ )
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ controlnet = self.controlnet._orig_mod if is_compiled_module(self.controlnet) else self.controlnet
+
+ if not isinstance(control_image, list):
+ control_image = [control_image]
+ else:
+ control_image = control_image.copy()
+
+ if not isinstance(control_mode, list):
+ control_mode = [control_mode]
+
+ if isinstance(controlnet, MultiControlNetUnionModel):
+ control_image = [[item] for item in control_image]
+ control_mode = [[item] for item in control_mode]
+
+ # align format for control guidance
+ if not isinstance(control_guidance_start, list) and isinstance(control_guidance_end, list):
+ control_guidance_start = len(control_guidance_end) * [control_guidance_start]
+ elif not isinstance(control_guidance_end, list) and isinstance(control_guidance_start, list):
+ control_guidance_end = len(control_guidance_start) * [control_guidance_end]
+ elif not isinstance(control_guidance_start, list) and not isinstance(control_guidance_end, list):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+ control_guidance_start, control_guidance_end = (
+ mult * [control_guidance_start],
+ mult * [control_guidance_end],
+ )
+
+ if isinstance(controlnet_conditioning_scale, float):
+ mult = len(controlnet.nets) if isinstance(controlnet, MultiControlNetUnionModel) else len(control_mode)
+ controlnet_conditioning_scale = [controlnet_conditioning_scale] * mult
+
+ # 1. Check inputs
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ control_image,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ controlnet_conditioning_scale,
+ control_guidance_start,
+ control_guidance_end,
+ control_mode,
+ callback_on_step_end_tensor_inputs,
+ )
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_type = torch.zeros(controlnet.config.num_control_type).scatter_(0, torch.tensor(control_mode), 1)
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_type = [
+ torch.zeros(controlnet_.config.num_control_type).scatter_(0, torch.tensor(control_mode_), 1)
+ for control_mode_, controlnet_ in zip(control_mode, self.controlnet.nets)
+ ]
+
+ self._guidance_scale = guidance_scale
+ self._clip_skip = clip_skip
+ self._cross_attention_kwargs = cross_attention_kwargs
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ global_pool_conditions = (
+ controlnet.config.global_pool_conditions
+ if isinstance(controlnet, ControlNetUnionModel)
+ else controlnet.nets[0].config.global_pool_conditions
+ )
+ guess_mode = guess_mode or global_pool_conditions
+
+ # 3.1. Encode input prompt
+ text_encoder_lora_scale = (
+ self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+ )
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt,
+ prompt_2,
+ device,
+ num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ clip_skip=self.clip_skip,
+ )
+
+ # 3.2 Encode ip_adapter_image
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ image_embeds = self.prepare_ip_adapter_image_embeds(
+ ip_adapter_image,
+ ip_adapter_image_embeds,
+ device,
+ batch_size * num_images_per_prompt,
+ self.do_classifier_free_guidance,
+ )
+
+ # 4.1 Prepare image
+ image = self.image_processor.preprocess(image, height=height, width=width).to(dtype=torch.float32)
+
+ # 4.2 Prepare control images
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_images = []
+
+ for image_ in control_image:
+ image_ = self.prepare_control_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ control_images.append(image_)
+
+ control_image = control_images
+ height, width = control_image[0].shape[-2:]
+
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ images = []
+
+ for image_ in control_image_:
+ image_ = self.prepare_control_image(
+ image=image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=controlnet.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=guess_mode,
+ )
+
+ images.append(image_)
+ control_images.append(images)
+
+ control_image = control_images
+ height, width = control_image[0][0].shape[-2:]
+
+ # 5. Prepare timesteps
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+ self._num_timesteps = len(timesteps)
+
+ # 6. Prepare latent variables
+ if latents is None:
+ latents = self.prepare_latents(
+ image,
+ latent_timestep,
+ batch_size,
+ num_images_per_prompt,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ True,
+ )
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 7.1 Create tensor stating which controlnets to keep
+ controlnet_keep = []
+ for i in range(len(timesteps)):
+ keeps = [
+ 1.0 - float(i / len(timesteps) < s or (i + 1) / len(timesteps) > e)
+ for s, e in zip(control_guidance_start, control_guidance_end)
+ ]
+ controlnet_keep.append(keeps)
+
+ # 7.2 Prepare added time ids & embeddings
+ original_size = original_size or (height, width)
+ target_size = target_size or (height, width)
+ for _image in control_image:
+ if isinstance(_image, torch.Tensor):
+ original_size = original_size or _image.shape[-2:]
+
+ if negative_original_size is None:
+ negative_original_size = original_size
+ if negative_target_size is None:
+ negative_target_size = target_size
+ add_text_embeds = pooled_prompt_embeds
+
+ if self.text_encoder_2 is None:
+ text_encoder_projection_dim = int(pooled_prompt_embeds.shape[-1])
+ else:
+ text_encoder_projection_dim = self.text_encoder_2.config.projection_dim
+
+ add_time_ids, add_neg_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ negative_original_size,
+ negative_crops_coords_top_left,
+ negative_target_size,
+ dtype=prompt_embeds.dtype,
+ text_encoder_projection_dim=text_encoder_projection_dim,
+ )
+ add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_neg_time_ids = add_neg_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+ add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device)
+
+ control_type_repeat_factor = (
+ batch_size * num_images_per_prompt * (2 if self.do_classifier_free_guidance else 1)
+ )
+
+ if isinstance(controlnet, ControlNetUnionModel):
+ control_type = (
+ control_type.reshape(1, -1)
+ .to(self._execution_device, dtype=prompt_embeds.dtype)
+ .repeat(control_type_repeat_factor, 1)
+ )
+ elif isinstance(controlnet, MultiControlNetUnionModel):
+ control_type = [
+ _control_type.reshape(1, -1)
+ .to(self._execution_device, dtype=prompt_embeds.dtype)
+ .repeat(control_type_repeat_factor, 1)
+ for _control_type in control_type
+ ]
+
+ # 8. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ added_cond_kwargs = {
+ "text_embeds": add_text_embeds,
+ "time_ids": add_time_ids,
+ }
+
+ # controlnet(s) inference
+ if guess_mode and self.do_classifier_free_guidance:
+ # Infer ControlNet only for the conditional batch.
+ control_model_input = latents
+ control_model_input = self.scheduler.scale_model_input(control_model_input, t)
+ controlnet_prompt_embeds = prompt_embeds.chunk(2)[1]
+ controlnet_added_cond_kwargs = {
+ "text_embeds": add_text_embeds.chunk(2)[1],
+ "time_ids": add_time_ids.chunk(2)[1],
+ }
+ else:
+ control_model_input = latent_model_input
+ controlnet_prompt_embeds = prompt_embeds
+ controlnet_added_cond_kwargs = added_cond_kwargs
+
+ if isinstance(controlnet_keep[i], list):
+ cond_scale = [c * s for c, s in zip(controlnet_conditioning_scale, controlnet_keep[i])]
+ else:
+ controlnet_cond_scale = controlnet_conditioning_scale
+ if isinstance(controlnet_cond_scale, list):
+ controlnet_cond_scale = controlnet_cond_scale[0]
+ cond_scale = controlnet_cond_scale * controlnet_keep[i]
+ down_block_res_samples, mid_block_res_sample = self.controlnet(
+ control_model_input,
+ t,
+ encoder_hidden_states=controlnet_prompt_embeds,
+ controlnet_cond=control_image,
+ control_type=control_type,
+ control_type_idx=control_mode,
+ conditioning_scale=cond_scale,
+ guess_mode=guess_mode,
+ added_cond_kwargs=controlnet_added_cond_kwargs,
+ return_dict=False,
+ )
+
+ if guess_mode and self.do_classifier_free_guidance:
+ # Inferred ControlNet only for the conditional batch.
+ # To apply the output of ControlNet to both the unconditional and conditional batches,
+ # add 0 to the unconditional batch to keep it unchanged.
+ down_block_res_samples = [torch.cat([torch.zeros_like(d), d]) for d in down_block_res_samples]
+ mid_block_res_sample = torch.cat([torch.zeros_like(mid_block_res_sample), mid_block_res_sample])
+
+ if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+ added_cond_kwargs["image_embeds"] = image_embeds
+
+ # predict the noise residual
+ noise_pred = self.unet(
+ latent_model_input,
+ t,
+ encoder_hidden_states=prompt_embeds,
+ cross_attention_kwargs=self.cross_attention_kwargs,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ added_cond_kwargs=added_cond_kwargs,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ add_text_embeds = callback_outputs.pop("add_text_embeds", add_text_embeds)
+ add_time_ids = callback_outputs.pop("add_time_ids", add_time_ids)
+ control_image = callback_outputs.pop("control_image", control_image)
+
+ # call the callback, if provided
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+ if callback is not None and i % callback_steps == 0:
+ step_idx = i // getattr(self.scheduler, "order", 1)
+ callback(step_idx, t, latents)
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ # If we do sequential model offloading, let's offload unet and controlnet
+ # manually for max memory savings
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.unet.to("cpu")
+ self.controlnet.to("cpu")
+ empty_device_cache()
+
+ if not output_type == "latent":
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+
+ if needs_upcasting:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # unscale/denormalize the latents
+ # denormalize with the mean and std if available and not None
+ has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None
+ has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None
+ if has_latents_mean and has_latents_std:
+ latents_mean = (
+ torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents_std = (
+ torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype)
+ )
+ latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean
+ else:
+ latents = latents / self.vae.config.scaling_factor
+
+ image = self.vae.decode(latents, return_dict=False)[0]
+
+ # cast back to fp16 if needed
+ if needs_upcasting:
+ self.vae.to(dtype=torch.float16)
+ else:
+ image = latents
+ return StableDiffusionXLPipelineOutput(images=image)
+
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_flax_controlnet.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_flax_controlnet.py
new file mode 100644
index 0000000000000000000000000000000000000000..1de1d4bde7f56e89317405ac96600e47f034659d
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet/pipeline_flax_controlnet.py
@@ -0,0 +1,535 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import warnings
+from functools import partial
+from typing import Dict, List, Optional, Union
+
+import jax
+import jax.numpy as jnp
+import numpy as np
+from flax.core.frozen_dict import FrozenDict
+from flax.jax_utils import unreplicate
+from flax.training.common_utils import shard
+from PIL import Image
+from transformers import CLIPImageProcessor, CLIPTokenizer, FlaxCLIPTextModel
+
+from ...models import FlaxAutoencoderKL, FlaxControlNetModel, FlaxUNet2DConditionModel
+from ...schedulers import (
+ FlaxDDIMScheduler,
+ FlaxDPMSolverMultistepScheduler,
+ FlaxLMSDiscreteScheduler,
+ FlaxPNDMScheduler,
+)
+from ...utils import PIL_INTERPOLATION, logging, replace_example_docstring
+from ..pipeline_flax_utils import FlaxDiffusionPipeline
+from ..stable_diffusion import FlaxStableDiffusionPipelineOutput
+from ..stable_diffusion.safety_checker_flax import FlaxStableDiffusionSafetyChecker
+
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+# Set to True to use python for loop instead of jax.fori_loop for easier debugging
+DEBUG = False
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> import jax
+ >>> import numpy as np
+ >>> import jax.numpy as jnp
+ >>> from flax.jax_utils import replicate
+ >>> from flax.training.common_utils import shard
+ >>> from diffusers.utils import load_image, make_image_grid
+ >>> from PIL import Image
+ >>> from diffusers import FlaxStableDiffusionControlNetPipeline, FlaxControlNetModel
+
+
+ >>> def create_key(seed=0):
+ ... return jax.random.PRNGKey(seed)
+
+
+ >>> rng = create_key(0)
+
+ >>> # get canny image
+ >>> canny_image = load_image(
+ ... "https://huggingface.co/datasets/YiYiXu/test-doc-assets/resolve/main/blog_post_cell_10_output_0.jpeg"
+ ... )
+
+ >>> prompts = "best quality, extremely detailed"
+ >>> negative_prompts = "monochrome, lowres, bad anatomy, worst quality, low quality"
+
+ >>> # load control net and stable diffusion v1-5
+ >>> controlnet, controlnet_params = FlaxControlNetModel.from_pretrained(
+ ... "lllyasviel/sd-controlnet-canny", from_pt=True, dtype=jnp.float32
+ ... )
+ >>> pipe, params = FlaxStableDiffusionControlNetPipeline.from_pretrained(
+ ... "stable-diffusion-v1-5/stable-diffusion-v1-5",
+ ... controlnet=controlnet,
+ ... revision="flax",
+ ... dtype=jnp.float32,
+ ... )
+ >>> params["controlnet"] = controlnet_params
+
+ >>> num_samples = jax.device_count()
+ >>> rng = jax.random.split(rng, jax.device_count())
+
+ >>> prompt_ids = pipe.prepare_text_inputs([prompts] * num_samples)
+ >>> negative_prompt_ids = pipe.prepare_text_inputs([negative_prompts] * num_samples)
+ >>> processed_image = pipe.prepare_image_inputs([canny_image] * num_samples)
+
+ >>> p_params = replicate(params)
+ >>> prompt_ids = shard(prompt_ids)
+ >>> negative_prompt_ids = shard(negative_prompt_ids)
+ >>> processed_image = shard(processed_image)
+
+ >>> output = pipe(
+ ... prompt_ids=prompt_ids,
+ ... image=processed_image,
+ ... params=p_params,
+ ... prng_seed=rng,
+ ... num_inference_steps=50,
+ ... neg_prompt_ids=negative_prompt_ids,
+ ... jit=True,
+ ... ).images
+
+ >>> output_images = pipe.numpy_to_pil(np.asarray(output.reshape((num_samples,) + output.shape[-3:])))
+ >>> output_images = make_image_grid(output_images, num_samples // 4, 4)
+ >>> output_images.save("generated_image.png")
+ ```
+"""
+
+
+class FlaxStableDiffusionControlNetPipeline(FlaxDiffusionPipeline):
+ r"""
+ Flax-based pipeline for text-to-image generation using Stable Diffusion with ControlNet Guidance.
+
+ This model inherits from [`FlaxDiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ Args:
+ vae ([`FlaxAutoencoderKL`]):
+ Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+ text_encoder ([`~transformers.FlaxCLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ tokenizer ([`~transformers.CLIPTokenizer`]):
+ A `CLIPTokenizer` to tokenize text.
+ unet ([`FlaxUNet2DConditionModel`]):
+ A `FlaxUNet2DConditionModel` to denoise the encoded image latents.
+ controlnet ([`FlaxControlNetModel`]:
+ Provides additional conditioning to the `unet` during the denoising process.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`FlaxDDIMScheduler`], [`FlaxLMSDiscreteScheduler`], [`FlaxPNDMScheduler`], or
+ [`FlaxDPMSolverMultistepScheduler`].
+ safety_checker ([`FlaxStableDiffusionSafetyChecker`]):
+ Classification module that estimates whether generated images could be considered offensive or harmful.
+ Please refer to the [model card](https://huggingface.co/stable-diffusion-v1-5/stable-diffusion-v1-5) for
+ more details about a model's potential harms.
+ feature_extractor ([`~transformers.CLIPImageProcessor`]):
+ A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+ """
+
+ def __init__(
+ self,
+ vae: FlaxAutoencoderKL,
+ text_encoder: FlaxCLIPTextModel,
+ tokenizer: CLIPTokenizer,
+ unet: FlaxUNet2DConditionModel,
+ controlnet: FlaxControlNetModel,
+ scheduler: Union[
+ FlaxDDIMScheduler, FlaxPNDMScheduler, FlaxLMSDiscreteScheduler, FlaxDPMSolverMultistepScheduler
+ ],
+ safety_checker: FlaxStableDiffusionSafetyChecker,
+ feature_extractor: CLIPImageProcessor,
+ dtype: jnp.dtype = jnp.float32,
+ ):
+ super().__init__()
+ self.dtype = dtype
+
+ if safety_checker is None:
+ logger.warning(
+ f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+ " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+ " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+ " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+ " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+ " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+ )
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ unet=unet,
+ controlnet=controlnet,
+ scheduler=scheduler,
+ safety_checker=safety_checker,
+ feature_extractor=feature_extractor,
+ )
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+
+ def prepare_text_inputs(self, prompt: Union[str, List[str]]):
+ if not isinstance(prompt, (str, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ text_input = self.tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=self.tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="np",
+ )
+
+ return text_input.input_ids
+
+ def prepare_image_inputs(self, image: Union[Image.Image, List[Image.Image]]):
+ if not isinstance(image, (Image.Image, list)):
+ raise ValueError(f"image has to be of type `PIL.Image.Image` or list but is {type(image)}")
+
+ if isinstance(image, Image.Image):
+ image = [image]
+
+ processed_images = jnp.concatenate([preprocess(img, jnp.float32) for img in image])
+
+ return processed_images
+
+ def _get_has_nsfw_concepts(self, features, params):
+ has_nsfw_concepts = self.safety_checker(features, params)
+ return has_nsfw_concepts
+
+ def _run_safety_checker(self, images, safety_model_params, jit=False):
+ # safety_model_params should already be replicated when jit is True
+ pil_images = [Image.fromarray(image) for image in images]
+ features = self.feature_extractor(pil_images, return_tensors="np").pixel_values
+
+ if jit:
+ features = shard(features)
+ has_nsfw_concepts = _p_get_has_nsfw_concepts(self, features, safety_model_params)
+ has_nsfw_concepts = unshard(has_nsfw_concepts)
+ safety_model_params = unreplicate(safety_model_params)
+ else:
+ has_nsfw_concepts = self._get_has_nsfw_concepts(features, safety_model_params)
+
+ images_was_copied = False
+ for idx, has_nsfw_concept in enumerate(has_nsfw_concepts):
+ if has_nsfw_concept:
+ if not images_was_copied:
+ images_was_copied = True
+ images = images.copy()
+
+ images[idx] = np.zeros(images[idx].shape, dtype=np.uint8) # black image
+
+ if any(has_nsfw_concepts):
+ warnings.warn(
+ "Potential NSFW content was detected in one or more images. A black image will be returned"
+ " instead. Try again with a different prompt and/or seed."
+ )
+
+ return images, has_nsfw_concepts
+
+ def _generate(
+ self,
+ prompt_ids: jnp.ndarray,
+ image: jnp.ndarray,
+ params: Union[Dict, FrozenDict],
+ prng_seed: jax.Array,
+ num_inference_steps: int,
+ guidance_scale: float,
+ latents: Optional[jnp.ndarray] = None,
+ neg_prompt_ids: Optional[jnp.ndarray] = None,
+ controlnet_conditioning_scale: float = 1.0,
+ ):
+ height, width = image.shape[-2:]
+ if height % 64 != 0 or width % 64 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 64 but are {height} and {width}.")
+
+ # get prompt text embeddings
+ prompt_embeds = self.text_encoder(prompt_ids, params=params["text_encoder"])[0]
+
+ # TODO: currently it is assumed `do_classifier_free_guidance = guidance_scale > 1.0`
+ # implement this conditional `do_classifier_free_guidance = guidance_scale > 1.0`
+ batch_size = prompt_ids.shape[0]
+
+ max_length = prompt_ids.shape[-1]
+
+ if neg_prompt_ids is None:
+ uncond_input = self.tokenizer(
+ [""] * batch_size, padding="max_length", max_length=max_length, return_tensors="np"
+ ).input_ids
+ else:
+ uncond_input = neg_prompt_ids
+ negative_prompt_embeds = self.text_encoder(uncond_input, params=params["text_encoder"])[0]
+ context = jnp.concatenate([negative_prompt_embeds, prompt_embeds])
+
+ image = jnp.concatenate([image] * 2)
+
+ latents_shape = (
+ batch_size,
+ self.unet.config.in_channels,
+ height // self.vae_scale_factor,
+ width // self.vae_scale_factor,
+ )
+ if latents is None:
+ latents = jax.random.normal(prng_seed, shape=latents_shape, dtype=jnp.float32)
+ else:
+ if latents.shape != latents_shape:
+ raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {latents_shape}")
+
+ def loop_body(step, args):
+ latents, scheduler_state = args
+ # For classifier free guidance, we need to do two forward passes.
+ # Here we concatenate the unconditional and text embeddings into a single batch
+ # to avoid doing two forward passes
+ latents_input = jnp.concatenate([latents] * 2)
+
+ t = jnp.array(scheduler_state.timesteps, dtype=jnp.int32)[step]
+ timestep = jnp.broadcast_to(t, latents_input.shape[0])
+
+ latents_input = self.scheduler.scale_model_input(scheduler_state, latents_input, t)
+
+ down_block_res_samples, mid_block_res_sample = self.controlnet.apply(
+ {"params": params["controlnet"]},
+ jnp.array(latents_input),
+ jnp.array(timestep, dtype=jnp.int32),
+ encoder_hidden_states=context,
+ controlnet_cond=image,
+ conditioning_scale=controlnet_conditioning_scale,
+ return_dict=False,
+ )
+
+ # predict the noise residual
+ noise_pred = self.unet.apply(
+ {"params": params["unet"]},
+ jnp.array(latents_input),
+ jnp.array(timestep, dtype=jnp.int32),
+ encoder_hidden_states=context,
+ down_block_additional_residuals=down_block_res_samples,
+ mid_block_additional_residual=mid_block_res_sample,
+ ).sample
+
+ # perform guidance
+ noise_pred_uncond, noise_prediction_text = jnp.split(noise_pred, 2, axis=0)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_prediction_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents, scheduler_state = self.scheduler.step(scheduler_state, noise_pred, t, latents).to_tuple()
+ return latents, scheduler_state
+
+ scheduler_state = self.scheduler.set_timesteps(
+ params["scheduler"], num_inference_steps=num_inference_steps, shape=latents_shape
+ )
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * params["scheduler"].init_noise_sigma
+
+ if DEBUG:
+ # run with python for loop
+ for i in range(num_inference_steps):
+ latents, scheduler_state = loop_body(i, (latents, scheduler_state))
+ else:
+ latents, _ = jax.lax.fori_loop(0, num_inference_steps, loop_body, (latents, scheduler_state))
+
+ # scale and decode the image latents with vae
+ latents = 1 / self.vae.config.scaling_factor * latents
+ image = self.vae.apply({"params": params["vae"]}, latents, method=self.vae.decode).sample
+
+ image = (image / 2 + 0.5).clip(0, 1).transpose(0, 2, 3, 1)
+ return image
+
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt_ids: jnp.ndarray,
+ image: jnp.ndarray,
+ params: Union[Dict, FrozenDict],
+ prng_seed: jax.Array,
+ num_inference_steps: int = 50,
+ guidance_scale: Union[float, jnp.ndarray] = 7.5,
+ latents: jnp.ndarray = None,
+ neg_prompt_ids: jnp.ndarray = None,
+ controlnet_conditioning_scale: Union[float, jnp.ndarray] = 1.0,
+ return_dict: bool = True,
+ jit: bool = False,
+ ):
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ prompt_ids (`jnp.ndarray`):
+ The prompt or prompts to guide the image generation.
+ image (`jnp.ndarray`):
+ Array representing the ControlNet input condition to provide guidance to the `unet` for generation.
+ params (`Dict` or `FrozenDict`):
+ Dictionary containing the model parameters/weights.
+ prng_seed (`jax.Array`):
+ Array containing random number generator key.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ latents (`jnp.ndarray`, *optional*):
+ Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+ generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+ array is generated by sampling using the supplied random `generator`.
+ controlnet_conditioning_scale (`float` or `jnp.ndarray`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] instead of
+ a plain tuple.
+ jit (`bool`, defaults to `False`):
+ Whether to run `pmap` versions of the generation and safety scoring functions.
+
+
+
+ This argument exists because `__call__` is not yet end-to-end pmap-able. It will be removed in a
+ future release.
+
+
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.FlaxStableDiffusionPipelineOutput`] is
+ returned, otherwise a `tuple` is returned where the first element is a list with the generated images
+ and the second element is a list of `bool`s indicating whether the corresponding generated image
+ contains "not-safe-for-work" (nsfw) content.
+ """
+
+ height, width = image.shape[-2:]
+
+ if isinstance(guidance_scale, float):
+ # Convert to a tensor so each device gets a copy. Follow the prompt_ids for
+ # shape information, as they may be sharded (when `jit` is `True`), or not.
+ guidance_scale = jnp.array([guidance_scale] * prompt_ids.shape[0])
+ if len(prompt_ids.shape) > 2:
+ # Assume sharded
+ guidance_scale = guidance_scale[:, None]
+
+ if isinstance(controlnet_conditioning_scale, float):
+ # Convert to a tensor so each device gets a copy. Follow the prompt_ids for
+ # shape information, as they may be sharded (when `jit` is `True`), or not.
+ controlnet_conditioning_scale = jnp.array([controlnet_conditioning_scale] * prompt_ids.shape[0])
+ if len(prompt_ids.shape) > 2:
+ # Assume sharded
+ controlnet_conditioning_scale = controlnet_conditioning_scale[:, None]
+
+ if jit:
+ images = _p_generate(
+ self,
+ prompt_ids,
+ image,
+ params,
+ prng_seed,
+ num_inference_steps,
+ guidance_scale,
+ latents,
+ neg_prompt_ids,
+ controlnet_conditioning_scale,
+ )
+ else:
+ images = self._generate(
+ prompt_ids,
+ image,
+ params,
+ prng_seed,
+ num_inference_steps,
+ guidance_scale,
+ latents,
+ neg_prompt_ids,
+ controlnet_conditioning_scale,
+ )
+
+ if self.safety_checker is not None:
+ safety_params = params["safety_checker"]
+ images_uint8_casted = (images * 255).round().astype("uint8")
+ num_devices, batch_size = images.shape[:2]
+
+ images_uint8_casted = np.asarray(images_uint8_casted).reshape(num_devices * batch_size, height, width, 3)
+ images_uint8_casted, has_nsfw_concept = self._run_safety_checker(images_uint8_casted, safety_params, jit)
+ images = np.array(images)
+
+ # block images
+ if any(has_nsfw_concept):
+ for i, is_nsfw in enumerate(has_nsfw_concept):
+ if is_nsfw:
+ images[i] = np.asarray(images_uint8_casted[i])
+
+ images = images.reshape(num_devices, batch_size, height, width, 3)
+ else:
+ images = np.asarray(images)
+ has_nsfw_concept = False
+
+ if not return_dict:
+ return (images, has_nsfw_concept)
+
+ return FlaxStableDiffusionPipelineOutput(images=images, nsfw_content_detected=has_nsfw_concept)
+
+
+# Static argnums are pipe, num_inference_steps. A change would trigger recompilation.
+# Non-static args are (sharded) input tensors mapped over their first dimension (hence, `0`).
+@partial(
+ jax.pmap,
+ in_axes=(None, 0, 0, 0, 0, None, 0, 0, 0, 0),
+ static_broadcasted_argnums=(0, 5),
+)
+def _p_generate(
+ pipe,
+ prompt_ids,
+ image,
+ params,
+ prng_seed,
+ num_inference_steps,
+ guidance_scale,
+ latents,
+ neg_prompt_ids,
+ controlnet_conditioning_scale,
+):
+ return pipe._generate(
+ prompt_ids,
+ image,
+ params,
+ prng_seed,
+ num_inference_steps,
+ guidance_scale,
+ latents,
+ neg_prompt_ids,
+ controlnet_conditioning_scale,
+ )
+
+
+@partial(jax.pmap, static_broadcasted_argnums=(0,))
+def _p_get_has_nsfw_concepts(pipe, features, params):
+ return pipe._get_has_nsfw_concepts(features, params)
+
+
+def unshard(x: jnp.ndarray):
+ # einops.rearrange(x, 'd b ... -> (d b) ...')
+ num_devices, batch_size = x.shape[:2]
+ rest = x.shape[2:]
+ return x.reshape(num_devices * batch_size, *rest)
+
+
+def preprocess(image, dtype):
+ image = image.convert("RGB")
+ w, h = image.size
+ w, h = (x - x % 64 for x in (w, h)) # resize to integer multiple of 64
+ image = image.resize((w, h), resample=PIL_INTERPOLATION["lanczos"])
+ image = jnp.array(image).astype(dtype) / 255.0
+ image = image[None].transpose(0, 3, 1, 2)
+ return image
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..34c59795de328f2fa2cbe610625d866e4710d4a2
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/__init__.py
@@ -0,0 +1,48 @@
+from typing import TYPE_CHECKING
+
+from ...utils import (
+ DIFFUSERS_SLOW_IMPORT,
+ OptionalDependencyNotAvailable,
+ _LazyModule,
+ get_objects_from_module,
+ is_torch_available,
+ is_transformers_available,
+)
+
+
+_dummy_objects = {}
+_import_structure = {}
+
+
+try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+ from ...utils import dummy_torch_and_transformers_objects # noqa F403
+
+ _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
+else:
+ _import_structure["pipeline_hunyuandit_controlnet"] = ["HunyuanDiTControlNetPipeline"]
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ try:
+ if not (is_transformers_available() and is_torch_available()):
+ raise OptionalDependencyNotAvailable()
+
+ except OptionalDependencyNotAvailable:
+ from ...utils.dummy_torch_and_transformers_objects import *
+ else:
+ from .pipeline_hunyuandit_controlnet import HunyuanDiTControlNetPipeline
+
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
+
+ for name, value in _dummy_objects.items():
+ setattr(sys.modules[__name__], name, value)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/__pycache__/pipeline_hunyuandit_controlnet.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/__pycache__/pipeline_hunyuandit_controlnet.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..e0c93274f8627b785b99acf48d80f0af977e2bfb
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/__pycache__/pipeline_hunyuandit_controlnet.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/pipeline_hunyuandit_controlnet.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/pipeline_hunyuandit_controlnet.py
new file mode 100644
index 0000000000000000000000000000000000000000..9b9adf49012539a8793e66c4cec8c91b4fb0de11
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/controlnet_hunyuandit/pipeline_hunyuandit_controlnet.py
@@ -0,0 +1,1058 @@
+# Copyright 2025 HunyuanDiT Authors and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import inspect
+from typing import Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import torch
+from transformers import BertModel, BertTokenizer, CLIPImageProcessor, MT5Tokenizer, T5EncoderModel
+
+from diffusers.pipelines.stable_diffusion import StableDiffusionPipelineOutput
+
+from ...callbacks import MultiPipelineCallbacks, PipelineCallback
+from ...image_processor import PipelineImageInput, VaeImageProcessor
+from ...models import AutoencoderKL, HunyuanDiT2DControlNetModel, HunyuanDiT2DModel, HunyuanDiT2DMultiControlNetModel
+from ...models.embeddings import get_2d_rotary_pos_embed
+from ...pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+from ...schedulers import DDPMScheduler
+from ...utils import (
+ is_torch_xla_available,
+ logging,
+ replace_example_docstring,
+)
+from ...utils.torch_utils import randn_tensor
+from ..pipeline_utils import DiffusionPipeline
+
+
+if is_torch_xla_available():
+ import torch_xla.core.xla_model as xm
+
+ XLA_AVAILABLE = True
+else:
+ XLA_AVAILABLE = False
+
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ from diffusers import HunyuanDiT2DControlNetModel, HunyuanDiTControlNetPipeline
+ import torch
+
+ controlnet = HunyuanDiT2DControlNetModel.from_pretrained(
+ "Tencent-Hunyuan/HunyuanDiT-v1.1-ControlNet-Diffusers-Canny", torch_dtype=torch.float16
+ )
+
+ pipe = HunyuanDiTControlNetPipeline.from_pretrained(
+ "Tencent-Hunyuan/HunyuanDiT-v1.1-Diffusers", controlnet=controlnet, torch_dtype=torch.float16
+ )
+ pipe.to("cuda")
+
+ from diffusers.utils import load_image
+
+ cond_image = load_image(
+ "https://huggingface.co/Tencent-Hunyuan/HunyuanDiT-v1.1-ControlNet-Diffusers-Canny/resolve/main/canny.jpg?download=true"
+ )
+
+ ## You may also use English prompt as HunyuanDiT supports both English and Chinese
+ prompt = "在夜晚的酒店门前，一座古老的中国风格的狮子雕像矗立着，它的眼睛闪烁着光芒，仿佛在守护着这座建筑。背景是夜晚的酒店前，构图方式是特写，平视，居中构图。这张照片呈现了真实摄影风格，蕴含了中国雕塑文化，同时展现了神秘氛围"
+ # prompt="At night, an ancient Chinese-style lion statue stands in front of the hotel, its eyes gleaming as if guarding the building. The background is the hotel entrance at night, with a close-up, eye-level, and centered composition. This photo presents a realistic photographic style, embodies Chinese sculpture culture, and reveals a mysterious atmosphere."
+ image = pipe(
+ prompt,
+ height=1024,
+ width=1024,
+ control_image=cond_image,
+ num_inference_steps=50,
+ ).images[0]
+ ```
+"""
+
+STANDARD_RATIO = np.array(
+ [
+ 1.0, # 1:1
+ 4.0 / 3.0, # 4:3
+ 3.0 / 4.0, # 3:4
+ 16.0 / 9.0, # 16:9
+ 9.0 / 16.0, # 9:16
+ ]
+)
+STANDARD_SHAPE = [
+ [(1024, 1024), (1280, 1280)], # 1:1
+ [(1024, 768), (1152, 864), (1280, 960)], # 4:3
+ [(768, 1024), (864, 1152), (960, 1280)], # 3:4
+ [(1280, 768)], # 16:9
+ [(768, 1280)], # 9:16
+]
+STANDARD_AREA = [np.array([w * h for w, h in shapes]) for shapes in STANDARD_SHAPE]
+SUPPORTED_SHAPE = [
+ (1024, 1024),
+ (1280, 1280), # 1:1
+ (1024, 768),
+ (1152, 864),
+ (1280, 960), # 4:3
+ (768, 1024),
+ (864, 1152),
+ (960, 1280), # 3:4
+ (1280, 768), # 16:9
+ (768, 1280), # 9:16
+]
+
+
+def map_to_standard_shapes(target_width, target_height):
+ target_ratio = target_width / target_height
+ closest_ratio_idx = np.argmin(np.abs(STANDARD_RATIO - target_ratio))
+ closest_area_idx = np.argmin(np.abs(STANDARD_AREA[closest_ratio_idx] - target_width * target_height))
+ width, height = STANDARD_SHAPE[closest_ratio_idx][closest_area_idx]
+ return width, height
+
+
+def get_resize_crop_region_for_grid(src, tgt_size):
+ th = tw = tgt_size
+ h, w = src
+
+ r = h / w
+
+ # resize
+ if r > 1:
+ resize_height = th
+ resize_width = int(round(th / h * w))
+ else:
+ resize_width = tw
+ resize_height = int(round(tw / w * h))
+
+ crop_top = int(round((th - resize_height) / 2.0))
+ crop_left = int(round((tw - resize_width) / 2.0))
+
+ return (crop_top, crop_left), (crop_top + resize_height, crop_left + resize_width)
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
+def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
+ r"""
+ Rescales `noise_cfg` tensor based on `guidance_rescale` to improve image quality and fix overexposure. Based on
+ Section 3.4 from [Common Diffusion Noise Schedules and Sample Steps are
+ Flawed](https://huggingface.co/papers/2305.08891).
+
+ Args:
+ noise_cfg (`torch.Tensor`):
+ The predicted noise tensor for the guided diffusion process.
+ noise_pred_text (`torch.Tensor`):
+ The predicted noise tensor for the text-guided diffusion process.
+ guidance_rescale (`float`, *optional*, defaults to 0.0):
+ A rescale factor applied to the noise predictions.
+
+ Returns:
+ noise_cfg (`torch.Tensor`): The rescaled noise prediction tensor.
+ """
+ std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
+ std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
+ # rescale the results from guidance (fixes overexposure)
+ noise_pred_rescaled = noise_cfg * (std_text / std_cfg)
+ # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images
+ noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg
+ return noise_cfg
+
+
+class HunyuanDiTControlNetPipeline(DiffusionPipeline):
+ r"""
+ Pipeline for English/Chinese-to-image generation using HunyuanDiT.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ HunyuanDiT uses two text encoders: [mT5](https://huggingface.co/google/mt5-base) and [bilingual CLIP](fine-tuned by
+ ourselves)
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. We use
+ `sdxl-vae-fp16-fix`.
+ text_encoder (Optional[`~transformers.BertModel`, `~transformers.CLIPTextModel`]):
+ Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+ HunyuanDiT uses a fine-tuned [bilingual CLIP].
+ tokenizer (Optional[`~transformers.BertTokenizer`, `~transformers.CLIPTokenizer`]):
+ A `BertTokenizer` or `CLIPTokenizer` to tokenize text.
+ transformer ([`HunyuanDiT2DModel`]):
+ The HunyuanDiT model designed by Tencent Hunyuan.
+ text_encoder_2 (`T5EncoderModel`):
+ The mT5 embedder. Specifically, it is 't5-v1_1-xxl'.
+ tokenizer_2 (`MT5Tokenizer`):
+ The tokenizer for the mT5 embedder.
+ scheduler ([`DDPMScheduler`]):
+ A scheduler to be used in combination with HunyuanDiT to denoise the encoded image latents.
+ controlnet ([`HunyuanDiT2DControlNetModel`] or `List[HunyuanDiT2DControlNetModel]` or [`HunyuanDiT2DControlNetModel`]):
+ Provides additional conditioning to the `unet` during the denoising process. If you set multiple
+ ControlNets as a list, the outputs from each ControlNet are added together to create one combined
+ additional conditioning.
+ """
+
+ model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae"
+ _optional_components = [
+ "safety_checker",
+ "feature_extractor",
+ "text_encoder_2",
+ "tokenizer_2",
+ "text_encoder",
+ "tokenizer",
+ ]
+ _exclude_from_cpu_offload = ["safety_checker"]
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ "prompt_embeds_2",
+ "negative_prompt_embeds_2",
+ ]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: BertModel,
+ tokenizer: BertTokenizer,
+ transformer: HunyuanDiT2DModel,
+ scheduler: DDPMScheduler,
+ safety_checker: StableDiffusionSafetyChecker,
+ feature_extractor: CLIPImageProcessor,
+ controlnet: Union[
+ HunyuanDiT2DControlNetModel,
+ List[HunyuanDiT2DControlNetModel],
+ Tuple[HunyuanDiT2DControlNetModel],
+ HunyuanDiT2DMultiControlNetModel,
+ ],
+ text_encoder_2: Optional[T5EncoderModel] = None,
+ tokenizer_2: Optional[MT5Tokenizer] = None,
+ requires_safety_checker: bool = True,
+ ):
+ super().__init__()
+ if isinstance(controlnet, (list, tuple)):
+ controlnet = HunyuanDiT2DMultiControlNetModel(controlnet)
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ transformer=transformer,
+ scheduler=scheduler,
+ safety_checker=safety_checker,
+ feature_extractor=feature_extractor,
+ text_encoder_2=text_encoder_2,
+ controlnet=controlnet,
+ )
+
+ if safety_checker is None and requires_safety_checker:
+ logger.warning(
+ f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+ " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+ " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+ " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+ " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+ " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+ )
+
+ if safety_checker is not None and feature_extractor is None:
+ raise ValueError(
+ "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+ " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+ )
+
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+ self.register_to_config(requires_safety_checker=requires_safety_checker)
+ self.default_sample_size = (
+ self.transformer.config.sample_size
+ if hasattr(self, "transformer") and self.transformer is not None
+ else 128
+ )
+
+ # Copied from diffusers.pipelines.hunyuandit.pipeline_hunyuandit.HunyuanDiTPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ device: torch.device = None,
+ dtype: torch.dtype = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ max_sequence_length: Optional[int] = None,
+ text_encoder_index: int = 0,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ device: (`torch.device`):
+ torch device
+ dtype (`torch.dtype`):
+ torch dtype
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+ provided, text embeddings will be generated from `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ prompt_attention_mask (`torch.Tensor`, *optional*):
+ Attention mask for the prompt. Required when `prompt_embeds` is passed directly.
+ negative_prompt_attention_mask (`torch.Tensor`, *optional*):
+ Attention mask for the negative prompt. Required when `negative_prompt_embeds` is passed directly.
+ max_sequence_length (`int`, *optional*): maximum sequence length to use for the prompt.
+ text_encoder_index (`int`, *optional*):
+ Index of the text encoder to use. `0` for clip and `1` for T5.
+ """
+ if dtype is None:
+ if self.text_encoder_2 is not None:
+ dtype = self.text_encoder_2.dtype
+ elif self.transformer is not None:
+ dtype = self.transformer.dtype
+ else:
+ dtype = None
+
+ if device is None:
+ device = self._execution_device
+
+ tokenizers = [self.tokenizer, self.tokenizer_2]
+ text_encoders = [self.text_encoder, self.text_encoder_2]
+
+ tokenizer = tokenizers[text_encoder_index]
+ text_encoder = text_encoders[text_encoder_index]
+
+ if max_sequence_length is None:
+ if text_encoder_index == 0:
+ max_length = 77
+ if text_encoder_index == 1:
+ max_length = 256
+ else:
+ max_length = max_sequence_length
+
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ if prompt_embeds is None:
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_attention_mask=True,
+ return_tensors="pt",
+ )
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+ if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+ text_input_ids, untruncated_ids
+ ):
+ removed_text = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_attention_mask = text_inputs.attention_mask.to(device)
+ prompt_embeds = text_encoder(
+ text_input_ids.to(device),
+ attention_mask=prompt_attention_mask,
+ )
+ prompt_embeds = prompt_embeds[0]
+ prompt_attention_mask = prompt_attention_mask.repeat(num_images_per_prompt, 1)
+
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ # get unconditional embeddings for classifier free guidance
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ uncond_tokens: List[str]
+ if negative_prompt is None:
+ uncond_tokens = [""] * batch_size
+ elif prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif isinstance(negative_prompt, str):
+ uncond_tokens = [negative_prompt]
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = negative_prompt
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ uncond_tokens,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_attention_mask = uncond_input.attention_mask.to(device)
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ attention_mask=negative_prompt_attention_mask,
+ )
+ negative_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(num_images_per_prompt, 1)
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ return prompt_embeds, negative_prompt_embeds, prompt_attention_mask, negative_prompt_attention_mask
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.run_safety_checker
+ def run_safety_checker(self, image, device, dtype):
+ if self.safety_checker is None:
+ has_nsfw_concept = None
+ else:
+ if torch.is_tensor(image):
+ feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
+ else:
+ feature_extractor_input = self.image_processor.numpy_to_pil(image)
+ safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
+ image, has_nsfw_concept = self.safety_checker(
+ images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+ )
+ return image, has_nsfw_concept
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://huggingface.co/papers/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ # Copied from diffusers.pipelines.hunyuandit.pipeline_hunyuandit.HunyuanDiTPipeline.check_inputs
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ negative_prompt=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ prompt_attention_mask=None,
+ negative_prompt_attention_mask=None,
+ prompt_embeds_2=None,
+ negative_prompt_embeds_2=None,
+ prompt_attention_mask_2=None,
+ negative_prompt_attention_mask_2=None,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if height % 8 != 0 or width % 8 != 0:
+ raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+ if callback_on_step_end_tensor_inputs is not None and not all(
+ k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+ ):
+ raise ValueError(
+ f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is None and prompt_embeds_2 is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds_2`. Cannot leave both `prompt` and `prompt_embeds_2` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+ if prompt_embeds is not None and prompt_attention_mask is None:
+ raise ValueError("Must provide `prompt_attention_mask` when specifying `prompt_embeds`.")
+
+ if prompt_embeds_2 is not None and prompt_attention_mask_2 is None:
+ raise ValueError("Must provide `prompt_attention_mask_2` when specifying `prompt_embeds_2`.")
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if negative_prompt_embeds is not None and negative_prompt_attention_mask is None:
+ raise ValueError("Must provide `negative_prompt_attention_mask` when specifying `negative_prompt_embeds`.")
+
+ if negative_prompt_embeds_2 is not None and negative_prompt_attention_mask_2 is None:
+ raise ValueError(
+ "Must provide `negative_prompt_attention_mask_2` when specifying `negative_prompt_embeds_2`."
+ )
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+ if prompt_embeds_2 is not None and negative_prompt_embeds_2 is not None:
+ if prompt_embeds_2.shape != negative_prompt_embeds_2.shape:
+ raise ValueError(
+ "`prompt_embeds_2` and `negative_prompt_embeds_2` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds_2` {prompt_embeds_2.shape} != `negative_prompt_embeds_2`"
+ f" {negative_prompt_embeds_2.shape}."
+ )
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
+ def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ if latents is None:
+ latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+ else:
+ latents = latents.to(device)
+
+ # scale the initial noise by the standard deviation required by the scheduler
+ latents = latents * self.scheduler.init_noise_sigma
+ return latents
+
+ # Copied from diffusers.pipelines.controlnet_sd3.pipeline_stable_diffusion_3_controlnet.StableDiffusion3ControlNetPipeline.prepare_image
+ def prepare_image(
+ self,
+ image,
+ width,
+ height,
+ batch_size,
+ num_images_per_prompt,
+ device,
+ dtype,
+ do_classifier_free_guidance=False,
+ guess_mode=False,
+ ):
+ if isinstance(image, torch.Tensor):
+ pass
+ else:
+ image = self.image_processor.preprocess(image, height=height, width=width)
+
+ image_batch_size = image.shape[0]
+
+ if image_batch_size == 1:
+ repeat_by = batch_size
+ else:
+ # image batch size is the same as prompt batch size
+ repeat_by = num_images_per_prompt
+
+ image = image.repeat_interleave(repeat_by, dim=0)
+
+ image = image.to(device=device, dtype=dtype)
+
+ if do_classifier_free_guidance and not guess_mode:
+ image = torch.cat([image] * 2)
+
+ return image
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def guidance_rescale(self):
+ return self._guidance_rescale
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://huggingface.co/papers/2205.11487 . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @property
+ def interrupt(self):
+ return self._interrupt
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ num_inference_steps: Optional[int] = 50,
+ guidance_scale: Optional[float] = 5.0,
+ control_image: PipelineImageInput = None,
+ controlnet_conditioning_scale: Union[float, List[float]] = 1.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ num_images_per_prompt: Optional[int] = 1,
+ eta: Optional[float] = 0.0,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_embeds_2: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds_2: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ prompt_attention_mask_2: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask_2: Optional[torch.Tensor] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ guidance_rescale: float = 0.0,
+ original_size: Optional[Tuple[int, int]] = (1024, 1024),
+ target_size: Optional[Tuple[int, int]] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ use_resolution_binning: bool = True,
+ ):
+ r"""
+ The call function to the pipeline for generation with HunyuanDiT.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+ height (`int`):
+ The height in pixels of the generated image.
+ width (`int`):
+ The width in pixels of the generated image.
+ num_inference_steps (`int`, *optional*, defaults to 50):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference. This parameter is modulated by `strength`.
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ A higher guidance scale value encourages the model to generate images closely linked to the text
+ `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+ control_guidance_start (`float` or `List[float]`, *optional*, defaults to 0.0):
+ The percentage of total steps at which the ControlNet starts applying.
+ control_guidance_end (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The percentage of total steps at which the ControlNet stops applying.
+ control_image (`torch.Tensor`, `PIL.Image.Image`, `np.ndarray`, `List[torch.Tensor]`, `List[PIL.Image.Image]`, `List[np.ndarray]`,:
+ `List[List[torch.Tensor]]`, `List[List[np.ndarray]]` or `List[List[PIL.Image.Image]]`):
+ The ControlNet input condition to provide guidance to the `unet` for generation. If the type is
+ specified as `torch.Tensor`, it is passed to ControlNet as is. `PIL.Image.Image` can also be accepted
+ as an image. The dimensions of the output image defaults to `image`'s dimensions. If height and/or
+ width are passed, `image` is resized accordingly. If multiple ControlNets are specified in `init`,
+ images must be passed as a list such that each element of the list can be correctly batched for input
+ to a single ControlNet.
+ controlnet_conditioning_scale (`float` or `List[float]`, *optional*, defaults to 1.0):
+ The outputs of the ControlNet are multiplied by `controlnet_conditioning_scale` before they are added
+ to the residual in the original `unet`. If multiple ControlNets are specified in `init`, you can set
+ the corresponding scale as a list.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+ pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) from the [DDIM](https://huggingface.co/papers/2010.02502) paper. Only
+ applies to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+ generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ prompt_embeds_2 (`torch.Tensor`, *optional*):
+ Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+ provided, text embeddings are generated from the `prompt` input argument.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ negative_prompt_embeds_2 (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+ not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+ prompt_attention_mask (`torch.Tensor`, *optional*):
+ Attention mask for the prompt. Required when `prompt_embeds` is passed directly.
+ prompt_attention_mask_2 (`torch.Tensor`, *optional*):
+ Attention mask for the prompt. Required when `prompt_embeds_2` is passed directly.
+ negative_prompt_attention_mask (`torch.Tensor`, *optional*):
+ Attention mask for the negative prompt. Required when `negative_prompt_embeds` is passed directly.
+ negative_prompt_attention_mask_2 (`torch.Tensor`, *optional*):
+ Attention mask for the negative prompt. Required when `negative_prompt_embeds_2` is passed directly.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+ plain tuple.
+ callback_on_step_end (`Callable[[int, int, Dict], None]`, `PipelineCallback`, `MultiPipelineCallbacks`, *optional*):
+ A callback function or a list of callback functions to be called at the end of each denoising step.
+ callback_on_step_end_tensor_inputs (`List[str]`, *optional*):
+ A list of tensor inputs that should be passed to the callback function. If not defined, all tensor
+ inputs will be passed.
+ guidance_rescale (`float`, *optional*, defaults to 0.0):
+ Rescale the noise_cfg according to `guidance_rescale`. Based on findings of [Common Diffusion Noise
+ Schedules and Sample Steps are Flawed](https://huggingface.co/papers/2305.08891). See Section 3.4
+ original_size (`Tuple[int, int]`, *optional*, defaults to `(1024, 1024)`):
+ The original size of the image. Used to calculate the time ids.
+ target_size (`Tuple[int, int]`, *optional*):
+ The target size of the image. Used to calculate the time ids.
+ crops_coords_top_left (`Tuple[int, int]`, *optional*, defaults to `(0, 0)`):
+ The top left coordinates of the crop. Used to calculate the time ids.
+ use_resolution_binning (`bool`, *optional*, defaults to `True`):
+ Whether to use resolution binning or not. If `True`, the input resolution will be mapped to the closest
+ standard resolution. Supported resolutions are 1024x1024, 1280x1280, 1024x768, 1152x864, 1280x960,
+ 768x1024, 864x1152, 960x1280, 1280x768, and 768x1280. It is recommended to set this to `True`.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+ otherwise a `tuple` is returned where the first element is a list with the generated images and the
+ second element is a list of `bool`s indicating whether the corresponding generated image contains
+ "not-safe-for-work" (nsfw) content.
+ """
+
+ if isinstance(callback_on_step_end, (PipelineCallback, MultiPipelineCallbacks)):
+ callback_on_step_end_tensor_inputs = callback_on_step_end.tensor_inputs
+
+ # 0. default height and width
+ height = height or self.default_sample_size * self.vae_scale_factor
+ width = width or self.default_sample_size * self.vae_scale_factor
+ height = int((height // 16) * 16)
+ width = int((width // 16) * 16)
+
+ if use_resolution_binning and (height, width) not in SUPPORTED_SHAPE:
+ width, height = map_to_standard_shapes(width, height)
+ height = int(height)
+ width = int(width)
+ logger.warning(f"Reshaped to (height, width)=({height}, {width}), Supported shapes are {SUPPORTED_SHAPE}")
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ height,
+ width,
+ negative_prompt,
+ prompt_embeds,
+ negative_prompt_embeds,
+ prompt_attention_mask,
+ negative_prompt_attention_mask,
+ prompt_embeds_2,
+ negative_prompt_embeds_2,
+ prompt_attention_mask_2,
+ negative_prompt_attention_mask_2,
+ callback_on_step_end_tensor_inputs,
+ )
+ self._guidance_scale = guidance_scale
+ self._guidance_rescale = guidance_rescale
+ self._interrupt = False
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # 3. Encode input prompt
+
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ prompt_attention_mask,
+ negative_prompt_attention_mask,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ device=device,
+ dtype=self.transformer.dtype,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ prompt_attention_mask=prompt_attention_mask,
+ negative_prompt_attention_mask=negative_prompt_attention_mask,
+ max_sequence_length=77,
+ text_encoder_index=0,
+ )
+ (
+ prompt_embeds_2,
+ negative_prompt_embeds_2,
+ prompt_attention_mask_2,
+ negative_prompt_attention_mask_2,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ device=device,
+ dtype=self.transformer.dtype,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ prompt_embeds=prompt_embeds_2,
+ negative_prompt_embeds=negative_prompt_embeds_2,
+ prompt_attention_mask=prompt_attention_mask_2,
+ negative_prompt_attention_mask=negative_prompt_attention_mask_2,
+ max_sequence_length=256,
+ text_encoder_index=1,
+ )
+
+ # 4. Prepare control image
+ if isinstance(self.controlnet, HunyuanDiT2DControlNetModel):
+ control_image = self.prepare_image(
+ image=control_image,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=self.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=False,
+ )
+ height, width = control_image.shape[-2:]
+
+ control_image = self.vae.encode(control_image).latent_dist.sample()
+ control_image = control_image * self.vae.config.scaling_factor
+
+ elif isinstance(self.controlnet, HunyuanDiT2DMultiControlNetModel):
+ control_images = []
+
+ for control_image_ in control_image:
+ control_image_ = self.prepare_image(
+ image=control_image_,
+ width=width,
+ height=height,
+ batch_size=batch_size * num_images_per_prompt,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ dtype=self.dtype,
+ do_classifier_free_guidance=self.do_classifier_free_guidance,
+ guess_mode=False,
+ )
+
+ control_image_ = self.vae.encode(control_image_).latent_dist.sample()
+ control_image_ = control_image_ * self.vae.config.scaling_factor
+
+ control_images.append(control_image_)
+
+ control_image = control_images
+ else:
+ assert False
+
+ # 5. Prepare timesteps
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps = self.scheduler.timesteps
+
+ # 6. Prepare latent variables
+ num_channels_latents = self.transformer.config.in_channels
+ latents = self.prepare_latents(
+ batch_size * num_images_per_prompt,
+ num_channels_latents,
+ height,
+ width,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ latents,
+ )
+
+ # 7. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+ extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+ # 8. create image_rotary_emb, style embedding & time ids
+ grid_height = height // 8 // self.transformer.config.patch_size
+ grid_width = width // 8 // self.transformer.config.patch_size
+ base_size = 512 // 8 // self.transformer.config.patch_size
+ grid_crops_coords = get_resize_crop_region_for_grid((grid_height, grid_width), base_size)
+ image_rotary_emb = get_2d_rotary_pos_embed(
+ self.transformer.inner_dim // self.transformer.num_heads,
+ grid_crops_coords,
+ (grid_height, grid_width),
+ device=device,
+ output_type="pt",
+ )
+
+ style = torch.tensor([0], device=device)
+
+ target_size = target_size or (height, width)
+ add_time_ids = list(original_size + target_size + crops_coords_top_left)
+ add_time_ids = torch.tensor([add_time_ids], dtype=prompt_embeds.dtype)
+
+ if self.do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+ prompt_attention_mask = torch.cat([negative_prompt_attention_mask, prompt_attention_mask])
+ prompt_embeds_2 = torch.cat([negative_prompt_embeds_2, prompt_embeds_2])
+ prompt_attention_mask_2 = torch.cat([negative_prompt_attention_mask_2, prompt_attention_mask_2])
+ add_time_ids = torch.cat([add_time_ids] * 2, dim=0)
+ style = torch.cat([style] * 2, dim=0)
+
+ prompt_embeds = prompt_embeds.to(device=device)
+ prompt_attention_mask = prompt_attention_mask.to(device=device)
+ prompt_embeds_2 = prompt_embeds_2.to(device=device)
+ prompt_attention_mask_2 = prompt_attention_mask_2.to(device=device)
+ add_time_ids = add_time_ids.to(dtype=prompt_embeds.dtype, device=device).repeat(
+ batch_size * num_images_per_prompt, 1
+ )
+ style = style.to(device=device).repeat(batch_size * num_images_per_prompt)
+
+ # 9. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ self._num_timesteps = len(timesteps)
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ if self.interrupt:
+ continue
+
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # expand scalar t to 1-D tensor to match the 1st dim of latent_model_input
+ t_expand = torch.tensor([t] * latent_model_input.shape[0], device=device).to(
+ dtype=latent_model_input.dtype
+ )
+
+ # controlnet(s) inference
+ control_block_samples = self.controlnet(
+ latent_model_input,
+ t_expand,
+ encoder_hidden_states=prompt_embeds,
+ text_embedding_mask=prompt_attention_mask,
+ encoder_hidden_states_t5=prompt_embeds_2,
+ text_embedding_mask_t5=prompt_attention_mask_2,
+ image_meta_size=add_time_ids,
+ style=style,
+ image_rotary_emb=image_rotary_emb,
+ return_dict=False,
+ controlnet_cond=control_image,
+ conditioning_scale=controlnet_conditioning_scale,
+ )[0]
+
+ # predict the noise residual
+ noise_pred = self.transformer(
+ latent_model_input,
+ t_expand,
+ encoder_hidden_states=prompt_embeds,
+ text_embedding_mask=prompt_attention_mask,
+ encoder_hidden_states_t5=prompt_embeds_2,
+ text_embedding_mask_t5=prompt_attention_mask_2,
+ image_meta_size=add_time_ids,
+ style=style,
+ image_rotary_emb=image_rotary_emb,
+ return_dict=False,
+ controlnet_block_samples=control_block_samples,
+ )[0]
+
+ noise_pred, _ = noise_pred.chunk(2, dim=1)
+
+ # perform guidance
+ if self.do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ if self.do_classifier_free_guidance and guidance_rescale > 0.0:
+ # Based on 3.4. in https://huggingface.co/papers/2305.08891
+ noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+
+ if callback_on_step_end is not None:
+ callback_kwargs = {}
+ for k in callback_on_step_end_tensor_inputs:
+ callback_kwargs[k] = locals()[k]
+ callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+ latents = callback_outputs.pop("latents", latents)
+ prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+ negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+ prompt_embeds_2 = callback_outputs.pop("prompt_embeds_2", prompt_embeds_2)
+ negative_prompt_embeds_2 = callback_outputs.pop(
+ "negative_prompt_embeds_2", negative_prompt_embeds_2
+ )
+
+ if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+ progress_bar.update()
+
+ if XLA_AVAILABLE:
+ xm.mark_step()
+
+ if not output_type == "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+ image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+ else:
+ image = latents
+ has_nsfw_concept = None
+
+ if has_nsfw_concept is None:
+ do_denormalize = [True] * image.shape[0]
+ else:
+ do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+ image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+ # Offload all models
+ self.maybe_free_model_hooks()
+
+ if not return_dict:
+ return (image, has_nsfw_concept)
+
+ return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/alt_diffusion/__pycache__/pipeline_output.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/alt_diffusion/__pycache__/pipeline_output.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..6286f48012016be0bbc4eaa127d8fb309ed07c90
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/alt_diffusion/__pycache__/pipeline_output.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/pndm/__pycache__/__init__.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/pndm/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..04e927bc90b2f93a0cb179c300b3bedc565f25f1
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/pndm/__pycache__/__init__.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/pndm/__pycache__/pipeline_pndm.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/pndm/__pycache__/pipeline_pndm.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..22a367193e433fffe66571d3cb4dbc9995056083
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/pndm/__pycache__/pipeline_pndm.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..2c6b04af52d40e8a2bfa2aa5812b9fb8b1da06f5
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__init__.py
@@ -0,0 +1,19 @@
+from typing import TYPE_CHECKING
+
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+
+
+_import_structure = {"pipeline_repaint": ["RePaintPipeline"]}
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ from .pipeline_repaint import RePaintPipeline
+
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__pycache__/__init__.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__pycache__/__init__.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..10bde94ad7db8dacfc1ea405033fbe33de541ffa
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__pycache__/__init__.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__pycache__/pipeline_repaint.cpython-310.pyc b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__pycache__/pipeline_repaint.cpython-310.pyc
new file mode 100644
index 0000000000000000000000000000000000000000..ccaf4f5ae9abc1da41c6309d6d146b820e216fc4
Binary files /dev/null and b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/__pycache__/pipeline_repaint.cpython-310.pyc differ
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/pipeline_repaint.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/pipeline_repaint.py
new file mode 100644
index 0000000000000000000000000000000000000000..56c6007ae886f96d0f5142a655f59588a8d91efc
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/repaint/pipeline_repaint.py
@@ -0,0 +1,231 @@
+# Copyright 2025 ETH Zurich Computer Vision Lab and The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from typing import List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+
+from ....models import UNet2DModel
+from ....schedulers import RePaintScheduler
+from ....utils import PIL_INTERPOLATION, deprecate, logging
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.preprocess
+def _preprocess_image(image: Union[List, PIL.Image.Image, torch.Tensor]):
+ deprecation_message = "The preprocess method is deprecated and will be removed in diffusers 1.0.0. Please use VaeImageProcessor.preprocess(...) instead"
+ deprecate("preprocess", "1.0.0", deprecation_message, standard_warn=False)
+ if isinstance(image, torch.Tensor):
+ return image
+ elif isinstance(image, PIL.Image.Image):
+ image = [image]
+
+ if isinstance(image[0], PIL.Image.Image):
+ w, h = image[0].size
+ w, h = (x - x % 8 for x in (w, h)) # resize to integer multiple of 8
+
+ image = [np.array(i.resize((w, h), resample=PIL_INTERPOLATION["lanczos"]))[None, :] for i in image]
+ image = np.concatenate(image, axis=0)
+ image = np.array(image).astype(np.float32) / 255.0
+ image = image.transpose(0, 3, 1, 2)
+ image = 2.0 * image - 1.0
+ image = torch.from_numpy(image)
+ elif isinstance(image[0], torch.Tensor):
+ image = torch.cat(image, dim=0)
+ return image
+
+
+def _preprocess_mask(mask: Union[List, PIL.Image.Image, torch.Tensor]):
+ if isinstance(mask, torch.Tensor):
+ return mask
+ elif isinstance(mask, PIL.Image.Image):
+ mask = [mask]
+
+ if isinstance(mask[0], PIL.Image.Image):
+ w, h = mask[0].size
+ w, h = (x - x % 32 for x in (w, h)) # resize to integer multiple of 32
+ mask = [np.array(m.convert("L").resize((w, h), resample=PIL_INTERPOLATION["nearest"]))[None, :] for m in mask]
+ mask = np.concatenate(mask, axis=0)
+ mask = mask.astype(np.float32) / 255.0
+ mask[mask < 0.5] = 0
+ mask[mask >= 0.5] = 1
+ mask = torch.from_numpy(mask)
+ elif isinstance(mask[0], torch.Tensor):
+ mask = torch.cat(mask, dim=0)
+ return mask
+
+
+class RePaintPipeline(DiffusionPipeline):
+ r"""
+ Pipeline for image inpainting using RePaint.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ Parameters:
+ unet ([`UNet2DModel`]):
+ A `UNet2DModel` to denoise the encoded image latents.
+ scheduler ([`RePaintScheduler`]):
+ A `RePaintScheduler` to be used in combination with `unet` to denoise the encoded image.
+ """
+
+ unet: UNet2DModel
+ scheduler: RePaintScheduler
+ model_cpu_offload_seq = "unet"
+
+ def __init__(self, unet: UNet2DModel, scheduler: RePaintScheduler):
+ super().__init__()
+ self.register_modules(unet=unet, scheduler=scheduler)
+
+ @torch.no_grad()
+ def __call__(
+ self,
+ image: Union[torch.Tensor, PIL.Image.Image],
+ mask_image: Union[torch.Tensor, PIL.Image.Image],
+ num_inference_steps: int = 250,
+ eta: float = 0.0,
+ jump_length: int = 10,
+ jump_n_sample: int = 10,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ ) -> Union[ImagePipelineOutput, Tuple]:
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ image (`torch.Tensor` or `PIL.Image.Image`):
+ The original image to inpaint on.
+ mask_image (`torch.Tensor` or `PIL.Image.Image`):
+ The mask_image where 0.0 define which part of the original image to inpaint.
+ num_inference_steps (`int`, *optional*, defaults to 1000):
+ The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+ expense of slower inference.
+ eta (`float`):
+ The weight of the added noise in a diffusion step. Its value is between 0.0 and 1.0; 0.0 corresponds to
+ DDIM and 1.0 is the DDPM scheduler.
+ jump_length (`int`, *optional*, defaults to 10):
+ The number of steps taken forward in time before going backward in time for a single jump ("j" in
+ RePaint paper). Take a look at Figure 9 and 10 in the
+ [paper](https://huggingface.co/papers/2201.09865).
+ jump_n_sample (`int`, *optional*, defaults to 10):
+ The number of times to make a forward time jump for a given chosen time sample. Take a look at Figure 9
+ and 10 in the [paper](https://huggingface.co/papers/2201.09865).
+ generator (`torch.Generator`, *optional*):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ output_type (`str`, `optional`, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`ImagePipelineOutput`] instead of a plain tuple.
+
+ Example:
+
+ ```py
+ >>> from io import BytesIO
+ >>> import torch
+ >>> import PIL
+ >>> import requests
+ >>> from diffusers import RePaintPipeline, RePaintScheduler
+
+
+ >>> def download_image(url):
+ ... response = requests.get(url)
+ ... return PIL.Image.open(BytesIO(response.content)).convert("RGB")
+
+
+ >>> img_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/repaint/celeba_hq_256.png"
+ >>> mask_url = "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/repaint/mask_256.png"
+
+ >>> # Load the original image and the mask as PIL images
+ >>> original_image = download_image(img_url).resize((256, 256))
+ >>> mask_image = download_image(mask_url).resize((256, 256))
+
+ >>> # Load the RePaint scheduler and pipeline based on a pretrained DDPM model
+ >>> scheduler = RePaintScheduler.from_pretrained("google/ddpm-ema-celebahq-256")
+ >>> pipe = RePaintPipeline.from_pretrained("google/ddpm-ema-celebahq-256", scheduler=scheduler)
+ >>> pipe = pipe.to("cuda")
+
+ >>> generator = torch.Generator(device="cuda").manual_seed(0)
+ >>> output = pipe(
+ ... image=original_image,
+ ... mask_image=mask_image,
+ ... num_inference_steps=250,
+ ... eta=0.0,
+ ... jump_length=10,
+ ... jump_n_sample=10,
+ ... generator=generator,
+ ... )
+ >>> inpainted_image = output.images[0]
+ ```
+
+ Returns:
+ [`~pipelines.ImagePipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is
+ returned where the first element is a list with the generated images.
+ """
+
+ original_image = image
+
+ original_image = _preprocess_image(original_image)
+ original_image = original_image.to(device=self._execution_device, dtype=self.unet.dtype)
+ mask_image = _preprocess_mask(mask_image)
+ mask_image = mask_image.to(device=self._execution_device, dtype=self.unet.dtype)
+
+ batch_size = original_image.shape[0]
+
+ # sample gaussian noise to begin the loop
+ if isinstance(generator, list) and len(generator) != batch_size:
+ raise ValueError(
+ f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+ f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+ )
+
+ image_shape = original_image.shape
+ image = randn_tensor(image_shape, generator=generator, device=self._execution_device, dtype=self.unet.dtype)
+
+ # set step values
+ self.scheduler.set_timesteps(num_inference_steps, jump_length, jump_n_sample, self._execution_device)
+ self.scheduler.eta = eta
+
+ t_last = self.scheduler.timesteps[0] + 1
+ generator = generator[0] if isinstance(generator, list) else generator
+ for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)):
+ if t < t_last:
+ # predict the noise residual
+ model_output = self.unet(image, t).sample
+ # compute previous image: x_t -> x_t-1
+ image = self.scheduler.step(model_output, t, image, original_image, mask_image, generator).prev_sample
+
+ else:
+ # compute the reverse: x_t-1 -> x_t
+ image = self.scheduler.undo_step(image, t_last, generator)
+ t_last = t
+
+ image = (image / 2 + 0.5).clamp(0, 1)
+ image = image.cpu().permute(0, 2, 3, 1).numpy()
+ if output_type == "pil":
+ image = self.numpy_to_pil(image)
+
+ if not return_dict:
+ return (image,)
+
+ return ImagePipelineOutput(images=image)
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/score_sde_ve/__init__.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/score_sde_ve/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..87c167c3dbd26e0408a41ef197a42dc5eb7038d7
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/score_sde_ve/__init__.py
@@ -0,0 +1,19 @@
+from typing import TYPE_CHECKING
+
+from ....utils import DIFFUSERS_SLOW_IMPORT, _LazyModule
+
+
+_import_structure = {"pipeline_score_sde_ve": ["ScoreSdeVePipeline"]}
+
+if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
+ from .pipeline_score_sde_ve import ScoreSdeVePipeline
+
+else:
+ import sys
+
+ sys.modules[__name__] = _LazyModule(
+ __name__,
+ globals()["__file__"],
+ _import_structure,
+ module_spec=__spec__,
+ )
diff --git a/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/score_sde_ve/pipeline_score_sde_ve.py b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/score_sde_ve/pipeline_score_sde_ve.py
new file mode 100644
index 0000000000000000000000000000000000000000..3f04db7ad6990863edcc7b19c905475a07171cae
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/diffusers/pipelines/deprecated/score_sde_ve/pipeline_score_sde_ve.py
@@ -0,0 +1,109 @@
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from typing import List, Optional, Tuple, Union
+
+import torch
+
+from ....models import UNet2DModel
+from ....schedulers import ScoreSdeVeScheduler
+from ....utils.torch_utils import randn_tensor
+from ...pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+
+
+class ScoreSdeVePipeline(DiffusionPipeline):
+ r"""
+ Pipeline for unconditional image generation.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+ implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+ Parameters:
+ unet ([`UNet2DModel`]):
+ A `UNet2DModel` to denoise the encoded image.
+ scheduler ([`ScoreSdeVeScheduler`]):
+ A `ScoreSdeVeScheduler` to be used in combination with `unet` to denoise the encoded image.
+ """
+
+ unet: UNet2DModel
+ scheduler: ScoreSdeVeScheduler
+
+ def __init__(self, unet: UNet2DModel, scheduler: ScoreSdeVeScheduler):
+ super().__init__()
+ self.register_modules(unet=unet, scheduler=scheduler)
+
+ @torch.no_grad()
+ def __call__(
+ self,
+ batch_size: int = 1,
+ num_inference_steps: int = 2000,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ **kwargs,
+ ) -> Union[ImagePipelineOutput, Tuple]:
+ r"""
+ The call function to the pipeline for generation.
+
+ Args:
+ batch_size (`int`, *optional*, defaults to 1):
+ The number of images to generate.
+ generator (`torch.Generator`, `optional`):
+ A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+ generation deterministic.
+ output_type (`str`, `optional`, defaults to `"pil"`):
+ The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`ImagePipelineOutput`] instead of a plain tuple.
+
+ Returns:
+ [`~pipelines.ImagePipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is
+ returned where the first element is a list with the generated images.
+ """
+
+ img_size = self.unet.config.sample_size
+ shape = (batch_size, 3, img_size, img_size)
+
+ model = self.unet
+
+ sample = randn_tensor(shape, generator=generator) * self.scheduler.init_noise_sigma
+ sample = sample.to(self.device)
+
+ self.scheduler.set_timesteps(num_inference_steps)
+ self.scheduler.set_sigmas(num_inference_steps)
+
+ for i, t in enumerate(self.progress_bar(self.scheduler.timesteps)):
+ sigma_t = self.scheduler.sigmas[i] * torch.ones(shape[0], device=self.device)
+
+ # correction step
+ for _ in range(self.scheduler.config.correct_steps):
+ model_output = self.unet(sample, sigma_t).sample
+ sample = self.scheduler.step_correct(model_output, sample, generator=generator).prev_sample
+
+ # prediction step
+ model_output = model(sample, sigma_t).sample
+ output = self.scheduler.step_pred(model_output, t, sample, generator=generator)
+
+ sample, sample_mean = output.prev_sample, output.prev_sample_mean
+
+ sample = sample_mean.clamp(0, 1)
+ sample = sample.cpu().permute(0, 2, 3, 1).numpy()
+ if output_type == "pil":
+ sample = self.numpy_to_pil(sample)
+
+ if not return_dict:
+ return (sample,)
+
+ return ImagePipelineOutput(images=sample)
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/node/test_where_long_example/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/node/test_where_long_example/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..4d772c60395580f9b96a626d3d1f02b41c5e365b
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/node/test_where_long_example/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5e005879d530fd98aaa0595cbec4bb1ec22824d64f166f8313d071eaa65e17bd
+size 43
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..50a337900304ca56585d2279842817205589709e
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c6f15ce3daa142241c12d73939925d7a9ad0d95ecdc68e44ea5fb9181c836a3
+size 331
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..92082c77f2013b57d744364d8aad3312cd6e71b1
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e65ee1ba0ecea1aea5ff5a58efe04f24040f7b4d5811a9e5dfb50a11a95cb6f5
+size 411
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1size1/test_data_set_0/input_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1size1/test_data_set_0/input_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..ccc68470e033faf2d79ce3e039f4e82aaee560b6
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad1size1/test_data_set_0/input_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:055f4bf74f8c8f0fe1285268d4a0ec7244ecf4a35f0c46183733ad1031006c46
+size 26
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/model.onnx b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/model.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..a57220edbbe79fecfe37a7816951fa2265e7a53c
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e81b5f92462708e1ff39d75006bd9f8ba35b6f08e8f585823caa271710815bb0
+size 685
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/input_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/input_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a88ce3dfbbf2047fef2819f97f16784bde8ad4b6
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/input_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b02153a44147d942cc958b858e926c5043a8ebb6a71b542f479eeeb76686e54
+size 331
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..4db1a456872fa23215e550c6f2e432e13392b06b
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3fc2370839a5bd4562879eb3e02047be8fe1c2afc53f802d4da6f89d889e3f86
+size 411
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/model.onnx b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/model.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..11dbf841114f55944fcb9b8fbb38a571b3bb85c4
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65645f48212b3a34cfd6f1d96109421ccbf7a680aa1cd757948650d908556a46
+size 601
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/test_data_set_0/input_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/test_data_set_0/input_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..7df21a5ff73c17751a71f7664f2e9a559eb2e4ba
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_pad2size1/test_data_set_0/input_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b92fdfbfff19258ca9a1ee5df8404b38b85a3d0e004c4fca3c73884371aa7ada
+size 26
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/model.onnx b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/model.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..76b8e58da4d1c0ea3793425a7d09263e5698df29
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6d8e22558acd9e17f71883bc09b90cb4d6449d6772d98eea082e5da149d9d1e
+size 525
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/input_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/input_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..f963d176847f4b0c07003be12f4abdd88506c850
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/input_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f421cb144abf8a4fd5e4b6aa4fb99ccfa48d78738886e22ebff8efc9335246b9
+size 331
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..69b707a88eac6ae6b84b2ffb98e94580dadd9c37
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv1d_stride/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d29361ed8d237ae03f4460b5deb6828a9675aa2a99bbe633edc2f58c084e9448
+size 171
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/model.onnx b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/model.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..b7126ab14f7a09fcb1debad56ab5a4aa1cea5772
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb8df62b22401aa644e46e13b55b7ac5f3c3814e002ff939a4bbe112720fc066
+size 593
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/input_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/input_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d67a0a2316c2f6ad5b2215ee81a60ee0502fbc51
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/input_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4e066b8dac7d27dadf0d18760045c8db6ad2488da2b9b4d437af9449f6d7795
+size 853
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..5d3ae14acba23c7395142c60bf936636ac0f954c
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26593b3a6d313f14df4c3dc44649b5540a34c7f9563f5334d6fb4343e30e1870
+size 653
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..4dbdd5ab694bdd605ed72690a402e5c9022ffad8
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd13e6427aa4a23e1a55cf5b40884789a4938204c4deddc0b5411f2f50e175d3
+size 525
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/model.onnx b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/model.onnx
new file mode 100644
index 0000000000000000000000000000000000000000..d96078087bcf015503fa06e1ba662fe88ff756ff
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/model.onnx
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7e91a334e84ec5e532732b3c8a4323d3ae6294cef7fcb5b6c906f3da1f59aa84
+size 449
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/test_data_set_0/output_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/test_data_set_0/output_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..a9b54e444dabcf5070de83fc0aedc75d596a834e
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_padded/test_data_set_0/output_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c73bbba981d2a7548cffec8d1a37ea07d69ac56c40e858a1e1514b83e6e9cd0
+size 1165
diff --git a/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_with_multiplier/test_data_set_0/input_0.pb b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_with_multiplier/test_data_set_0/input_0.pb
new file mode 100644
index 0000000000000000000000000000000000000000..d70e50ce2e12ccafdb644e28215833e9368314c2
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/onnx/backend/test/data/pytorch-converted/test_Conv2d_depthwise_with_multiplier/test_data_set_0/input_0.pb
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e103ad686b0bd49ced278ea064dc53a0d2adfa1cef0bf6fe81abdb53e253aef7
+size 1165
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/cublas64_11.dll b/pythonProject/.venv/Lib/site-packages/torch/lib/cublas64_11.dll
new file mode 100644
index 0000000000000000000000000000000000000000..6ce85ca2b714cb077206238c6ac817ee1e69deea
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/cublas64_11.dll
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ca516b96b29c2fba2344909a896bc1cd7951f6cd11fe595a8a3929c02cccbed
+size 88668160
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/cudnn_heuristic64_9.dll b/pythonProject/.venv/Lib/site-packages/torch/lib/cudnn_heuristic64_9.dll
new file mode 100644
index 0000000000000000000000000000000000000000..7f3560045ac83e2ed7e0768f2eb445f836bb6074
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/cudnn_heuristic64_9.dll
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cb55585a02e1c8265c76bb82d4cf96401f4bc26390d18b46d1841feb42f237c
+size 85742136
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/curand64_10.dll b/pythonProject/.venv/Lib/site-packages/torch/lib/curand64_10.dll
new file mode 100644
index 0000000000000000000000000000000000000000..9318b9b51e384cefd149eefbb9bb0613d9b25e67
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/curand64_10.dll
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:047867bb615b0bfa986d237bf4cc1b78cf24065297863d1f730ae0b5398add2e
+size 65375744
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/fbgemm.lib b/pythonProject/.venv/Lib/site-packages/torch/lib/fbgemm.lib
new file mode 100644
index 0000000000000000000000000000000000000000..cf7f25bfb98637027d39a428c6168d1d92f2b248
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/fbgemm.lib
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7eef33a8cb398728d2e9bad2abbae2cef6d2d3128fe18b4dd98fcf455079537c
+size 1409606
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/fmt.lib b/pythonProject/.venv/Lib/site-packages/torch/lib/fmt.lib
new file mode 100644
index 0000000000000000000000000000000000000000..018015f2b067b58e41f3dc98114c8a91073cd644
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/fmt.lib
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c528fe692da844db553cb07aa5c1dbd03212a54b3ed3b47052bec971147dcbd
+size 1668942
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/kineto.lib b/pythonProject/.venv/Lib/site-packages/torch/lib/kineto.lib
new file mode 100644
index 0000000000000000000000000000000000000000..5504fa6b9250654d82e8721e2effbaefd449ff9c
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/kineto.lib
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b5a28336a6f4efe1a58137552f44bb86327207c3351a10691a1acc47969cd4f
+size 22359482
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/libiomp5md.dll b/pythonProject/.venv/Lib/site-packages/torch/lib/libiomp5md.dll
new file mode 100644
index 0000000000000000000000000000000000000000..88756476e508d2052b30405d4edc53842a95e663
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/libiomp5md.dll
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61f8b815afa84395ce892f4ba4430fc6702c87e9c21ae0e17e0c3f14bc4d6906
+size 1250312
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/libomp140.x86_64.dll b/pythonProject/.venv/Lib/site-packages/torch/lib/libomp140.x86_64.dll
new file mode 100644
index 0000000000000000000000000000000000000000..da89fdebea77a57b28dc2eb524f58e37a0a90475
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/libomp140.x86_64.dll
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3b884bf8be99d47edbae2afc9bd5fb3431b7be4ae937fb3ce02ba82f6608560
+size 698888
diff --git a/pythonProject/.venv/Lib/site-packages/torch/lib/libprotobuf-lite.lib b/pythonProject/.venv/Lib/site-packages/torch/lib/libprotobuf-lite.lib
new file mode 100644
index 0000000000000000000000000000000000000000..2c9c6e94b3ac6d91dbf9d358a2e1f8cad9dd7207
--- /dev/null
+++ b/pythonProject/.venv/Lib/site-packages/torch/lib/libprotobuf-lite.lib
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdac4fa9a60d7912851021fff10bbfd050791b1b813aaca2e7caf4bd0cf00f49
+size 5168824
diff --git a/pythonProject/diffusers-main/utils/check_inits.py b/pythonProject/diffusers-main/utils/check_inits.py
new file mode 100644
index 0000000000000000000000000000000000000000..8208fa634186536767054e8fb5f0d6f81bee7ba0
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/check_inits.py
@@ -0,0 +1,299 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import collections
+import importlib.util
+import os
+import re
+from pathlib import Path
+
+
+PATH_TO_TRANSFORMERS = "src/transformers"
+
+
+# Matches is_xxx_available()
+_re_backend = re.compile(r"is\_([a-z_]*)_available()")
+# Catches a one-line _import_struct = {xxx}
+_re_one_line_import_struct = re.compile(r"^_import_structure\s+=\s+\{([^\}]+)\}")
+# Catches a line with a key-values pattern: "bla": ["foo", "bar"]
+_re_import_struct_key_value = re.compile(r'\s+"\S*":\s+\[([^\]]*)\]')
+# Catches a line if not is_foo_available
+_re_test_backend = re.compile(r"^\s*if\s+not\s+is\_[a-z_]*\_available\(\)")
+# Catches a line _import_struct["bla"].append("foo")
+_re_import_struct_add_one = re.compile(r'^\s*_import_structure\["\S*"\]\.append\("(\S*)"\)')
+# Catches a line _import_struct["bla"].extend(["foo", "bar"]) or _import_struct["bla"] = ["foo", "bar"]
+_re_import_struct_add_many = re.compile(r"^\s*_import_structure\[\S*\](?:\.extend\(|\s*=\s+)\[([^\]]*)\]")
+# Catches a line with an object between quotes and a comma: "MyModel",
+_re_quote_object = re.compile(r'^\s+"([^"]+)",')
+# Catches a line with objects between brackets only: ["foo", "bar"],
+_re_between_brackets = re.compile(r"^\s+\[([^\]]+)\]")
+# Catches a line with from foo import bar, bla, boo
+_re_import = re.compile(r"\s+from\s+\S*\s+import\s+([^\(\s].*)\n")
+# Catches a line with try:
+_re_try = re.compile(r"^\s*try:")
+# Catches a line with else:
+_re_else = re.compile(r"^\s*else:")
+
+
+def find_backend(line):
+ """Find one (or multiple) backend in a code line of the init."""
+ if _re_test_backend.search(line) is None:
+ return None
+ backends = [b[0] for b in _re_backend.findall(line)]
+ backends.sort()
+ return "_and_".join(backends)
+
+
+def parse_init(init_file):
+ """
+ Read an init_file and parse (per backend) the _import_structure objects defined and the TYPE_CHECKING objects
+ defined
+ """
+ with open(init_file, "r", encoding="utf-8", newline="\n") as f:
+ lines = f.readlines()
+
+ line_index = 0
+ while line_index < len(lines) and not lines[line_index].startswith("_import_structure = {"):
+ line_index += 1
+
+ # If this is a traditional init, just return.
+ if line_index >= len(lines):
+ return None
+
+ # First grab the objects without a specific backend in _import_structure
+ objects = []
+ while not lines[line_index].startswith("if TYPE_CHECKING") and find_backend(lines[line_index]) is None:
+ line = lines[line_index]
+ # If we have everything on a single line, let's deal with it.
+ if _re_one_line_import_struct.search(line):
+ content = _re_one_line_import_struct.search(line).groups()[0]
+ imports = re.findall(r"\[([^\]]+)\]", content)
+ for imp in imports:
+ objects.extend([obj[1:-1] for obj in imp.split(", ")])
+ line_index += 1
+ continue
+ single_line_import_search = _re_import_struct_key_value.search(line)
+ if single_line_import_search is not None:
+ imports = [obj[1:-1] for obj in single_line_import_search.groups()[0].split(", ") if len(obj) > 0]
+ objects.extend(imports)
+ elif line.startswith(" " * 8 + '"'):
+ objects.append(line[9:-3])
+ line_index += 1
+
+ import_dict_objects = {"none": objects}
+ # Let's continue with backend-specific objects in _import_structure
+ while not lines[line_index].startswith("if TYPE_CHECKING"):
+ # If the line is an if not is_backend_available, we grab all objects associated.
+ backend = find_backend(lines[line_index])
+ # Check if the backend declaration is inside a try block:
+ if _re_try.search(lines[line_index - 1]) is None:
+ backend = None
+
+ if backend is not None:
+ line_index += 1
+
+ # Scroll until we hit the else block of try-except-else
+ while _re_else.search(lines[line_index]) is None:
+ line_index += 1
+
+ line_index += 1
+
+ objects = []
+ # Until we unindent, add backend objects to the list
+ while len(lines[line_index]) <= 1 or lines[line_index].startswith(" " * 4):
+ line = lines[line_index]
+ if _re_import_struct_add_one.search(line) is not None:
+ objects.append(_re_import_struct_add_one.search(line).groups()[0])
+ elif _re_import_struct_add_many.search(line) is not None:
+ imports = _re_import_struct_add_many.search(line).groups()[0].split(", ")
+ imports = [obj[1:-1] for obj in imports if len(obj) > 0]
+ objects.extend(imports)
+ elif _re_between_brackets.search(line) is not None:
+ imports = _re_between_brackets.search(line).groups()[0].split(", ")
+ imports = [obj[1:-1] for obj in imports if len(obj) > 0]
+ objects.extend(imports)
+ elif _re_quote_object.search(line) is not None:
+ objects.append(_re_quote_object.search(line).groups()[0])
+ elif line.startswith(" " * 8 + '"'):
+ objects.append(line[9:-3])
+ elif line.startswith(" " * 12 + '"'):
+ objects.append(line[13:-3])
+ line_index += 1
+
+ import_dict_objects[backend] = objects
+ else:
+ line_index += 1
+
+ # At this stage we are in the TYPE_CHECKING part, first grab the objects without a specific backend
+ objects = []
+ while (
+ line_index < len(lines)
+ and find_backend(lines[line_index]) is None
+ and not lines[line_index].startswith("else")
+ ):
+ line = lines[line_index]
+ single_line_import_search = _re_import.search(line)
+ if single_line_import_search is not None:
+ objects.extend(single_line_import_search.groups()[0].split(", "))
+ elif line.startswith(" " * 8):
+ objects.append(line[8:-2])
+ line_index += 1
+
+ type_hint_objects = {"none": objects}
+ # Let's continue with backend-specific objects
+ while line_index < len(lines):
+ # If the line is an if is_backend_available, we grab all objects associated.
+ backend = find_backend(lines[line_index])
+ # Check if the backend declaration is inside a try block:
+ if _re_try.search(lines[line_index - 1]) is None:
+ backend = None
+
+ if backend is not None:
+ line_index += 1
+
+ # Scroll until we hit the else block of try-except-else
+ while _re_else.search(lines[line_index]) is None:
+ line_index += 1
+
+ line_index += 1
+
+ objects = []
+ # Until we unindent, add backend objects to the list
+ while len(lines[line_index]) <= 1 or lines[line_index].startswith(" " * 8):
+ line = lines[line_index]
+ single_line_import_search = _re_import.search(line)
+ if single_line_import_search is not None:
+ objects.extend(single_line_import_search.groups()[0].split(", "))
+ elif line.startswith(" " * 12):
+ objects.append(line[12:-2])
+ line_index += 1
+
+ type_hint_objects[backend] = objects
+ else:
+ line_index += 1
+
+ return import_dict_objects, type_hint_objects
+
+
+def analyze_results(import_dict_objects, type_hint_objects):
+ """
+ Analyze the differences between _import_structure objects and TYPE_CHECKING objects found in an init.
+ """
+
+ def find_duplicates(seq):
+ return [k for k, v in collections.Counter(seq).items() if v > 1]
+
+ if list(import_dict_objects.keys()) != list(type_hint_objects.keys()):
+ return ["Both sides of the init do not have the same backends!"]
+
+ errors = []
+ for key in import_dict_objects.keys():
+ duplicate_imports = find_duplicates(import_dict_objects[key])
+ if duplicate_imports:
+ errors.append(f"Duplicate _import_structure definitions for: {duplicate_imports}")
+ duplicate_type_hints = find_duplicates(type_hint_objects[key])
+ if duplicate_type_hints:
+ errors.append(f"Duplicate TYPE_CHECKING objects for: {duplicate_type_hints}")
+
+ if sorted(set(import_dict_objects[key])) != sorted(set(type_hint_objects[key])):
+ name = "base imports" if key == "none" else f"{key} backend"
+ errors.append(f"Differences for {name}:")
+ for a in type_hint_objects[key]:
+ if a not in import_dict_objects[key]:
+ errors.append(f" {a} in TYPE_HINT but not in _import_structure.")
+ for a in import_dict_objects[key]:
+ if a not in type_hint_objects[key]:
+ errors.append(f" {a} in _import_structure but not in TYPE_HINT.")
+ return errors
+
+
+def check_all_inits():
+ """
+ Check all inits in the transformers repo and raise an error if at least one does not define the same objects in
+ both halves.
+ """
+ failures = []
+ for root, _, files in os.walk(PATH_TO_TRANSFORMERS):
+ if "__init__.py" in files:
+ fname = os.path.join(root, "__init__.py")
+ objects = parse_init(fname)
+ if objects is not None:
+ errors = analyze_results(*objects)
+ if len(errors) > 0:
+ errors[0] = f"Problem in {fname}, both halves do not define the same objects.\n{errors[0]}"
+ failures.append("\n".join(errors))
+ if len(failures) > 0:
+ raise ValueError("\n\n".join(failures))
+
+
+def get_transformers_submodules():
+ """
+ Returns the list of Transformers submodules.
+ """
+ submodules = []
+ for path, directories, files in os.walk(PATH_TO_TRANSFORMERS):
+ for folder in directories:
+ # Ignore private modules
+ if folder.startswith("_"):
+ directories.remove(folder)
+ continue
+ # Ignore leftovers from branches (empty folders apart from pycache)
+ if len(list((Path(path) / folder).glob("*.py"))) == 0:
+ continue
+ short_path = str((Path(path) / folder).relative_to(PATH_TO_TRANSFORMERS))
+ submodule = short_path.replace(os.path.sep, ".")
+ submodules.append(submodule)
+ for fname in files:
+ if fname == "__init__.py":
+ continue
+ short_path = str((Path(path) / fname).relative_to(PATH_TO_TRANSFORMERS))
+ submodule = short_path.replace(".py", "").replace(os.path.sep, ".")
+ if len(submodule.split(".")) == 1:
+ submodules.append(submodule)
+ return submodules
+
+
+IGNORE_SUBMODULES = [
+ "convert_pytorch_checkpoint_to_tf2",
+ "modeling_flax_pytorch_utils",
+]
+
+
+def check_submodules():
+ # This is to make sure the transformers module imported is the one in the repo.
+ spec = importlib.util.spec_from_file_location(
+ "transformers",
+ os.path.join(PATH_TO_TRANSFORMERS, "__init__.py"),
+ submodule_search_locations=[PATH_TO_TRANSFORMERS],
+ )
+ transformers = spec.loader.load_module()
+
+ module_not_registered = [
+ module
+ for module in get_transformers_submodules()
+ if module not in IGNORE_SUBMODULES and module not in transformers._import_structure.keys()
+ ]
+ if len(module_not_registered) > 0:
+ list_of_modules = "\n".join(f"- {module}" for module in module_not_registered)
+ raise ValueError(
+ "The following submodules are not properly registered in the main init of Transformers:\n"
+ f"{list_of_modules}\n"
+ "Make sure they appear somewhere in the keys of `_import_structure` with an empty list as value."
+ )
+
+
+if __name__ == "__main__":
+ check_all_inits()
+ check_submodules()
diff --git a/pythonProject/diffusers-main/utils/check_repo.py b/pythonProject/diffusers-main/utils/check_repo.py
new file mode 100644
index 0000000000000000000000000000000000000000..14bdbe60adf06e4b6a3f79378f68393cc083b6a3
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/check_repo.py
@@ -0,0 +1,755 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib
+import inspect
+import os
+import re
+import warnings
+from collections import OrderedDict
+from difflib import get_close_matches
+from pathlib import Path
+
+from diffusers.models.auto import get_values
+from diffusers.utils import ENV_VARS_TRUE_VALUES, is_flax_available, is_torch_available
+
+
+# All paths are set with the intent you should run this script from the root of the repo with the command
+# python utils/check_repo.py
+PATH_TO_DIFFUSERS = "src/diffusers"
+PATH_TO_TESTS = "tests"
+PATH_TO_DOC = "docs/source/en"
+
+# Update this list with models that are supposed to be private.
+PRIVATE_MODELS = [
+ "DPRSpanPredictor",
+ "RealmBertModel",
+ "T5Stack",
+ "TFDPRSpanPredictor",
+]
+
+# Update this list for models that are not tested with a comment explaining the reason it should not be.
+# Being in this list is an exception and should **not** be the rule.
+IGNORE_NON_TESTED = PRIVATE_MODELS.copy() + [
+ # models to ignore for not tested
+ "OPTDecoder", # Building part of bigger (tested) model.
+ "DecisionTransformerGPT2Model", # Building part of bigger (tested) model.
+ "SegformerDecodeHead", # Building part of bigger (tested) model.
+ "PLBartEncoder", # Building part of bigger (tested) model.
+ "PLBartDecoder", # Building part of bigger (tested) model.
+ "PLBartDecoderWrapper", # Building part of bigger (tested) model.
+ "BigBirdPegasusEncoder", # Building part of bigger (tested) model.
+ "BigBirdPegasusDecoder", # Building part of bigger (tested) model.
+ "BigBirdPegasusDecoderWrapper", # Building part of bigger (tested) model.
+ "DetrEncoder", # Building part of bigger (tested) model.
+ "DetrDecoder", # Building part of bigger (tested) model.
+ "DetrDecoderWrapper", # Building part of bigger (tested) model.
+ "M2M100Encoder", # Building part of bigger (tested) model.
+ "M2M100Decoder", # Building part of bigger (tested) model.
+ "Speech2TextEncoder", # Building part of bigger (tested) model.
+ "Speech2TextDecoder", # Building part of bigger (tested) model.
+ "LEDEncoder", # Building part of bigger (tested) model.
+ "LEDDecoder", # Building part of bigger (tested) model.
+ "BartDecoderWrapper", # Building part of bigger (tested) model.
+ "BartEncoder", # Building part of bigger (tested) model.
+ "BertLMHeadModel", # Needs to be setup as decoder.
+ "BlenderbotSmallEncoder", # Building part of bigger (tested) model.
+ "BlenderbotSmallDecoderWrapper", # Building part of bigger (tested) model.
+ "BlenderbotEncoder", # Building part of bigger (tested) model.
+ "BlenderbotDecoderWrapper", # Building part of bigger (tested) model.
+ "MBartEncoder", # Building part of bigger (tested) model.
+ "MBartDecoderWrapper", # Building part of bigger (tested) model.
+ "MegatronBertLMHeadModel", # Building part of bigger (tested) model.
+ "MegatronBertEncoder", # Building part of bigger (tested) model.
+ "MegatronBertDecoder", # Building part of bigger (tested) model.
+ "MegatronBertDecoderWrapper", # Building part of bigger (tested) model.
+ "PegasusEncoder", # Building part of bigger (tested) model.
+ "PegasusDecoderWrapper", # Building part of bigger (tested) model.
+ "DPREncoder", # Building part of bigger (tested) model.
+ "ProphetNetDecoderWrapper", # Building part of bigger (tested) model.
+ "RealmBertModel", # Building part of bigger (tested) model.
+ "RealmReader", # Not regular model.
+ "RealmScorer", # Not regular model.
+ "RealmForOpenQA", # Not regular model.
+ "ReformerForMaskedLM", # Needs to be setup as decoder.
+ "Speech2Text2DecoderWrapper", # Building part of bigger (tested) model.
+ "TFDPREncoder", # Building part of bigger (tested) model.
+ "TFElectraMainLayer", # Building part of bigger (tested) model (should it be a TFModelMixin ?)
+ "TFRobertaForMultipleChoice", # TODO: fix
+ "TrOCRDecoderWrapper", # Building part of bigger (tested) model.
+ "SeparableConv1D", # Building part of bigger (tested) model.
+ "FlaxBartForCausalLM", # Building part of bigger (tested) model.
+ "FlaxBertForCausalLM", # Building part of bigger (tested) model. Tested implicitly through FlaxRobertaForCausalLM.
+ "OPTDecoderWrapper",
+]
+
+# Update this list with test files that don't have a tester with a `all_model_classes` variable and which don't
+# trigger the common tests.
+TEST_FILES_WITH_NO_COMMON_TESTS = [
+ "models/decision_transformer/test_modeling_decision_transformer.py",
+ "models/camembert/test_modeling_camembert.py",
+ "models/mt5/test_modeling_flax_mt5.py",
+ "models/mbart/test_modeling_mbart.py",
+ "models/mt5/test_modeling_mt5.py",
+ "models/pegasus/test_modeling_pegasus.py",
+ "models/camembert/test_modeling_tf_camembert.py",
+ "models/mt5/test_modeling_tf_mt5.py",
+ "models/xlm_roberta/test_modeling_tf_xlm_roberta.py",
+ "models/xlm_roberta/test_modeling_flax_xlm_roberta.py",
+ "models/xlm_prophetnet/test_modeling_xlm_prophetnet.py",
+ "models/xlm_roberta/test_modeling_xlm_roberta.py",
+ "models/vision_text_dual_encoder/test_modeling_vision_text_dual_encoder.py",
+ "models/vision_text_dual_encoder/test_modeling_flax_vision_text_dual_encoder.py",
+ "models/decision_transformer/test_modeling_decision_transformer.py",
+]
+
+# Update this list for models that are not in any of the auto MODEL_XXX_MAPPING. Being in this list is an exception and
+# should **not** be the rule.
+IGNORE_NON_AUTO_CONFIGURED = PRIVATE_MODELS.copy() + [
+ # models to ignore for model xxx mapping
+ "DPTForDepthEstimation",
+ "DecisionTransformerGPT2Model",
+ "GLPNForDepthEstimation",
+ "ViltForQuestionAnswering",
+ "ViltForImagesAndTextClassification",
+ "ViltForImageAndTextRetrieval",
+ "ViltForMaskedLM",
+ "XGLMEncoder",
+ "XGLMDecoder",
+ "XGLMDecoderWrapper",
+ "PerceiverForMultimodalAutoencoding",
+ "PerceiverForOpticalFlow",
+ "SegformerDecodeHead",
+ "FlaxBeitForMaskedImageModeling",
+ "PLBartEncoder",
+ "PLBartDecoder",
+ "PLBartDecoderWrapper",
+ "BeitForMaskedImageModeling",
+ "CLIPTextModel",
+ "CLIPVisionModel",
+ "TFCLIPTextModel",
+ "TFCLIPVisionModel",
+ "FlaxCLIPTextModel",
+ "FlaxCLIPVisionModel",
+ "FlaxWav2Vec2ForCTC",
+ "DetrForSegmentation",
+ "DPRReader",
+ "FlaubertForQuestionAnswering",
+ "FlavaImageCodebook",
+ "FlavaTextModel",
+ "FlavaImageModel",
+ "FlavaMultimodalModel",
+ "GPT2DoubleHeadsModel",
+ "LukeForMaskedLM",
+ "LukeForEntityClassification",
+ "LukeForEntityPairClassification",
+ "LukeForEntitySpanClassification",
+ "OpenAIGPTDoubleHeadsModel",
+ "RagModel",
+ "RagSequenceForGeneration",
+ "RagTokenForGeneration",
+ "RealmEmbedder",
+ "RealmForOpenQA",
+ "RealmScorer",
+ "RealmReader",
+ "TFDPRReader",
+ "TFGPT2DoubleHeadsModel",
+ "TFOpenAIGPTDoubleHeadsModel",
+ "TFRagModel",
+ "TFRagSequenceForGeneration",
+ "TFRagTokenForGeneration",
+ "Wav2Vec2ForCTC",
+ "HubertForCTC",
+ "SEWForCTC",
+ "SEWDForCTC",
+ "XLMForQuestionAnswering",
+ "XLNetForQuestionAnswering",
+ "SeparableConv1D",
+ "VisualBertForRegionToPhraseAlignment",
+ "VisualBertForVisualReasoning",
+ "VisualBertForQuestionAnswering",
+ "VisualBertForMultipleChoice",
+ "TFWav2Vec2ForCTC",
+ "TFHubertForCTC",
+ "MaskFormerForInstanceSegmentation",
+]
+
+# Update this list for models that have multiple model types for the same
+# model doc
+MODEL_TYPE_TO_DOC_MAPPING = OrderedDict(
+ [
+ ("data2vec-text", "data2vec"),
+ ("data2vec-audio", "data2vec"),
+ ("data2vec-vision", "data2vec"),
+ ]
+)
+
+
+# This is to make sure the transformers module imported is the one in the repo.
+spec = importlib.util.spec_from_file_location(
+ "diffusers",
+ os.path.join(PATH_TO_DIFFUSERS, "__init__.py"),
+ submodule_search_locations=[PATH_TO_DIFFUSERS],
+)
+diffusers = spec.loader.load_module()
+
+
+def check_model_list():
+ """Check the model list inside the transformers library."""
+ # Get the models from the directory structure of `src/diffusers/models/`
+ models_dir = os.path.join(PATH_TO_DIFFUSERS, "models")
+ _models = []
+ for model in os.listdir(models_dir):
+ model_dir = os.path.join(models_dir, model)
+ if os.path.isdir(model_dir) and "__init__.py" in os.listdir(model_dir):
+ _models.append(model)
+
+ # Get the models from the directory structure of `src/transformers/models/`
+ models = [model for model in dir(diffusers.models) if not model.startswith("__")]
+
+ missing_models = sorted(set(_models).difference(models))
+ if missing_models:
+ raise Exception(
+ f"The following models should be included in {models_dir}/__init__.py: {','.join(missing_models)}."
+ )
+
+
+# If some modeling modules should be ignored for all checks, they should be added in the nested list
+# _ignore_modules of this function.
+def get_model_modules():
+ """Get the model modules inside the transformers library."""
+ _ignore_modules = [
+ "modeling_auto",
+ "modeling_encoder_decoder",
+ "modeling_marian",
+ "modeling_mmbt",
+ "modeling_outputs",
+ "modeling_retribert",
+ "modeling_utils",
+ "modeling_flax_auto",
+ "modeling_flax_encoder_decoder",
+ "modeling_flax_utils",
+ "modeling_speech_encoder_decoder",
+ "modeling_flax_speech_encoder_decoder",
+ "modeling_flax_vision_encoder_decoder",
+ "modeling_transfo_xl_utilities",
+ "modeling_tf_auto",
+ "modeling_tf_encoder_decoder",
+ "modeling_tf_outputs",
+ "modeling_tf_pytorch_utils",
+ "modeling_tf_utils",
+ "modeling_tf_transfo_xl_utilities",
+ "modeling_tf_vision_encoder_decoder",
+ "modeling_vision_encoder_decoder",
+ ]
+ modules = []
+ for model in dir(diffusers.models):
+ # There are some magic dunder attributes in the dir, we ignore them
+ if not model.startswith("__"):
+ model_module = getattr(diffusers.models, model)
+ for submodule in dir(model_module):
+ if submodule.startswith("modeling") and submodule not in _ignore_modules:
+ modeling_module = getattr(model_module, submodule)
+ if inspect.ismodule(modeling_module):
+ modules.append(modeling_module)
+ return modules
+
+
+def get_models(module, include_pretrained=False):
+ """Get the objects in module that are models."""
+ models = []
+ model_classes = (diffusers.ModelMixin, diffusers.TFModelMixin, diffusers.FlaxModelMixin)
+ for attr_name in dir(module):
+ if not include_pretrained and ("Pretrained" in attr_name or "PreTrained" in attr_name):
+ continue
+ attr = getattr(module, attr_name)
+ if isinstance(attr, type) and issubclass(attr, model_classes) and attr.__module__ == module.__name__:
+ models.append((attr_name, attr))
+ return models
+
+
+def is_a_private_model(model):
+ """Returns True if the model should not be in the main init."""
+ if model in PRIVATE_MODELS:
+ return True
+
+ # Wrapper, Encoder and Decoder are all privates
+ if model.endswith("Wrapper"):
+ return True
+ if model.endswith("Encoder"):
+ return True
+ if model.endswith("Decoder"):
+ return True
+ return False
+
+
+def check_models_are_in_init():
+ """Checks all models defined in the library are in the main init."""
+ models_not_in_init = []
+ dir_transformers = dir(diffusers)
+ for module in get_model_modules():
+ models_not_in_init += [
+ model[0] for model in get_models(module, include_pretrained=True) if model[0] not in dir_transformers
+ ]
+
+ # Remove private models
+ models_not_in_init = [model for model in models_not_in_init if not is_a_private_model(model)]
+ if len(models_not_in_init) > 0:
+ raise Exception(f"The following models should be in the main init: {','.join(models_not_in_init)}.")
+
+
+# If some test_modeling files should be ignored when checking models are all tested, they should be added in the
+# nested list _ignore_files of this function.
+def get_model_test_files():
+ """Get the model test files.
+
+ The returned files should NOT contain the `tests` (i.e. `PATH_TO_TESTS` defined in this script). They will be
+ considered as paths relative to `tests`. A caller has to use `os.path.join(PATH_TO_TESTS, ...)` to access the files.
+ """
+
+ _ignore_files = [
+ "test_modeling_common",
+ "test_modeling_encoder_decoder",
+ "test_modeling_flax_encoder_decoder",
+ "test_modeling_flax_speech_encoder_decoder",
+ "test_modeling_marian",
+ "test_modeling_tf_common",
+ "test_modeling_tf_encoder_decoder",
+ ]
+ test_files = []
+ # Check both `PATH_TO_TESTS` and `PATH_TO_TESTS/models`
+ model_test_root = os.path.join(PATH_TO_TESTS, "models")
+ model_test_dirs = []
+ for x in os.listdir(model_test_root):
+ x = os.path.join(model_test_root, x)
+ if os.path.isdir(x):
+ model_test_dirs.append(x)
+
+ for target_dir in [PATH_TO_TESTS] + model_test_dirs:
+ for file_or_dir in os.listdir(target_dir):
+ path = os.path.join(target_dir, file_or_dir)
+ if os.path.isfile(path):
+ filename = os.path.split(path)[-1]
+ if "test_modeling" in filename and os.path.splitext(filename)[0] not in _ignore_files:
+ file = os.path.join(*path.split(os.sep)[1:])
+ test_files.append(file)
+
+ return test_files
+
+
+# This is a bit hacky but I didn't find a way to import the test_file as a module and read inside the tester class
+# for the all_model_classes variable.
+def find_tested_models(test_file):
+ """Parse the content of test_file to detect what's in all_model_classes"""
+ # This is a bit hacky but I didn't find a way to import the test_file as a module and read inside the class
+ with open(os.path.join(PATH_TO_TESTS, test_file), "r", encoding="utf-8", newline="\n") as f:
+ content = f.read()
+ all_models = re.findall(r"all_model_classes\s+=\s+\(\s*\(([^\)]*)\)", content)
+ # Check with one less parenthesis as well
+ all_models += re.findall(r"all_model_classes\s+=\s+\(([^\)]*)\)", content)
+ if len(all_models) > 0:
+ model_tested = []
+ for entry in all_models:
+ for line in entry.split(","):
+ name = line.strip()
+ if len(name) > 0:
+ model_tested.append(name)
+ return model_tested
+
+
+def check_models_are_tested(module, test_file):
+ """Check models defined in module are tested in test_file."""
+ # XxxModelMixin are not tested
+ defined_models = get_models(module)
+ tested_models = find_tested_models(test_file)
+ if tested_models is None:
+ if test_file.replace(os.path.sep, "/") in TEST_FILES_WITH_NO_COMMON_TESTS:
+ return
+ return [
+ f"{test_file} should define `all_model_classes` to apply common tests to the models it tests. "
+ + "If this intentional, add the test filename to `TEST_FILES_WITH_NO_COMMON_TESTS` in the file "
+ + "`utils/check_repo.py`."
+ ]
+ failures = []
+ for model_name, _ in defined_models:
+ if model_name not in tested_models and model_name not in IGNORE_NON_TESTED:
+ failures.append(
+ f"{model_name} is defined in {module.__name__} but is not tested in "
+ + f"{os.path.join(PATH_TO_TESTS, test_file)}. Add it to the all_model_classes in that file."
+ + "If common tests should not applied to that model, add its name to `IGNORE_NON_TESTED`"
+ + "in the file `utils/check_repo.py`."
+ )
+ return failures
+
+
+def check_all_models_are_tested():
+ """Check all models are properly tested."""
+ modules = get_model_modules()
+ test_files = get_model_test_files()
+ failures = []
+ for module in modules:
+ test_file = [file for file in test_files if f"test_{module.__name__.split('.')[-1]}.py" in file]
+ if len(test_file) == 0:
+ failures.append(f"{module.__name__} does not have its corresponding test file {test_file}.")
+ elif len(test_file) > 1:
+ failures.append(f"{module.__name__} has several test files: {test_file}.")
+ else:
+ test_file = test_file[0]
+ new_failures = check_models_are_tested(module, test_file)
+ if new_failures is not None:
+ failures += new_failures
+ if len(failures) > 0:
+ raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))
+
+
+def get_all_auto_configured_models():
+ """Return the list of all models in at least one auto class."""
+ result = set() # To avoid duplicates we concatenate all model classes in a set.
+ if is_torch_available():
+ for attr_name in dir(diffusers.models.auto.modeling_auto):
+ if attr_name.startswith("MODEL_") and attr_name.endswith("MAPPING_NAMES"):
+ result = result | set(get_values(getattr(diffusers.models.auto.modeling_auto, attr_name)))
+ if is_flax_available():
+ for attr_name in dir(diffusers.models.auto.modeling_flax_auto):
+ if attr_name.startswith("FLAX_MODEL_") and attr_name.endswith("MAPPING_NAMES"):
+ result = result | set(get_values(getattr(diffusers.models.auto.modeling_flax_auto, attr_name)))
+ return list(result)
+
+
+def ignore_unautoclassed(model_name):
+ """Rules to determine if `name` should be in an auto class."""
+ # Special white list
+ if model_name in IGNORE_NON_AUTO_CONFIGURED:
+ return True
+ # Encoder and Decoder should be ignored
+ if "Encoder" in model_name or "Decoder" in model_name:
+ return True
+ return False
+
+
+def check_models_are_auto_configured(module, all_auto_models):
+ """Check models defined in module are each in an auto class."""
+ defined_models = get_models(module)
+ failures = []
+ for model_name, _ in defined_models:
+ if model_name not in all_auto_models and not ignore_unautoclassed(model_name):
+ failures.append(
+ f"{model_name} is defined in {module.__name__} but is not present in any of the auto mapping. "
+ "If that is intended behavior, add its name to `IGNORE_NON_AUTO_CONFIGURED` in the file "
+ "`utils/check_repo.py`."
+ )
+ return failures
+
+
+def check_all_models_are_auto_configured():
+ """Check all models are each in an auto class."""
+ missing_backends = []
+ if not is_torch_available():
+ missing_backends.append("PyTorch")
+ if not is_flax_available():
+ missing_backends.append("Flax")
+ if len(missing_backends) > 0:
+ missing = ", ".join(missing_backends)
+ if os.getenv("TRANSFORMERS_IS_CI", "").upper() in ENV_VARS_TRUE_VALUES:
+ raise Exception(
+ "Full quality checks require all backends to be installed (with `pip install -e .[dev]` in the "
+ f"Transformers repo, the following are missing: {missing}."
+ )
+ else:
+ warnings.warn(
+ "Full quality checks require all backends to be installed (with `pip install -e .[dev]` in the "
+ f"Transformers repo, the following are missing: {missing}. While it's probably fine as long as you "
+ "didn't make any change in one of those backends modeling files, you should probably execute the "
+ "command above to be on the safe side."
+ )
+ modules = get_model_modules()
+ all_auto_models = get_all_auto_configured_models()
+ failures = []
+ for module in modules:
+ new_failures = check_models_are_auto_configured(module, all_auto_models)
+ if new_failures is not None:
+ failures += new_failures
+ if len(failures) > 0:
+ raise Exception(f"There were {len(failures)} failures:\n" + "\n".join(failures))
+
+
+_re_decorator = re.compile(r"^\s*@(\S+)\s+$")
+
+
+def check_decorator_order(filename):
+ """Check that in the test file `filename` the slow decorator is always last."""
+ with open(filename, "r", encoding="utf-8", newline="\n") as f:
+ lines = f.readlines()
+ decorator_before = None
+ errors = []
+ for i, line in enumerate(lines):
+ search = _re_decorator.search(line)
+ if search is not None:
+ decorator_name = search.groups()[0]
+ if decorator_before is not None and decorator_name.startswith("parameterized"):
+ errors.append(i)
+ decorator_before = decorator_name
+ elif decorator_before is not None:
+ decorator_before = None
+ return errors
+
+
+def check_all_decorator_order():
+ """Check that in all test files, the slow decorator is always last."""
+ errors = []
+ for fname in os.listdir(PATH_TO_TESTS):
+ if fname.endswith(".py"):
+ filename = os.path.join(PATH_TO_TESTS, fname)
+ new_errors = check_decorator_order(filename)
+ errors += [f"- {filename}, line {i}" for i in new_errors]
+ if len(errors) > 0:
+ msg = "\n".join(errors)
+ raise ValueError(
+ "The parameterized decorator (and its variants) should always be first, but this is not the case in the"
+ f" following files:\n{msg}"
+ )
+
+
+def find_all_documented_objects():
+ """Parse the content of all doc files to detect which classes and functions it documents"""
+ documented_obj = []
+ for doc_file in Path(PATH_TO_DOC).glob("**/*.rst"):
+ with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
+ content = f.read()
+ raw_doc_objs = re.findall(r"(?:autoclass|autofunction):: transformers.(\S+)\s+", content)
+ documented_obj += [obj.split(".")[-1] for obj in raw_doc_objs]
+ for doc_file in Path(PATH_TO_DOC).glob("**/*.md"):
+ with open(doc_file, "r", encoding="utf-8", newline="\n") as f:
+ content = f.read()
+ raw_doc_objs = re.findall(r"\[\[autodoc\]\]\s+(\S+)\s+", content)
+ documented_obj += [obj.split(".")[-1] for obj in raw_doc_objs]
+ return documented_obj
+
+
+# One good reason for not being documented is to be deprecated. Put in this list deprecated objects.
+DEPRECATED_OBJECTS = [
+ "AutoModelWithLMHead",
+ "BartPretrainedModel",
+ "DataCollator",
+ "DataCollatorForSOP",
+ "GlueDataset",
+ "GlueDataTrainingArguments",
+ "LineByLineTextDataset",
+ "LineByLineWithRefDataset",
+ "LineByLineWithSOPTextDataset",
+ "PretrainedBartModel",
+ "PretrainedFSMTModel",
+ "SingleSentenceClassificationProcessor",
+ "SquadDataTrainingArguments",
+ "SquadDataset",
+ "SquadExample",
+ "SquadFeatures",
+ "SquadV1Processor",
+ "SquadV2Processor",
+ "TFAutoModelWithLMHead",
+ "TFBartPretrainedModel",
+ "TextDataset",
+ "TextDatasetForNextSentencePrediction",
+ "Wav2Vec2ForMaskedLM",
+ "Wav2Vec2Tokenizer",
+ "glue_compute_metrics",
+ "glue_convert_examples_to_features",
+ "glue_output_modes",
+ "glue_processors",
+ "glue_tasks_num_labels",
+ "squad_convert_examples_to_features",
+ "xnli_compute_metrics",
+ "xnli_output_modes",
+ "xnli_processors",
+ "xnli_tasks_num_labels",
+ "TFTrainer",
+ "TFTrainingArguments",
+]
+
+# Exceptionally, some objects should not be documented after all rules passed.
+# ONLY PUT SOMETHING IN THIS LIST AS A LAST RESORT!
+UNDOCUMENTED_OBJECTS = [
+ "AddedToken", # This is a tokenizers class.
+ "BasicTokenizer", # Internal, should never have been in the main init.
+ "CharacterTokenizer", # Internal, should never have been in the main init.
+ "DPRPretrainedReader", # Like an Encoder.
+ "DummyObject", # Just picked by mistake sometimes.
+ "MecabTokenizer", # Internal, should never have been in the main init.
+ "ModelCard", # Internal type.
+ "SqueezeBertModule", # Internal building block (should have been called SqueezeBertLayer)
+ "TFDPRPretrainedReader", # Like an Encoder.
+ "TransfoXLCorpus", # Internal type.
+ "WordpieceTokenizer", # Internal, should never have been in the main init.
+ "absl", # External module
+ "add_end_docstrings", # Internal, should never have been in the main init.
+ "add_start_docstrings", # Internal, should never have been in the main init.
+ "cached_path", # Internal used for downloading models.
+ "convert_tf_weight_name_to_pt_weight_name", # Internal used to convert model weights
+ "logger", # Internal logger
+ "logging", # External module
+ "requires_backends", # Internal function
+]
+
+# This list should be empty. Objects in it should get their own doc page.
+SHOULD_HAVE_THEIR_OWN_PAGE = [
+ # Benchmarks
+ "PyTorchBenchmark",
+ "PyTorchBenchmarkArguments",
+ "TensorFlowBenchmark",
+ "TensorFlowBenchmarkArguments",
+]
+
+
+def ignore_undocumented(name):
+ """Rules to determine if `name` should be undocumented."""
+ # NOT DOCUMENTED ON PURPOSE.
+ # Constants uppercase are not documented.
+ if name.isupper():
+ return True
+ # ModelMixins / Encoders / Decoders / Layers / Embeddings / Attention are not documented.
+ if (
+ name.endswith("ModelMixin")
+ or name.endswith("Decoder")
+ or name.endswith("Encoder")
+ or name.endswith("Layer")
+ or name.endswith("Embeddings")
+ or name.endswith("Attention")
+ ):
+ return True
+ # Submodules are not documented.
+ if os.path.isdir(os.path.join(PATH_TO_DIFFUSERS, name)) or os.path.isfile(
+ os.path.join(PATH_TO_DIFFUSERS, f"{name}.py")
+ ):
+ return True
+ # All load functions are not documented.
+ if name.startswith("load_tf") or name.startswith("load_pytorch"):
+ return True
+ # is_xxx_available functions are not documented.
+ if name.startswith("is_") and name.endswith("_available"):
+ return True
+ # Deprecated objects are not documented.
+ if name in DEPRECATED_OBJECTS or name in UNDOCUMENTED_OBJECTS:
+ return True
+ # MMBT model does not really work.
+ if name.startswith("MMBT"):
+ return True
+ if name in SHOULD_HAVE_THEIR_OWN_PAGE:
+ return True
+ return False
+
+
+def check_all_objects_are_documented():
+ """Check all models are properly documented."""
+ documented_objs = find_all_documented_objects()
+ modules = diffusers._modules
+ objects = [c for c in dir(diffusers) if c not in modules and not c.startswith("_")]
+ undocumented_objs = [c for c in objects if c not in documented_objs and not ignore_undocumented(c)]
+ if len(undocumented_objs) > 0:
+ raise Exception(
+ "The following objects are in the public init so should be documented:\n - "
+ + "\n - ".join(undocumented_objs)
+ )
+ check_docstrings_are_in_md()
+ check_model_type_doc_match()
+
+
+def check_model_type_doc_match():
+ """Check all doc pages have a corresponding model type."""
+ model_doc_folder = Path(PATH_TO_DOC) / "model_doc"
+ model_docs = [m.stem for m in model_doc_folder.glob("*.md")]
+
+ model_types = list(diffusers.models.auto.configuration_auto.MODEL_NAMES_MAPPING.keys())
+ model_types = [MODEL_TYPE_TO_DOC_MAPPING[m] if m in MODEL_TYPE_TO_DOC_MAPPING else m for m in model_types]
+
+ errors = []
+ for m in model_docs:
+ if m not in model_types and m != "auto":
+ close_matches = get_close_matches(m, model_types)
+ error_message = f"{m} is not a proper model identifier."
+ if len(close_matches) > 0:
+ close_matches = "/".join(close_matches)
+ error_message += f" Did you mean {close_matches}?"
+ errors.append(error_message)
+
+ if len(errors) > 0:
+ raise ValueError(
+ "Some model doc pages do not match any existing model type:\n"
+ + "\n".join(errors)
+ + "\nYou can add any missing model type to the `MODEL_NAMES_MAPPING` constant in "
+ "models/auto/configuration_auto.py."
+ )
+
+
+# Re pattern to catch :obj:`xx`, :class:`xx`, :func:`xx` or :meth:`xx`.
+_re_rst_special_words = re.compile(r":(?:obj|func|class|meth):`([^`]+)`")
+# Re pattern to catch things between double backquotes.
+_re_double_backquotes = re.compile(r"(^|[^`])``([^`]+)``([^`]|$)")
+# Re pattern to catch example introduction.
+_re_rst_example = re.compile(r"^\s*Example.*::\s*$", flags=re.MULTILINE)
+
+
+def is_rst_docstring(docstring):
+ """
+ Returns `True` if `docstring` is written in rst.
+ """
+ if _re_rst_special_words.search(docstring) is not None:
+ return True
+ if _re_double_backquotes.search(docstring) is not None:
+ return True
+ if _re_rst_example.search(docstring) is not None:
+ return True
+ return False
+
+
+def check_docstrings_are_in_md():
+ """Check all docstrings are in md"""
+ files_with_rst = []
+ for file in Path(PATH_TO_DIFFUSERS).glob("**/*.py"):
+ with open(file, "r") as f:
+ code = f.read()
+ docstrings = code.split('"""')
+
+ for idx, docstring in enumerate(docstrings):
+ if idx % 2 == 0 or not is_rst_docstring(docstring):
+ continue
+ files_with_rst.append(file)
+ break
+
+ if len(files_with_rst) > 0:
+ raise ValueError(
+ "The following files have docstrings written in rst:\n"
+ + "\n".join([f"- {f}" for f in files_with_rst])
+ + "\nTo fix this run `doc-builder convert path_to_py_file` after installing `doc-builder`\n"
+ "(`pip install git+https://github.com/huggingface/doc-builder`)"
+ )
+
+
+def check_repo_quality():
+ """Check all models are properly tested and documented."""
+ print("Checking all models are included.")
+ check_model_list()
+ print("Checking all models are public.")
+ check_models_are_in_init()
+ print("Checking all models are properly tested.")
+ check_all_decorator_order()
+ check_all_models_are_tested()
+ print("Checking all objects are properly documented.")
+ check_all_objects_are_documented()
+ print("Checking all models are in at least one auto class.")
+ check_all_models_are_auto_configured()
+
+
+if __name__ == "__main__":
+ check_repo_quality()
diff --git a/pythonProject/diffusers-main/utils/check_support_list.py b/pythonProject/diffusers-main/utils/check_support_list.py
new file mode 100644
index 0000000000000000000000000000000000000000..ade9df3b64faa2e9daa227e36c036df4e69116bc
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/check_support_list.py
@@ -0,0 +1,124 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+"""
+Utility that checks that modules like attention processors are listed in the documentation file.
+
+```bash
+python utils/check_support_list.py
+```
+
+It has no auto-fix mode.
+"""
+
+import os
+import re
+
+
+# All paths are set with the intent that you run this script from the root of the repo
+REPO_PATH = "."
+
+
+def read_documented_classes(doc_path, autodoc_regex=r"\[\[autodoc\]\]\s([^\n]+)"):
+ """
+ Reads documented classes from a doc file using a regex to find lines like [[autodoc]] my.module.Class.
+ Returns a list of documented class names (just the class name portion).
+ """
+ with open(os.path.join(REPO_PATH, doc_path), "r") as f:
+ doctext = f.read()
+ matches = re.findall(autodoc_regex, doctext)
+ return [match.split(".")[-1] for match in matches]
+
+
+def read_source_classes(src_path, class_regex, exclude_conditions=None):
+ """
+ Reads class names from a source file using a regex that captures class definitions.
+ Optionally exclude classes based on a list of conditions (functions that take class name and return bool).
+ """
+ if exclude_conditions is None:
+ exclude_conditions = []
+ with open(os.path.join(REPO_PATH, src_path), "r") as f:
+ doctext = f.read()
+ classes = re.findall(class_regex, doctext)
+ # Filter out classes that meet any of the exclude conditions
+ filtered_classes = [c for c in classes if not any(cond(c) for cond in exclude_conditions)]
+ return filtered_classes
+
+
+def check_documentation(doc_path, src_path, doc_regex, src_regex, exclude_conditions=None):
+ """
+ Generic function to check if all classes defined in `src_path` are documented in `doc_path`.
+ Returns a set of undocumented class names.
+ """
+ documented = set(read_documented_classes(doc_path, doc_regex))
+ source_classes = set(read_source_classes(src_path, src_regex, exclude_conditions=exclude_conditions))
+
+ # Find which classes in source are not documented in a deterministic way.
+ undocumented = sorted(source_classes - documented)
+ return undocumented
+
+
+if __name__ == "__main__":
+ # Define the checks we need to perform
+ checks = {
+ "Attention Processors": {
+ "doc_path": "docs/source/en/api/attnprocessor.md",
+ "src_path": "src/diffusers/models/attention_processor.py",
+ "doc_regex": r"\[\[autodoc\]\]\s([^\n]+)",
+ "src_regex": r"class\s+(\w+Processor(?:\d*_?\d*))[:(]",
+ "exclude_conditions": [lambda c: "LoRA" in c, lambda c: c == "Attention"],
+ },
+ "Image Processors": {
+ "doc_path": "docs/source/en/api/image_processor.md",
+ "src_path": "src/diffusers/image_processor.py",
+ "doc_regex": r"\[\[autodoc\]\]\s([^\n]+)",
+ "src_regex": r"class\s+(\w+Processor(?:\d*_?\d*))[:(]",
+ },
+ "Activations": {
+ "doc_path": "docs/source/en/api/activations.md",
+ "src_path": "src/diffusers/models/activations.py",
+ "doc_regex": r"\[\[autodoc\]\]\s([^\n]+)",
+ "src_regex": r"class\s+(\w+)\s*\(.*?nn\.Module.*?\):",
+ },
+ "Normalizations": {
+ "doc_path": "docs/source/en/api/normalization.md",
+ "src_path": "src/diffusers/models/normalization.py",
+ "doc_regex": r"\[\[autodoc\]\]\s([^\n]+)",
+ "src_regex": r"class\s+(\w+)\s*\(.*?nn\.Module.*?\):",
+ "exclude_conditions": [
+ # Exclude LayerNorm as it's an intentional exception
+ lambda c: c == "LayerNorm"
+ ],
+ },
+ "LoRA Mixins": {
+ "doc_path": "docs/source/en/api/loaders/lora.md",
+ "src_path": "src/diffusers/loaders/lora_pipeline.py",
+ "doc_regex": r"\[\[autodoc\]\]\s([^\n]+)",
+ "src_regex": r"class\s+(\w+LoraLoaderMixin(?:\d*_?\d*))[:(]",
+ },
+ }
+
+ missing_items = {}
+ for category, params in checks.items():
+ undocumented = check_documentation(
+ doc_path=params["doc_path"],
+ src_path=params["src_path"],
+ doc_regex=params["doc_regex"],
+ src_regex=params["src_regex"],
+ exclude_conditions=params.get("exclude_conditions"),
+ )
+ if undocumented:
+ missing_items[category] = undocumented
+
+ # If we have any missing items, raise a single combined error
+ if missing_items:
+ error_msg = ["Some classes are not documented properly:\n"]
+ for category, classes in missing_items.items():
+ error_msg.append(f"- {category}: {', '.join(sorted(classes))}")
+ raise ValueError("\n".join(error_msg))
diff --git a/pythonProject/diffusers-main/utils/check_table.py b/pythonProject/diffusers-main/utils/check_table.py
new file mode 100644
index 0000000000000000000000000000000000000000..83c29aa74eca042df81386a73c12ffaf128fc4c2
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/check_table.py
@@ -0,0 +1,185 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+import collections
+import importlib.util
+import os
+import re
+
+
+# All paths are set with the intent you should run this script from the root of the repo with the command
+# python utils/check_table.py
+TRANSFORMERS_PATH = "src/diffusers"
+PATH_TO_DOCS = "docs/source/en"
+REPO_PATH = "."
+
+
+def _find_text_in_file(filename, start_prompt, end_prompt):
+ """
+ Find the text in `filename` between a line beginning with `start_prompt` and before `end_prompt`, removing empty
+ lines.
+ """
+ with open(filename, "r", encoding="utf-8", newline="\n") as f:
+ lines = f.readlines()
+ # Find the start prompt.
+ start_index = 0
+ while not lines[start_index].startswith(start_prompt):
+ start_index += 1
+ start_index += 1
+
+ end_index = start_index
+ while not lines[end_index].startswith(end_prompt):
+ end_index += 1
+ end_index -= 1
+
+ while len(lines[start_index]) <= 1:
+ start_index += 1
+ while len(lines[end_index]) <= 1:
+ end_index -= 1
+ end_index += 1
+ return "".join(lines[start_index:end_index]), start_index, end_index, lines
+
+
+# Add here suffixes that are used to identify models, separated by |
+ALLOWED_MODEL_SUFFIXES = "Model|Encoder|Decoder|ForConditionalGeneration"
+# Regexes that match TF/Flax/PT model names.
+_re_tf_models = re.compile(r"TF(.*)(?:Model|Encoder|Decoder|ForConditionalGeneration)")
+_re_flax_models = re.compile(r"Flax(.*)(?:Model|Encoder|Decoder|ForConditionalGeneration)")
+# Will match any TF or Flax model too so need to be in an else branch afterthe two previous regexes.
+_re_pt_models = re.compile(r"(.*)(?:Model|Encoder|Decoder|ForConditionalGeneration)")
+
+
+# This is to make sure the diffusers module imported is the one in the repo.
+spec = importlib.util.spec_from_file_location(
+ "diffusers",
+ os.path.join(TRANSFORMERS_PATH, "__init__.py"),
+ submodule_search_locations=[TRANSFORMERS_PATH],
+)
+diffusers_module = spec.loader.load_module()
+
+
+# Thanks to https://stackoverflow.com/questions/29916065/how-to-do-camelcase-split-in-python
+def camel_case_split(identifier):
+ """Split a camelcased `identifier` into words."""
+ matches = re.finditer(".+?(?:(?<=[a-z])(?=[A-Z])|(?<=[A-Z])(?=[A-Z][a-z])|$)", identifier)
+ return [m.group(0) for m in matches]
+
+
+def _center_text(text, width):
+ text_length = 2 if text == "✅" or text == "❌" else len(text)
+ left_indent = (width - text_length) // 2
+ right_indent = width - text_length - left_indent
+ return " " * left_indent + text + " " * right_indent
+
+
+def get_model_table_from_auto_modules():
+ """Generates an up-to-date model table from the content of the auto modules."""
+ # Dictionary model names to config.
+ config_mapping_names = diffusers_module.models.auto.configuration_auto.CONFIG_MAPPING_NAMES
+ model_name_to_config = {
+ name: config_mapping_names[code]
+ for code, name in diffusers_module.MODEL_NAMES_MAPPING.items()
+ if code in config_mapping_names
+ }
+ model_name_to_prefix = {name: config.replace("ConfigMixin", "") for name, config in model_name_to_config.items()}
+
+ # Dictionaries flagging if each model prefix has a slow/fast tokenizer, backend in PT/TF/Flax.
+ slow_tokenizers = collections.defaultdict(bool)
+ fast_tokenizers = collections.defaultdict(bool)
+ pt_models = collections.defaultdict(bool)
+ tf_models = collections.defaultdict(bool)
+ flax_models = collections.defaultdict(bool)
+
+ # Let's lookup through all diffusers object (once).
+ for attr_name in dir(diffusers_module):
+ lookup_dict = None
+ if attr_name.endswith("Tokenizer"):
+ lookup_dict = slow_tokenizers
+ attr_name = attr_name[:-9]
+ elif attr_name.endswith("TokenizerFast"):
+ lookup_dict = fast_tokenizers
+ attr_name = attr_name[:-13]
+ elif _re_tf_models.match(attr_name) is not None:
+ lookup_dict = tf_models
+ attr_name = _re_tf_models.match(attr_name).groups()[0]
+ elif _re_flax_models.match(attr_name) is not None:
+ lookup_dict = flax_models
+ attr_name = _re_flax_models.match(attr_name).groups()[0]
+ elif _re_pt_models.match(attr_name) is not None:
+ lookup_dict = pt_models
+ attr_name = _re_pt_models.match(attr_name).groups()[0]
+
+ if lookup_dict is not None:
+ while len(attr_name) > 0:
+ if attr_name in model_name_to_prefix.values():
+ lookup_dict[attr_name] = True
+ break
+ # Try again after removing the last word in the name
+ attr_name = "".join(camel_case_split(attr_name)[:-1])
+
+ # Let's build that table!
+ model_names = list(model_name_to_config.keys())
+ model_names.sort(key=str.lower)
+ columns = ["Model", "Tokenizer slow", "Tokenizer fast", "PyTorch support", "TensorFlow support", "Flax Support"]
+ # We'll need widths to properly display everything in the center (+2 is to leave one extra space on each side).
+ widths = [len(c) + 2 for c in columns]
+ widths[0] = max([len(name) for name in model_names]) + 2
+
+ # Build the table per se
+ table = "|" + "|".join([_center_text(c, w) for c, w in zip(columns, widths)]) + "|\n"
+ # Use ":-----:" format to center-aligned table cell texts
+ table += "|" + "|".join([":" + "-" * (w - 2) + ":" for w in widths]) + "|\n"
+
+ check = {True: "✅", False: "❌"}
+ for name in model_names:
+ prefix = model_name_to_prefix[name]
+ line = [
+ name,
+ check[slow_tokenizers[prefix]],
+ check[fast_tokenizers[prefix]],
+ check[pt_models[prefix]],
+ check[tf_models[prefix]],
+ check[flax_models[prefix]],
+ ]
+ table += "|" + "|".join([_center_text(l, w) for l, w in zip(line, widths)]) + "|\n"
+ return table
+
+
+def check_model_table(overwrite=False):
+ """Check the model table in the index.rst is consistent with the state of the lib and maybe `overwrite`."""
+ current_table, start_index, end_index, lines = _find_text_in_file(
+ filename=os.path.join(PATH_TO_DOCS, "index.md"),
+ start_prompt="",
+ )
+ new_table = get_model_table_from_auto_modules()
+
+ if current_table != new_table:
+ if overwrite:
+ with open(os.path.join(PATH_TO_DOCS, "index.md"), "w", encoding="utf-8", newline="\n") as f:
+ f.writelines(lines[:start_index] + [new_table] + lines[end_index:])
+ else:
+ raise ValueError(
+ "The model table in the `index.md` has not been updated. Run `make fix-copies` to fix this."
+ )
+
+
+if __name__ == "__main__":
+ parser = argparse.ArgumentParser()
+ parser.add_argument("--fix_and_overwrite", action="store_true", help="Whether to fix inconsistencies.")
+ args = parser.parse_args()
+
+ check_model_table(args.fix_and_overwrite)
diff --git a/pythonProject/diffusers-main/utils/consolidated_test_report.py b/pythonProject/diffusers-main/utils/consolidated_test_report.py
new file mode 100644
index 0000000000000000000000000000000000000000..134fecf721e44ab50c16d8ed3469c94c74ffd408
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/consolidated_test_report.py
@@ -0,0 +1,789 @@
+#!/usr/bin/env python
+import argparse
+import glob
+import os
+import re
+from datetime import date, datetime
+
+from slack_sdk import WebClient
+from tabulate import tabulate
+
+
+MAX_LEN_MESSAGE = 3001 # slack endpoint has a limit of 3001 characters
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--slack_channel_name", default="diffusers-ci-nightly")
+parser.add_argument(
+ "--reports_dir",
+ default="reports",
+ help="Directory containing test reports (will search recursively in all subdirectories)",
+)
+parser.add_argument("--output_file", default=None, help="Path to save the consolidated report (markdown format)")
+
+
+def parse_stats_file(file_path):
+ """Parse a stats file to extract test statistics."""
+ try:
+ with open(file_path, "r") as f:
+ content = f.read()
+
+ # Extract the numbers using regex
+ tests_pattern = r"collected (\d+) items"
+ passed_pattern = r"(\d+) passed"
+ failed_pattern = r"(\d+) failed"
+ skipped_pattern = r"(\d+) skipped"
+ xpassed_pattern = r"(\d+) xpassed"
+
+ tests_match = re.search(tests_pattern, content)
+ passed_match = re.search(passed_pattern, content)
+ failed_match = re.search(failed_pattern, content)
+ skipped_match = re.search(skipped_pattern, content)
+ xpassed_match = re.search(xpassed_pattern, content)
+
+ passed = int(passed_match.group(1)) if passed_match else 0
+ failed = int(failed_match.group(1)) if failed_match else 0
+ skipped = int(skipped_match.group(1)) if skipped_match else 0
+ xpassed = int(xpassed_match.group(1)) if xpassed_match else 0
+
+ # If tests_match exists, use it, otherwise calculate from passed/failed/skipped
+ if tests_match:
+ tests = int(tests_match.group(1))
+ else:
+ tests = passed + failed + skipped + xpassed
+
+ # Extract timing information if available
+ timing_pattern = r"slowest \d+ test durations[\s\S]*?\n([\s\S]*?)={70}"
+ timing_match = re.search(timing_pattern, content, re.MULTILINE)
+ slowest_tests = []
+
+ if timing_match:
+ timing_text = timing_match.group(1).strip()
+ test_timing_lines = timing_text.split("\n")
+ for line in test_timing_lines:
+ if line.strip():
+ # Format is typically: 10.37s call tests/path/to/test.py::TestClass::test_method
+ parts = line.strip().split()
+ if len(parts) >= 3:
+ time_str = parts[0]
+ test_path = " ".join(parts[2:])
+
+ # Skip entries with "< 0.05 secs were omitted" or similar
+ if "secs were omitted" in test_path:
+ continue
+
+ try:
+ time_seconds = float(time_str.rstrip("s"))
+ slowest_tests.append({"test": test_path, "duration": time_seconds})
+ except ValueError:
+ pass
+
+ return {
+ "tests": tests,
+ "passed": passed,
+ "failed": failed,
+ "skipped": skipped,
+ "slowest_tests": slowest_tests,
+ }
+ except Exception as e:
+ print(f"Error parsing {file_path}: {e}")
+ return {"tests": 0, "passed": 0, "failed": 0, "skipped": 0, "slowest_tests": []}
+
+
+def parse_durations_file(file_path):
+ """Parse a durations file to extract test timing information."""
+ slowest_tests = []
+ try:
+ durations_file = file_path.replace("_stats.txt", "_durations.txt")
+ if os.path.exists(durations_file):
+ with open(durations_file, "r") as f:
+ content = f.read()
+
+ # Skip the header line
+ for line in content.split("\n")[1:]:
+ if line.strip():
+ # Format is typically: 10.37s call tests/path/to/test.py::TestClass::test_method
+ parts = line.strip().split()
+ if len(parts) >= 3:
+ time_str = parts[0]
+ test_path = " ".join(parts[2:])
+
+ # Skip entries with "< 0.05 secs were omitted" or similar
+ if "secs were omitted" in test_path:
+ continue
+
+ try:
+ time_seconds = float(time_str.rstrip("s"))
+ slowest_tests.append({"test": test_path, "duration": time_seconds})
+ except ValueError:
+ # If time_str is not a valid float, it might be a different format
+ # For example, some pytest formats show "< 0.05s" or similar
+ if test_path.startswith("<") and "secs were omitted" in test_path:
+ # Extract the time value from test_path if it's in the format "< 0.05 secs were omitted"
+ try:
+ # This handles entries where the time is in the test_path itself
+ dur_match = re.search(r"(\d+(?:\.\d+)?)", test_path)
+ if dur_match:
+ time_seconds = float(dur_match.group(1))
+ slowest_tests.append({"test": test_path, "duration": time_seconds})
+ except ValueError:
+ pass
+ except Exception as e:
+ print(f"Error parsing durations file {file_path.replace('_stats.txt', '_durations.txt')}: {e}")
+
+ return slowest_tests
+
+
+def parse_failures_file(file_path):
+ """Parse a failures file to extract failed test details."""
+ failures = []
+ try:
+ with open(file_path, "r") as f:
+ content = f.read()
+
+ # We don't need the base file name anymore as we're getting test paths from summary
+
+ # Check if it's a short stack format
+ if "============================= FAILURES SHORT STACK =============================" in content:
+ # First, look for pytest-style failure headers with underscores and clean them up
+ test_headers = re.findall(r"_{5,}\s+([^_\n]+?)\s+_{5,}", content)
+
+ for test_name in test_headers:
+ test_name = test_name.strip()
+ # Make sure it's a valid test name (contains a dot and doesn't look like a number)
+ if "." in test_name and not test_name.replace(".", "").isdigit():
+ # For test names missing the full path, check if we can reconstruct it from failures_line.txt
+ # This is a best effort - we won't always have the line file available
+ if not test_name.endswith(".py") and "::" not in test_name and "/" not in test_name:
+ # Try to look for a corresponding line file
+ line_file = file_path.replace("_failures_short.txt", "_failures_line.txt")
+ if os.path.exists(line_file):
+ try:
+ with open(line_file, "r") as lf:
+ line_content = lf.read()
+ # Look for test name in line file which might have the full path
+ path_match = re.search(
+ r"(tests/[\w/]+\.py::[^:]+::" + test_name.split(".")[-1] + ")",
+ line_content,
+ )
+ if path_match:
+ test_name = path_match.group(1)
+ except Exception:
+ pass # If we can't read the line file, just use what we have
+
+ failures.append(
+ {
+ "test": test_name,
+ "error": "Error occurred",
+ "original_test_name": test_name, # Keep original for reference
+ }
+ )
+
+ # If we didn't find any pytest-style headers, try other formats
+ if not failures:
+ # Look for test names at the beginning of the file (in first few lines)
+ first_lines = content.split("\n")[:20] # Look at first 20 lines
+ for line in first_lines:
+ # Look for test names in various formats
+ # Format: tests/file.py::TestClass::test_method
+ path_match = re.search(r"(tests/[\w/]+\.py::[\w\.]+::\w+)", line)
+ # Format: TestClass.test_method
+ class_match = re.search(r"([A-Za-z][A-Za-z0-9_]+\.[A-Za-z][A-Za-z0-9_]+)", line)
+
+ if path_match:
+ test_name = path_match.group(1)
+ failures.append(
+ {"test": test_name, "error": "Error occurred", "original_test_name": test_name}
+ )
+ break # Found a full path, stop looking
+ elif class_match and "test" in line.lower():
+ test_name = class_match.group(1)
+ # Make sure it's likely a test name (contains test in method name)
+ if "test" in test_name.lower():
+ failures.append(
+ {"test": test_name, "error": "Error occurred", "original_test_name": test_name}
+ )
+ else:
+ # Standard format - try to extract from standard pytest output
+ failure_blocks = re.split(r"={70}", content)
+
+ for block in failure_blocks:
+ if not block.strip():
+ continue
+
+ # Look for test paths in the format: path/to/test.py::TestClass::test_method
+ path_matches = re.findall(r"([\w/]+\.py::[\w\.]+::\w+)", block)
+ if path_matches:
+ for test_name in path_matches:
+ failures.append(
+ {"test": test_name, "error": "Error occurred", "original_test_name": test_name}
+ )
+ else:
+ # Try alternative format: TestClass.test_method
+ class_matches = re.findall(r"([A-Za-z][A-Za-z0-9_]+\.[A-Za-z][A-Za-z0-9_]+)", block)
+ for test_name in class_matches:
+ # Filter out things that don't look like test names
+ if (
+ not test_name.startswith(("e.g", "i.e", "etc."))
+ and not test_name.isdigit()
+ and "test" in test_name.lower()
+ ):
+ failures.append(
+ {"test": test_name, "error": "Error occurred", "original_test_name": test_name}
+ )
+
+ except Exception as e:
+ print(f"Error parsing failures in {file_path}: {e}")
+
+ return failures
+
+
+def consolidate_reports(reports_dir):
+ """Consolidate test reports from multiple test runs, including from subdirectories."""
+ # Get all stats files, including those in subdirectories
+ stats_files = glob.glob(f"{reports_dir}/**/*_stats.txt", recursive=True)
+
+ results = {}
+ total_stats = {"tests": 0, "passed": 0, "failed": 0, "skipped": 0}
+
+ # Collect all slow tests across all test suites
+ all_slow_tests = []
+
+ # Process each stats file and its corresponding failures file
+ for stats_file in stats_files:
+ # Extract test suite name from filename (e.g., tests_pipeline_allegro_cuda_stats.txt -> pipeline_allegro_cuda)
+ base_name = os.path.basename(stats_file).replace("_stats.txt", "")
+
+ # Include parent directory in suite name if it's in a subdirectory
+ rel_path = os.path.relpath(os.path.dirname(stats_file), reports_dir)
+ if rel_path and rel_path != ".":
+ # Remove 'test_reports' suffix from directory name if present
+ dir_name = os.path.basename(rel_path)
+ if dir_name.endswith("_test_reports"):
+ dir_name = dir_name[:-13] # Remove '_test_reports' suffix
+ base_name = f"{dir_name}/{base_name}"
+
+ # Parse stats
+ stats = parse_stats_file(stats_file)
+
+ # If no slowest tests found in stats file, try the durations file directly
+ if not stats.get("slowest_tests"):
+ stats["slowest_tests"] = parse_durations_file(stats_file)
+
+ # Update total stats
+ for key in ["tests", "passed", "failed", "skipped"]:
+ total_stats[key] += stats[key]
+
+ # Collect slowest tests with their suite name
+ for slow_test in stats.get("slowest_tests", []):
+ all_slow_tests.append({"test": slow_test["test"], "duration": slow_test["duration"], "suite": base_name})
+
+ # Parse failures if there are any
+ failures = []
+ if stats["failed"] > 0:
+ # First try to get test paths from summary_short.txt which has the best format
+ summary_file = stats_file.replace("_stats.txt", "_summary_short.txt")
+ if os.path.exists(summary_file):
+ try:
+ with open(summary_file, "r") as f:
+ content = f.read()
+ # Look for full lines with test path and error message: "FAILED test_path - error_msg"
+ failed_test_lines = re.findall(
+ r"FAILED\s+(tests/[\w/]+\.py::[A-Za-z0-9_\.]+::[A-Za-z0-9_]+)(?:\s+-\s+(.+))?", content
+ )
+
+ if failed_test_lines:
+ for match in failed_test_lines:
+ test_path = match[0]
+ error_msg = match[1] if len(match) > 1 and match[1] else "No error message"
+
+ failures.append({"test": test_path, "error": error_msg})
+ except Exception as e:
+ print(f"Error parsing summary file: {e}")
+
+ # If no failures found in summary, try other failure files
+ if not failures:
+ failure_patterns = ["_failures_short.txt", "_failures.txt", "_failures_line.txt", "_failures_long.txt"]
+
+ for pattern in failure_patterns:
+ failures_file = stats_file.replace("_stats.txt", pattern)
+ if os.path.exists(failures_file):
+ failures = parse_failures_file(failures_file)
+ if failures:
+ break
+
+ # No debug output needed
+
+ # Store results for this test suite
+ results[base_name] = {"stats": stats, "failures": failures}
+
+ # Filter out entries with "secs were omitted"
+ filtered_slow_tests = [test for test in all_slow_tests if "secs were omitted" not in test["test"]]
+
+ # Sort all slow tests by duration (descending)
+ filtered_slow_tests.sort(key=lambda x: x["duration"], reverse=True)
+
+ # Get the number of slowest tests to show from environment variable or default to 10
+ num_slowest_tests = int(os.environ.get("SHOW_SLOWEST_TESTS", "10"))
+ top_slowest_tests = filtered_slow_tests[:num_slowest_tests] if filtered_slow_tests else []
+
+ # Calculate additional duration statistics
+ total_duration = sum(test["duration"] for test in all_slow_tests)
+
+ # Calculate duration per suite
+ suite_durations = {}
+ for test in all_slow_tests:
+ suite_name = test["suite"]
+ if suite_name not in suite_durations:
+ suite_durations[suite_name] = 0
+ suite_durations[suite_name] += test["duration"]
+
+ # Removed duration categories
+
+ return {
+ "total_stats": total_stats,
+ "test_suites": results,
+ "slowest_tests": top_slowest_tests,
+ "duration_stats": {"total_duration": total_duration, "suite_durations": suite_durations},
+ }
+
+
+def generate_report(consolidated_data):
+ """Generate a comprehensive markdown report from consolidated data."""
+ report = []
+
+ # Add report header
+ report.append("# Diffusers Nightly Test Report")
+ report.append(f"Generated on: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}\n")
+
+ # Removed comparison section
+
+ # Add summary section
+ total = consolidated_data["total_stats"]
+ report.append("## Summary")
+
+ # Get duration stats if available
+ duration_stats = consolidated_data.get("duration_stats", {})
+ total_duration = duration_stats.get("total_duration", 0)
+
+ summary_table = [
+ ["Total Tests", total["tests"]],
+ ["Passed", total["passed"]],
+ ["Failed", total["failed"]],
+ ["Skipped", total["skipped"]],
+ ["Success Rate", f"{(total['passed'] / total['tests'] * 100):.2f}%" if total["tests"] > 0 else "N/A"],
+ ["Total Duration", f"{total_duration:.2f}s" if total_duration else "N/A"],
+ ]
+
+ report.append(tabulate(summary_table, tablefmt="pipe"))
+ report.append("")
+
+ # Removed duration distribution section
+
+ # Add test suites summary
+ report.append("## Test Suites")
+
+ # Include duration in test suites table if available
+ suite_durations = consolidated_data.get("duration_stats", {}).get("suite_durations", {})
+
+ if suite_durations:
+ suites_table = [["Test Suite", "Tests", "Passed", "Failed", "Skipped", "Success Rate", "Duration (s)"]]
+ else:
+ suites_table = [["Test Suite", "Tests", "Passed", "Failed", "Skipped", "Success Rate"]]
+
+ # Sort test suites by success rate (ascending - least successful first)
+ sorted_suites = sorted(
+ consolidated_data["test_suites"].items(),
+ key=lambda x: (x[1]["stats"]["passed"] / x[1]["stats"]["tests"] * 100) if x[1]["stats"]["tests"] > 0 else 0,
+ reverse=False,
+ )
+
+ for suite_name, suite_data in sorted_suites:
+ stats = suite_data["stats"]
+ success_rate = f"{(stats['passed'] / stats['tests'] * 100):.2f}%" if stats["tests"] > 0 else "N/A"
+
+ if suite_durations:
+ duration = suite_durations.get(suite_name, 0)
+ suites_table.append(
+ [
+ suite_name,
+ stats["tests"],
+ stats["passed"],
+ stats["failed"],
+ stats["skipped"],
+ success_rate,
+ f"{duration:.2f}",
+ ]
+ )
+ else:
+ suites_table.append(
+ [suite_name, stats["tests"], stats["passed"], stats["failed"], stats["skipped"], success_rate]
+ )
+
+ report.append(tabulate(suites_table, headers="firstrow", tablefmt="pipe"))
+ report.append("")
+
+ # Add slowest tests section
+ slowest_tests = consolidated_data.get("slowest_tests", [])
+ if slowest_tests:
+ report.append("## Slowest Tests")
+
+ slowest_table = [["Rank", "Test", "Duration (s)", "Test Suite"]]
+ for i, test in enumerate(slowest_tests, 1):
+ # Skip entries that don't contain actual test names
+ if "< 0.05 secs were omitted" in test["test"]:
+ continue
+ slowest_table.append([i, test["test"], f"{test['duration']:.2f}", test["suite"]])
+
+ report.append(tabulate(slowest_table, headers="firstrow", tablefmt="pipe"))
+ report.append("")
+
+ # Add failures section if there are any
+ failed_suites = [s for s in sorted_suites if s[1]["stats"]["failed"] > 0]
+
+ if failed_suites:
+ report.append("## Failures")
+
+ # Group failures by module for cleaner organization
+ failures_by_module = {}
+
+ for suite_name, suite_data in failed_suites:
+ # Extract failures data for this suite
+ for failure in suite_data.get("failures", []):
+ test_name = failure["test"]
+
+ # If test name doesn't look like a full path, try to reconstruct it
+ if not ("/" in test_name or "::" in test_name) and "." in test_name:
+ # For simple 'TestClass.test_method' format, try to get full path from suite name
+ # Form: tests__cuda -> tests//test_.py::TestClass::test_method
+ if suite_name.startswith("tests_") and "_cuda" in suite_name:
+ # Extract component name from suite
+ component = suite_name.replace("tests_", "").replace("_cuda", "")
+ if "." in test_name:
+ class_name, method_name = test_name.split(".", 1)
+ possible_path = f"tests/{component}/test_{component}.py::{class_name}::{method_name}"
+ # Use this constructed path if it seems reasonable
+ if "test_" in method_name:
+ test_name = possible_path
+
+ # Extract module name from test name
+ if "::" in test_name:
+ # For path/file.py::TestClass::test_method format
+ parts = test_name.split("::")
+ module_name = parts[-2] if len(parts) >= 2 else "Other" # TestClass
+ elif "." in test_name:
+ # For TestClass.test_method format
+ parts = test_name.split(".")
+ module_name = parts[0] # TestClass
+ else:
+ module_name = "Other"
+
+ # Skip module names that don't look like class/module names
+ if (
+ module_name.startswith(("e.g", "i.e", "etc"))
+ or module_name.replace(".", "").isdigit()
+ or len(module_name) < 3
+ ):
+ module_name = "Other"
+
+ # Add to the module group
+ if module_name not in failures_by_module:
+ failures_by_module[module_name] = []
+
+ # Prepend the suite name if the test name doesn't already have a full path
+ if "/" not in test_name and suite_name not in test_name:
+ full_test_name = f"{suite_name}::{test_name}"
+ else:
+ full_test_name = test_name
+
+ # Add this failure to the module group
+ failures_by_module[module_name].append(
+ {"test": full_test_name, "original_test": test_name, "error": failure["error"]}
+ )
+
+ # Create a list of failing tests for each module
+ if failures_by_module:
+ for module_name, failures in sorted(failures_by_module.items()):
+ report.append(f"### {module_name}")
+
+ # Put all failed tests in a single code block
+ report.append("```")
+ for failure in failures:
+ # Show test path and error message if available
+ if failure.get("error") and failure["error"] != "No error message":
+ report.append(f"{failure['test']} - {failure['error']}")
+ else:
+ report.append(failure["test"])
+ report.append("```")
+
+ report.append("") # Add space between modules
+ else:
+ report.append("*No detailed failure information available*")
+ report.append("")
+
+ return "\n".join(report)
+
+
+def create_test_groups_table(test_groups, total_tests, total_success_rate):
+ """Create a table-like format for test groups showing total tests and success rate."""
+ if not test_groups:
+ return None
+
+ # Sort by total test count (descending)
+ sorted_groups = sorted(test_groups.items(), key=lambda x: x[1]["total"], reverse=True)
+
+ # Create table lines
+ table_lines = ["```"]
+ table_lines.append("Test Results Summary")
+ table_lines.append("-------------------")
+ table_lines.append(f"Total Tests: {total_tests:,}")
+ table_lines.append(f"Success Rate: {total_success_rate}")
+ table_lines.append("")
+ table_lines.append("Category | Total Tests | Failed | Success Rate")
+ table_lines.append("------------------- | ----------- | ------ | ------------")
+
+ # Add rows
+ for category, stats in sorted_groups:
+ # Pad category name to fixed width (19 chars)
+ padded_cat = category[:19].ljust(19) # Truncate if too long
+ # Right-align counts
+ padded_total = str(stats["total"]).rjust(11)
+ padded_failed = str(stats["failed"]).rjust(6)
+ # Calculate and format success rate
+ if stats["total"] > 0:
+ cat_success_rate = f"{((stats['total'] - stats['failed']) / stats['total'] * 100):.1f}%"
+ else:
+ cat_success_rate = "N/A"
+ padded_rate = cat_success_rate.rjust(12)
+ table_lines.append(f"{padded_cat} | {padded_total} | {padded_failed} | {padded_rate}")
+
+ table_lines.append("```")
+
+ total_failures = sum(stats["failed"] for stats in test_groups.values())
+ return (
+ f"*Test Groups Summary ({total_failures} {'failure' if total_failures == 1 else 'failures'}):*\n"
+ + "\n".join(table_lines)
+ )
+
+
+def create_slack_payload(consolidated_data):
+ """Create a concise Slack message payload from consolidated data."""
+ total = consolidated_data["total_stats"]
+ success_rate = f"{(total['passed'] / total['tests'] * 100):.2f}%" if total["tests"] > 0 else "N/A"
+
+ # Determine emoji based on success rate
+ if total["failed"] == 0:
+ emoji = "✅"
+ elif total["failed"] / total["tests"] < 0.1:
+ emoji = "⚠️"
+ else:
+ emoji = "❌"
+
+ # Create a more compact summary section
+ summary = f"{emoji} *Diffusers Nightly Tests:* {success_rate} success ({total['passed']}/{total['tests']} tests"
+ if total["skipped"] > 0:
+ summary += f", {total['skipped']} skipped"
+ summary += ")"
+
+ # Create the test suites table in markdown format
+ # Build the markdown table with proper alignment
+ table_lines = []
+ table_lines.append("```")
+
+ # Sort test suites by success rate (ascending - least successful first)
+ sorted_suites = sorted(
+ consolidated_data["test_suites"].items(),
+ key=lambda x: (x[1]["stats"]["passed"] / x[1]["stats"]["tests"] * 100) if x[1]["stats"]["tests"] > 0 else 0,
+ reverse=False,
+ )
+
+ # Calculate max widths for proper alignment
+ max_suite_name_len = max(len(suite_name) for suite_name, _ in sorted_suites) if sorted_suites else 10
+ max_suite_name_len = max(max_suite_name_len, len("Test Suite")) # Ensure header fits
+
+ # Create header with proper spacing (only Tests, Failed, Success Rate)
+ header = f"| {'Test Suite'.ljust(max_suite_name_len)} | {'Tests'.rjust(6)} | {'Failed'.rjust(6)} | {'Success Rate'.ljust(12)} |"
+ separator = f"|:{'-' * max_suite_name_len}|{'-' * 7}:|{'-' * 7}:|:{'-' * 11}|"
+
+ table_lines.append(header)
+ table_lines.append(separator)
+
+ # Add data rows with proper alignment
+ for suite_name, suite_data in sorted_suites:
+ stats = suite_data["stats"]
+ suite_success_rate = f"{(stats['passed'] / stats['tests'] * 100):.2f}%" if stats["tests"] > 0 else "N/A"
+
+ row = f"| {suite_name.ljust(max_suite_name_len)} | {str(stats['tests']).rjust(6)} | {str(stats['failed']).rjust(6)} | {suite_success_rate.ljust(12)} |"
+
+ table_lines.append(row)
+
+ table_lines.append("```")
+
+ # Create the Slack payload with character limit enforcement
+ payload = [
+ {"type": "section", "text": {"type": "mrkdwn", "text": summary}},
+ {"type": "section", "text": {"type": "mrkdwn", "text": "\n".join(table_lines)}},
+ ]
+
+ # Add action button
+ if os.environ.get("GITHUB_RUN_ID"):
+ run_id = os.environ["GITHUB_RUN_ID"]
+ payload.append(
+ {
+ "type": "section",
+ "text": {
+ "type": "mrkdwn",
+ "text": f"**",
+ },
+ }
+ )
+
+ # Add date in more compact form
+ payload.append(
+ {
+ "type": "context",
+ "elements": [
+ {
+ "type": "plain_text",
+ "text": f"Results for {date.today()}",
+ },
+ ],
+ }
+ )
+
+ # Enforce 3001 character limit
+ payload_text = str(payload)
+ if len(payload_text) > MAX_LEN_MESSAGE:
+ # Truncate table if payload is too long
+ # Remove rows from the bottom until under limit
+ original_table_lines = table_lines[:]
+ while len(str(payload)) > MAX_LEN_MESSAGE and len(table_lines) > 3: # Keep at least header and separator
+ # Remove the last data row (but keep ``` at the end)
+ table_lines.pop(-2) # Remove second to last (last is the closing ```)
+
+ # Recreate payload with truncated table
+ payload[1] = {"type": "section", "text": {"type": "mrkdwn", "text": "\n".join(table_lines)}}
+
+ # Add note if we had to truncate
+ if len(table_lines) < len(original_table_lines):
+ truncated_count = len(original_table_lines) - len(table_lines)
+ table_lines.insert(-1, f"... {truncated_count} more test suites (truncated due to message limit)")
+ payload[1] = {"type": "section", "text": {"type": "mrkdwn", "text": "\n".join(table_lines)}}
+
+ return payload
+
+
+def create_failed_tests_by_suite_ordered(consolidated_data):
+ """Group failed tests by test suite, ordered by success rate (ascending)."""
+ # Sort test suites by success rate (ascending - least successful first)
+ sorted_suites = sorted(
+ consolidated_data["test_suites"].items(),
+ key=lambda x: (x[1]["stats"]["passed"] / x[1]["stats"]["tests"] * 100) if x[1]["stats"]["tests"] > 0 else 0,
+ reverse=False,
+ )
+
+ failed_suite_tests = []
+
+ # Process suites in order of success rate
+ for suite_name, suite_data in sorted_suites:
+ if suite_data["stats"]["failed"] > 0:
+ suite_failures = []
+
+ for failure in suite_data.get("failures", []):
+ test_name = failure["test"]
+
+ # Try to reconstruct full path if partial
+ if "::" in test_name and "/" in test_name:
+ full_test_name = test_name
+ elif "::" in test_name or "." in test_name:
+ if "/" not in test_name and suite_name not in test_name:
+ full_test_name = f"{suite_name}::{test_name}"
+ else:
+ full_test_name = test_name
+ else:
+ full_test_name = f"{suite_name}::{test_name}"
+
+ suite_failures.append(full_test_name)
+
+ # Sort and deduplicate tests within the suite
+ suite_failures = sorted(set(suite_failures))
+
+ if suite_failures:
+ failed_suite_tests.append(
+ {
+ "suite_name": suite_name,
+ "tests": suite_failures,
+ "success_rate": (suite_data["stats"]["passed"] / suite_data["stats"]["tests"] * 100)
+ if suite_data["stats"]["tests"] > 0
+ else 0,
+ }
+ )
+
+ return failed_suite_tests
+
+
+def main(args):
+ # Make sure reports directory exists
+ if not os.path.isdir(args.reports_dir):
+ print(f"Error: Reports directory '{args.reports_dir}' does not exist.")
+ return
+
+ # Consolidate reports
+ consolidated_data = consolidate_reports(args.reports_dir)
+
+ # Check if we found any test results
+ if consolidated_data["total_stats"]["tests"] == 0:
+ print(f"Warning: No test results found in '{args.reports_dir}' or its subdirectories.")
+
+ # Generate markdown report
+ report = generate_report(consolidated_data)
+
+ # Save report to file if specified
+ if args.output_file:
+ # Create parent directories if they don't exist
+ output_dir = os.path.dirname(args.output_file)
+ if output_dir and not os.path.exists(output_dir):
+ os.makedirs(output_dir)
+
+ with open(args.output_file, "w") as f:
+ f.write(report)
+
+ # Only print the report when saving to file
+ print(report)
+
+ # Send to Slack if token is available (optional, can be disabled)
+ slack_token = os.environ.get("SLACK_API_TOKEN")
+ if slack_token and args.slack_channel_name:
+ payload = create_slack_payload(consolidated_data)
+
+ try:
+ client = WebClient(token=slack_token)
+ # Send main message
+ response = client.chat_postMessage(channel=f"#{args.slack_channel_name}", blocks=payload)
+ print(f"Report sent to Slack channel: {args.slack_channel_name}")
+
+ # Send failed tests as separate threaded replies grouped by test suite (ordered by success rate)
+ total = consolidated_data["total_stats"]
+ if total["failed"] > 0:
+ failed_suites = create_failed_tests_by_suite_ordered(consolidated_data)
+ for suite_info in failed_suites:
+ suite_name = suite_info["suite_name"]
+ suite_tests = suite_info["tests"]
+ success_rate = suite_info["success_rate"]
+ message_text = (
+ f"**{suite_name}** (Success Rate: {success_rate:.2f}%)\n```\n"
+ + "\n".join(suite_tests)
+ + "\n```"
+ )
+ client.chat_postMessage(
+ channel=f"#{args.slack_channel_name}",
+ thread_ts=response["ts"], # Reply in thread
+ text=message_text, # Use text instead of blocks for markdown
+ )
+ print(f"Failed tests details sent as {len(failed_suites)} thread replies")
+ except Exception as e:
+ print(f"Error sending report to Slack: {e}")
+
+
+if __name__ == "__main__":
+ args = parser.parse_args()
+ main(args)
diff --git a/pythonProject/diffusers-main/utils/custom_init_isort.py b/pythonProject/diffusers-main/utils/custom_init_isort.py
new file mode 100644
index 0000000000000000000000000000000000000000..cc3bccb9bd63a0fbd28105f5d946024c1637b303
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/custom_init_isort.py
@@ -0,0 +1,330 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Utility that sorts the imports in the custom inits of Diffusers. Diffusers uses init files that delay the
+import of an object to when it's actually needed. This is to avoid the main init importing all models, which would
+make the line `import transformers` very slow when the user has all optional dependencies installed. The inits with
+delayed imports have two halves: one defining a dictionary `_import_structure` which maps modules to the name of the
+objects in each module, and one in `TYPE_CHECKING` which looks like a normal init for type-checkers. `isort` or `ruff`
+properly sort the second half which looks like traditionl imports, the goal of this script is to sort the first half.
+
+Use from the root of the repo with:
+
+```bash
+python utils/custom_init_isort.py
+```
+
+which will auto-sort the imports (used in `make style`).
+
+For a check only (as used in `make quality`) run:
+
+```bash
+python utils/custom_init_isort.py --check_only
+```
+"""
+
+import argparse
+import os
+import re
+from typing import Any, Callable, List, Optional
+
+
+# Path is defined with the intent you should run this script from the root of the repo.
+PATH_TO_TRANSFORMERS = "src/diffusers"
+
+# Pattern that looks at the indentation in a line.
+_re_indent = re.compile(r"^(\s*)\S")
+# Pattern that matches `"key":" and puts `key` in group 0.
+_re_direct_key = re.compile(r'^\s*"([^"]+)":')
+# Pattern that matches `_import_structure["key"]` and puts `key` in group 0.
+_re_indirect_key = re.compile(r'^\s*_import_structure\["([^"]+)"\]')
+# Pattern that matches `"key",` and puts `key` in group 0.
+_re_strip_line = re.compile(r'^\s*"([^"]+)",\s*$')
+# Pattern that matches any `[stuff]` and puts `stuff` in group 0.
+_re_bracket_content = re.compile(r"\[([^\]]+)\]")
+
+
+def get_indent(line: str) -> str:
+ """Returns the indent in given line (as string)."""
+ search = _re_indent.search(line)
+ return "" if search is None else search.groups()[0]
+
+
+def split_code_in_indented_blocks(
+ code: str, indent_level: str = "", start_prompt: Optional[str] = None, end_prompt: Optional[str] = None
+) -> List[str]:
+ """
+ Split some code into its indented blocks, starting at a given level.
+
+ Args:
+ code (`str`): The code to split.
+ indent_level (`str`): The indent level (as string) to use for identifying the blocks to split.
+ start_prompt (`str`, *optional*): If provided, only starts splitting at the line where this text is.
+ end_prompt (`str`, *optional*): If provided, stops splitting at a line where this text is.
+
+ Warning:
+ The text before `start_prompt` or after `end_prompt` (if provided) is not ignored, just not split. The input `code`
+ can thus be retrieved by joining the result.
+
+ Returns:
+ `List[str]`: The list of blocks.
+ """
+ # Let's split the code into lines and move to start_index.
+ index = 0
+ lines = code.split("\n")
+ if start_prompt is not None:
+ while not lines[index].startswith(start_prompt):
+ index += 1
+ blocks = ["\n".join(lines[:index])]
+ else:
+ blocks = []
+
+ # This variable contains the block treated at a given time.
+ current_block = [lines[index]]
+ index += 1
+ # We split into blocks until we get to the `end_prompt` (or the end of the file).
+ while index < len(lines) and (end_prompt is None or not lines[index].startswith(end_prompt)):
+ # We have a non-empty line with the proper indent -> start of a new block
+ if len(lines[index]) > 0 and get_indent(lines[index]) == indent_level:
+ # Store the current block in the result and rest. There are two cases: the line is part of the block (like
+ # a closing parenthesis) or not.
+ if len(current_block) > 0 and get_indent(current_block[-1]).startswith(indent_level + " "):
+ # Line is part of the current block
+ current_block.append(lines[index])
+ blocks.append("\n".join(current_block))
+ if index < len(lines) - 1:
+ current_block = [lines[index + 1]]
+ index += 1
+ else:
+ current_block = []
+ else:
+ # Line is not part of the current block
+ blocks.append("\n".join(current_block))
+ current_block = [lines[index]]
+ else:
+ # Just add the line to the current block
+ current_block.append(lines[index])
+ index += 1
+
+ # Adds current block if it's nonempty.
+ if len(current_block) > 0:
+ blocks.append("\n".join(current_block))
+
+ # Add final block after end_prompt if provided.
+ if end_prompt is not None and index < len(lines):
+ blocks.append("\n".join(lines[index:]))
+
+ return blocks
+
+
+def ignore_underscore_and_lowercase(key: Callable[[Any], str]) -> Callable[[Any], str]:
+ """
+ Wraps a key function (as used in a sort) to lowercase and ignore underscores.
+ """
+
+ def _inner(x):
+ return key(x).lower().replace("_", "")
+
+ return _inner
+
+
+def sort_objects(objects: List[Any], key: Optional[Callable[[Any], str]] = None) -> List[Any]:
+ """
+ Sort a list of objects following the rules of isort (all uppercased first, camel-cased second and lower-cased
+ last).
+
+ Args:
+ objects (`List[Any]`):
+ The list of objects to sort.
+ key (`Callable[[Any], str]`, *optional*):
+ A function taking an object as input and returning a string, used to sort them by alphabetical order.
+ If not provided, will default to noop (so a `key` must be provided if the `objects` are not of type string).
+
+ Returns:
+ `List[Any]`: The sorted list with the same elements as in the inputs
+ """
+
+ # If no key is provided, we use a noop.
+ def noop(x):
+ return x
+
+ if key is None:
+ key = noop
+ # Constants are all uppercase, they go first.
+ constants = [obj for obj in objects if key(obj).isupper()]
+ # Classes are not all uppercase but start with a capital, they go second.
+ classes = [obj for obj in objects if key(obj)[0].isupper() and not key(obj).isupper()]
+ # Functions begin with a lowercase, they go last.
+ functions = [obj for obj in objects if not key(obj)[0].isupper()]
+
+ # Then we sort each group.
+ key1 = ignore_underscore_and_lowercase(key)
+ return sorted(constants, key=key1) + sorted(classes, key=key1) + sorted(functions, key=key1)
+
+
+def sort_objects_in_import(import_statement: str) -> str:
+ """
+ Sorts the imports in a single import statement.
+
+ Args:
+ import_statement (`str`): The import statement in which to sort the imports.
+
+ Returns:
+ `str`: The same as the input, but with objects properly sorted.
+ """
+
+ # This inner function sort imports between [ ].
+ def _replace(match):
+ imports = match.groups()[0]
+ # If there is one import only, nothing to do.
+ if "," not in imports:
+ return f"[{imports}]"
+ keys = [part.strip().replace('"', "") for part in imports.split(",")]
+ # We will have a final empty element if the line finished with a comma.
+ if len(keys[-1]) == 0:
+ keys = keys[:-1]
+ return "[" + ", ".join([f'"{k}"' for k in sort_objects(keys)]) + "]"
+
+ lines = import_statement.split("\n")
+ if len(lines) > 3:
+ # Here we have to sort internal imports that are on several lines (one per name):
+ # key: [
+ # "object1",
+ # "object2",
+ # ...
+ # ]
+
+ # We may have to ignore one or two lines on each side.
+ idx = 2 if lines[1].strip() == "[" else 1
+ keys_to_sort = [(i, _re_strip_line.search(line).groups()[0]) for i, line in enumerate(lines[idx:-idx])]
+ sorted_indices = sort_objects(keys_to_sort, key=lambda x: x[1])
+ sorted_lines = [lines[x[0] + idx] for x in sorted_indices]
+ return "\n".join(lines[:idx] + sorted_lines + lines[-idx:])
+ elif len(lines) == 3:
+ # Here we have to sort internal imports that are on one separate line:
+ # key: [
+ # "object1", "object2", ...
+ # ]
+ if _re_bracket_content.search(lines[1]) is not None:
+ lines[1] = _re_bracket_content.sub(_replace, lines[1])
+ else:
+ keys = [part.strip().replace('"', "") for part in lines[1].split(",")]
+ # We will have a final empty element if the line finished with a comma.
+ if len(keys[-1]) == 0:
+ keys = keys[:-1]
+ lines[1] = get_indent(lines[1]) + ", ".join([f'"{k}"' for k in sort_objects(keys)])
+ return "\n".join(lines)
+ else:
+ # Finally we have to deal with imports fitting on one line
+ import_statement = _re_bracket_content.sub(_replace, import_statement)
+ return import_statement
+
+
+def sort_imports(file: str, check_only: bool = True):
+ """
+ Sort the imports defined in the `_import_structure` of a given init.
+
+ Args:
+ file (`str`): The path to the init to check/fix.
+ check_only (`bool`, *optional*, defaults to `True`): Whether or not to just check (and not auto-fix) the init.
+ """
+ with open(file, encoding="utf-8") as f:
+ code = f.read()
+
+ # If the file is not a custom init, there is nothing to do.
+ if "_import_structure" not in code:
+ return
+
+ # Blocks of indent level 0
+ main_blocks = split_code_in_indented_blocks(
+ code, start_prompt="_import_structure = {", end_prompt="if TYPE_CHECKING:"
+ )
+
+ # We ignore block 0 (everything until start_prompt) and the last block (everything after end_prompt).
+ for block_idx in range(1, len(main_blocks) - 1):
+ # Check if the block contains some `_import_structure`s thingy to sort.
+ block = main_blocks[block_idx]
+ block_lines = block.split("\n")
+
+ # Get to the start of the imports.
+ line_idx = 0
+ while line_idx < len(block_lines) and "_import_structure" not in block_lines[line_idx]:
+ # Skip dummy import blocks
+ if "import dummy" in block_lines[line_idx]:
+ line_idx = len(block_lines)
+ else:
+ line_idx += 1
+ if line_idx >= len(block_lines):
+ continue
+
+ # Ignore beginning and last line: they don't contain anything.
+ internal_block_code = "\n".join(block_lines[line_idx:-1])
+ indent = get_indent(block_lines[1])
+ # Slit the internal block into blocks of indent level 1.
+ internal_blocks = split_code_in_indented_blocks(internal_block_code, indent_level=indent)
+ # We have two categories of import key: list or _import_structure[key].append/extend
+ pattern = _re_direct_key if "_import_structure = {" in block_lines[0] else _re_indirect_key
+ # Grab the keys, but there is a trap: some lines are empty or just comments.
+ keys = [(pattern.search(b).groups()[0] if pattern.search(b) is not None else None) for b in internal_blocks]
+ # We only sort the lines with a key.
+ keys_to_sort = [(i, key) for i, key in enumerate(keys) if key is not None]
+ sorted_indices = [x[0] for x in sorted(keys_to_sort, key=lambda x: x[1])]
+
+ # We reorder the blocks by leaving empty lines/comments as they were and reorder the rest.
+ count = 0
+ reordered_blocks = []
+ for i in range(len(internal_blocks)):
+ if keys[i] is None:
+ reordered_blocks.append(internal_blocks[i])
+ else:
+ block = sort_objects_in_import(internal_blocks[sorted_indices[count]])
+ reordered_blocks.append(block)
+ count += 1
+
+ # And we put our main block back together with its first and last line.
+ main_blocks[block_idx] = "\n".join(block_lines[:line_idx] + reordered_blocks + [block_lines[-1]])
+
+ if code != "\n".join(main_blocks):
+ if check_only:
+ return True
+ else:
+ print(f"Overwriting {file}.")
+ with open(file, "w", encoding="utf-8") as f:
+ f.write("\n".join(main_blocks))
+
+
+def sort_imports_in_all_inits(check_only=True):
+ """
+ Sort the imports defined in the `_import_structure` of all inits in the repo.
+
+ Args:
+ check_only (`bool`, *optional*, defaults to `True`): Whether or not to just check (and not auto-fix) the init.
+ """
+ failures = []
+ for root, _, files in os.walk(PATH_TO_TRANSFORMERS):
+ if "__init__.py" in files:
+ result = sort_imports(os.path.join(root, "__init__.py"), check_only=check_only)
+ if result:
+ failures = [os.path.join(root, "__init__.py")]
+ if len(failures) > 0:
+ raise ValueError(f"Would overwrite {len(failures)} files, run `make style`.")
+
+
+if __name__ == "__main__":
+ parser = argparse.ArgumentParser()
+ parser.add_argument("--check_only", action="store_true", help="Whether to only check or fix style.")
+ args = parser.parse_args()
+
+ sort_imports_in_all_inits(check_only=args.check_only)
diff --git a/pythonProject/diffusers-main/utils/extract_tests_from_mixin.py b/pythonProject/diffusers-main/utils/extract_tests_from_mixin.py
new file mode 100644
index 0000000000000000000000000000000000000000..c8b65b96ee16071042d93b66071367acad658d9c
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/extract_tests_from_mixin.py
@@ -0,0 +1,61 @@
+import argparse
+import inspect
+import sys
+from pathlib import Path
+from typing import List, Type
+
+
+root_dir = Path(__file__).parent.parent.absolute()
+sys.path.insert(0, str(root_dir))
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--type", type=str, default=None)
+args = parser.parse_args()
+
+
+def get_test_methods_from_class(cls: Type) -> List[str]:
+ """
+ Get all test method names from a given class.
+ Only returns methods that start with 'test_'.
+ """
+ test_methods = []
+ for name, obj in inspect.getmembers(cls):
+ if name.startswith("test_") and inspect.isfunction(obj):
+ test_methods.append(name)
+ return sorted(test_methods)
+
+
+def generate_pytest_pattern(test_methods: List[str]) -> str:
+ """Generate pytest pattern string for the -k flag."""
+ return " or ".join(test_methods)
+
+
+def generate_pattern_for_mixin(mixin_class: Type) -> str:
+ """
+ Generate pytest pattern for a specific mixin class.
+ """
+ if mixin_cls is None:
+ return ""
+ test_methods = get_test_methods_from_class(mixin_class)
+ return generate_pytest_pattern(test_methods)
+
+
+if __name__ == "__main__":
+ mixin_cls = None
+ if args.type == "pipeline":
+ from tests.pipelines.test_pipelines_common import PipelineTesterMixin
+
+ mixin_cls = PipelineTesterMixin
+
+ elif args.type == "models":
+ from tests.models.test_modeling_common import ModelTesterMixin
+
+ mixin_cls = ModelTesterMixin
+
+ elif args.type == "lora":
+ from tests.lora.utils import PeftLoraLoaderMixinTests
+
+ mixin_cls = PeftLoraLoaderMixinTests
+
+ pattern = generate_pattern_for_mixin(mixin_cls)
+ print(pattern)
diff --git a/pythonProject/diffusers-main/utils/fetch_latest_release_branch.py b/pythonProject/diffusers-main/utils/fetch_latest_release_branch.py
new file mode 100644
index 0000000000000000000000000000000000000000..5b0be6253e1b96af3883da916fc06c5522321dcf
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/fetch_latest_release_branch.py
@@ -0,0 +1,73 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+import requests
+from packaging.version import parse
+
+
+# GitHub repository details
+USER = "huggingface"
+REPO = "diffusers"
+
+
+def fetch_all_branches(user, repo):
+ branches = [] # List to store all branches
+ page = 1 # Start from first page
+ while True:
+ # Make a request to the GitHub API for the branches
+ response = requests.get(
+ f"https://api.github.com/repos/{user}/{repo}/branches",
+ params={"page": page},
+ timeout=60,
+ )
+
+ # Check if the request was successful
+ if response.status_code == 200:
+ # Add the branches from the current page to the list
+ branches.extend([branch["name"] for branch in response.json()])
+
+ # Check if there is a 'next' link for pagination
+ if "next" in response.links:
+ page += 1 # Move to the next page
+ else:
+ break # Exit loop if there is no next page
+ else:
+ print("Failed to retrieve branches:", response.status_code)
+ break
+
+ return branches
+
+
+def main():
+ # Fetch all branches
+ branches = fetch_all_branches(USER, REPO)
+
+ # Filter branches.
+ # print(f"Total branches: {len(branches)}")
+ filtered_branches = []
+ for branch in branches:
+ if branch.startswith("v") and ("-release" in branch or "-patch" in branch):
+ filtered_branches.append(branch)
+ # print(f"Filtered: {branch}")
+
+ sorted_branches = sorted(filtered_branches, key=lambda x: parse(x.split("-")[0][1:]), reverse=True)
+ latest_branch = sorted_branches[0]
+ # print(f"Latest branch: {latest_branch}")
+ return latest_branch
+
+
+if __name__ == "__main__":
+ print(main())
diff --git a/pythonProject/diffusers-main/utils/fetch_torch_cuda_pipeline_test_matrix.py b/pythonProject/diffusers-main/utils/fetch_torch_cuda_pipeline_test_matrix.py
new file mode 100644
index 0000000000000000000000000000000000000000..196f35628ac19922671724b85f80d75fc19fb436
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/fetch_torch_cuda_pipeline_test_matrix.py
@@ -0,0 +1,99 @@
+import json
+import logging
+import os
+from collections import defaultdict
+from pathlib import Path
+
+from huggingface_hub import HfApi
+
+import diffusers
+
+
+PATH_TO_REPO = Path(__file__).parent.parent.resolve()
+ALWAYS_TEST_PIPELINE_MODULES = [
+ "controlnet",
+ "controlnet_flux",
+ "controlnet_sd3",
+ "stable_diffusion",
+ "stable_diffusion_2",
+ "stable_diffusion_3",
+ "stable_diffusion_xl",
+ "ip_adapters",
+ "flux",
+]
+PIPELINE_USAGE_CUTOFF = int(os.getenv("PIPELINE_USAGE_CUTOFF", 50000))
+
+logger = logging.getLogger(__name__)
+api = HfApi()
+
+
+def filter_pipelines(usage_dict, usage_cutoff=10000):
+ output = []
+ for diffusers_object, usage in usage_dict.items():
+ if usage < usage_cutoff:
+ continue
+
+ is_diffusers_pipeline = hasattr(diffusers.pipelines, diffusers_object)
+ if not is_diffusers_pipeline:
+ continue
+
+ output.append(diffusers_object)
+
+ return output
+
+
+def fetch_pipeline_objects():
+ models = api.list_models(library="diffusers")
+ downloads = defaultdict(int)
+
+ for model in models:
+ is_counted = False
+ for tag in model.tags:
+ if tag.startswith("diffusers:"):
+ is_counted = True
+ downloads[tag[len("diffusers:") :]] += model.downloads
+
+ if not is_counted:
+ downloads["other"] += model.downloads
+
+ # Remove 0 downloads
+ downloads = {k: v for k, v in downloads.items() if v > 0}
+ pipeline_objects = filter_pipelines(downloads, PIPELINE_USAGE_CUTOFF)
+
+ return pipeline_objects
+
+
+def fetch_pipeline_modules_to_test():
+ try:
+ pipeline_objects = fetch_pipeline_objects()
+ except Exception as e:
+ logger.error(e)
+ raise RuntimeError("Unable to fetch model list from HuggingFace Hub.")
+
+ test_modules = []
+ for pipeline_name in pipeline_objects:
+ module = getattr(diffusers, pipeline_name)
+
+ test_module = module.__module__.split(".")[-2].strip()
+ test_modules.append(test_module)
+
+ return test_modules
+
+
+def main():
+ test_modules = fetch_pipeline_modules_to_test()
+ test_modules.extend(ALWAYS_TEST_PIPELINE_MODULES)
+
+ # Get unique modules
+ test_modules = sorted(set(test_modules))
+ print(json.dumps(test_modules))
+
+ save_path = f"{PATH_TO_REPO}/reports"
+ os.makedirs(save_path, exist_ok=True)
+
+ with open(f"{save_path}/test-pipelines.json", "w") as f:
+ json.dump({"pipeline_test_modules": test_modules}, f)
+
+
+if __name__ == "__main__":
+ main()
diff --git a/pythonProject/diffusers-main/utils/get_modified_files.py b/pythonProject/diffusers-main/utils/get_modified_files.py
new file mode 100644
index 0000000000000000000000000000000000000000..e392e50c12d3af270df748987be33ee6c3937901
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/get_modified_files.py
@@ -0,0 +1,34 @@
+# coding=utf-8
+# Copyright 2025 The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# this script reports modified .py files under the desired list of top-level sub-dirs passed as a list of arguments, e.g.:
+# python ./utils/get_modified_files.py utils src tests examples
+#
+# it uses git to find the forking point and which files were modified - i.e. files not under git won't be considered
+# since the output of this script is fed into Makefile commands it doesn't print a newline after the results
+
+import re
+import subprocess
+import sys
+
+
+fork_point_sha = subprocess.check_output("git merge-base main HEAD".split()).decode("utf-8")
+modified_files = subprocess.check_output(f"git diff --name-only {fork_point_sha}".split()).decode("utf-8").split()
+
+joined_dirs = "|".join(sys.argv[1:])
+regex = re.compile(rf"^({joined_dirs}).*?\.py$")
+
+relevant_modified_files = [x for x in modified_files if regex.match(x)]
+print(" ".join(relevant_modified_files), end="")
diff --git a/pythonProject/diffusers-main/utils/log_reports.py b/pythonProject/diffusers-main/utils/log_reports.py
new file mode 100644
index 0000000000000000000000000000000000000000..5575c9ba8415a73358fe48e3d0ac41b2188691d5
--- /dev/null
+++ b/pythonProject/diffusers-main/utils/log_reports.py
@@ -0,0 +1,139 @@
+import argparse
+import json
+import os
+from datetime import date
+from pathlib import Path
+
+from slack_sdk import WebClient
+from tabulate import tabulate
+
+
+MAX_LEN_MESSAGE = 2900 # slack endpoint has a limit of 3001 characters
+
+parser = argparse.ArgumentParser()
+parser.add_argument("--slack_channel_name", default="diffusers-ci-nightly")
+
+
+def main(slack_channel_name=None):
+ failed = []
+ passed = []
+
+ group_info = []
+
+ total_num_failed = 0
+ empty_file = False or len(list(Path().glob("*.log"))) == 0
+
+ total_empty_files = []
+
+ for log in Path().glob("*.log"):
+ section_num_failed = 0
+ i = 0
+ with open(log) as f:
+ for line in f:
+ line = json.loads(line)
+ i += 1
+ if line.get("nodeid", "") != "":
+ test = line["nodeid"]
+ if line.get("duration", None) is not None:
+ duration = f"{line['duration']:.4f}"
+ if line.get("outcome", "") == "failed":
+ section_num_failed += 1
+ failed.append([test, duration, log.name.split("_")[0]])
+ total_num_failed += 1
+ else:
+ passed.append([test, duration, log.name.split("_")[0]])
+ empty_file = i == 0
+ group_info.append([str(log), section_num_failed, failed])
+ total_empty_files.append(empty_file)
+ os.remove(log)
+ failed = []
+ text = (
+ "🌞 There were no failures!"
+ if not any(total_empty_files)
+ else "Something went wrong there is at least one empty file - please check GH action results."
+ )
+ no_error_payload = {
+ "type": "section",
+ "text": {
+ "type": "plain_text",
+ "text": text,
+ "emoji": True,
+ },
+ }
+
+ message = ""
+ payload = [
+ {
+ "type": "header",
+ "text": {
+ "type": "plain_text",
+ "text": "🤗 Results of the Diffusers scheduled nightly tests.",
+ },
+ },
+ ]
+ if total_num_failed > 0:
+ for i, (name, num_failed, failed_tests) in enumerate(group_info):
+ if num_failed > 0:
+ if num_failed == 1:
+ message += f"*{name}: {num_failed} failed test*\n"
+ else:
+ message += f"*{name}: {num_failed} failed tests*\n"
+ failed_table = []
+ for test in failed_tests:
+ failed_table.append(test[0].split("::"))
+ failed_table = tabulate(
+ failed_table,
+ headers=["Test Location", "Test Case", "Test Name"],
+ showindex="always",
+ tablefmt="grid",
+ maxcolwidths=[12, 12, 12],
+ )
+ message += "\n```\n" + failed_table + "\n```"
+
+ if total_empty_files[i]:
+ message += f"\n*{name}: Warning! Empty file - please check the GitHub action job *\n"
+ print(f"### {message}")
+ else:
+ payload.append(no_error_payload)
+
+ if len(message) > MAX_LEN_MESSAGE:
+ print(f"Truncating long message from {len(message)} to {MAX_LEN_MESSAGE}")
+ message = message[:MAX_LEN_MESSAGE] + "..."
+
+ if len(message) != 0:
+ md_report = {
+ "type": "section",
+ "text": {"type": "mrkdwn", "text": message},
+ }
+ payload.append(md_report)
+ action_button = {
+ "type": "section",
+ "text": {"type": "mrkdwn", "text": "*For more details:*"},
+ "accessory": {
+ "type": "button",
+ "text": {"type": "plain_text", "text": "Check Action results", "emoji": True},
+ "url": f"https://github.com/huggingface/diffusers/actions/runs/{os.environ['GITHUB_RUN_ID']}",
+ },
+ }
+ payload.append(action_button)
+
+ date_report = {
+ "type": "context",
+ "elements": [
+ {
+ "type": "plain_text",
+ "text": f"Nightly test results for {date.today()}",
+ },
+ ],
+ }
+ payload.append(date_report)
+
+ print(payload)
+
+ client = WebClient(token=os.environ.get("SLACK_API_TOKEN"))
+ client.chat_postMessage(channel=f"#{slack_channel_name}", text=message, blocks=payload)
+
+
+if __name__ == "__main__":
+ args = parser.parse_args()
+ main(args.slack_channel_name)