Spaces:
Paused
Paused
File size: 17,379 Bytes
cd458ae | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 | # Copyright 2026 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 ...utils import logging
from ..modular_pipeline import AutoPipelineBlocks, SequentialPipelineBlocks
from ..modular_pipeline_utils import OutputParam
from .before_denoise import (
WanAdditionalInputsStep,
WanPrepareLatentsStep,
WanSetTimestepsStep,
WanTextInputStep,
)
from .decoders import WanVaeDecoderStep
from .denoise import (
WanImage2VideoDenoiseStep,
)
from .encoders import (
WanFirstLastFrameImageEncoderStep,
WanFirstLastFrameVaeEncoderStep,
WanImageCropResizeStep,
WanImageEncoderStep,
WanImageResizeStep,
WanPrepareFirstFrameLatentsStep,
WanPrepareFirstLastFrameLatentsStep,
WanTextEncoderStep,
WanVaeEncoderStep,
)
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
# ====================
# 1. IMAGE ENCODER
# ====================
# wan2.1 I2V (first frame only)
# auto_docstring
class WanImage2VideoImageEncoderStep(SequentialPipelineBlocks):
"""
Image2Video Image Encoder step that resize the image and encode the image to generate the image embeddings
Components:
image_processor (`CLIPImageProcessor`) image_encoder (`CLIPVisionModel`)
Inputs:
image (`Image`):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
Outputs:
resized_image (`Image`):
TODO: Add description.
image_embeds (`Tensor`):
The image embeddings
"""
model_name = "wan-i2v"
block_classes = [WanImageResizeStep, WanImageEncoderStep]
block_names = ["image_resize", "image_encoder"]
@property
def description(self):
return "Image2Video Image Encoder step that resize the image and encode the image to generate the image embeddings"
# wan2.1 FLF2V (first and last frame)
# auto_docstring
class WanFLF2VImageEncoderStep(SequentialPipelineBlocks):
"""
FLF2V Image Encoder step that resize and encode and encode the first and last frame images to generate the image
embeddings
Components:
image_processor (`CLIPImageProcessor`) image_encoder (`CLIPVisionModel`)
Inputs:
image (`Image`):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
last_image (`Image`):
The last frameimage
Outputs:
resized_image (`Image`):
TODO: Add description.
resized_last_image (`Image`):
TODO: Add description.
image_embeds (`Tensor`):
The image embeddings
"""
model_name = "wan-i2v"
block_classes = [WanImageResizeStep, WanImageCropResizeStep, WanFirstLastFrameImageEncoderStep]
block_names = ["image_resize", "last_image_resize", "image_encoder"]
@property
def description(self):
return "FLF2V Image Encoder step that resize and encode and encode the first and last frame images to generate the image embeddings"
# wan2.1 Auto Image Encoder
# auto_docstring
class WanAutoImageEncoderStep(AutoPipelineBlocks):
"""
Image Encoder step that encode the image to generate the image embeddingsThis is an auto pipeline block that works
for image2video tasks. - `WanFLF2VImageEncoderStep` (flf2v) is used when `last_image` is provided. -
`WanImage2VideoImageEncoderStep` (image2video) is used when `image` is provided. - if `last_image` or `image` is
not provided, step will be skipped.
Components:
image_processor (`CLIPImageProcessor`) image_encoder (`CLIPVisionModel`)
Inputs:
image (`Image`, *optional*):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
last_image (`Image`, *optional*):
The last frameimage
Outputs:
resized_image (`Image`):
TODO: Add description.
resized_last_image (`Image`):
TODO: Add description.
image_embeds (`Tensor`):
The image embeddings
"""
block_classes = [WanFLF2VImageEncoderStep, WanImage2VideoImageEncoderStep]
block_names = ["flf2v_image_encoder", "image2video_image_encoder"]
block_trigger_inputs = ["last_image", "image"]
model_name = "wan-i2v"
@property
def description(self):
return (
"Image Encoder step that encode the image to generate the image embeddings"
+ "This is an auto pipeline block that works for image2video tasks."
+ " - `WanFLF2VImageEncoderStep` (flf2v) is used when `last_image` is provided."
+ " - `WanImage2VideoImageEncoderStep` (image2video) is used when `image` is provided."
+ " - if `last_image` or `image` is not provided, step will be skipped."
)
# ====================
# 2. VAE ENCODER
# ====================
# wan2.1 I2V (first frame only)
# auto_docstring
class WanImage2VideoVaeEncoderStep(SequentialPipelineBlocks):
"""
Image2Video Vae Image Encoder step that resize the image and encode the first frame image to its latent
representation
Components:
vae (`AutoencoderKLWan`) video_processor (`VideoProcessor`)
Inputs:
image (`Image`):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
num_frames (`int`, *optional*, defaults to 81):
TODO: Add description.
generator (`None`, *optional*):
TODO: Add description.
Outputs:
resized_image (`Image`):
TODO: Add description.
first_frame_latents (`Tensor`):
video latent representation with the first frame image condition
image_condition_latents (`Tensor | NoneType`):
TODO: Add description.
"""
model_name = "wan-i2v"
block_classes = [WanImageResizeStep, WanVaeEncoderStep, WanPrepareFirstFrameLatentsStep]
block_names = ["image_resize", "vae_encoder", "prepare_first_frame_latents"]
@property
def description(self):
return "Image2Video Vae Image Encoder step that resize the image and encode the first frame image to its latent representation"
# wan2.1 FLF2V (first and last frame)
# auto_docstring
class WanFLF2VVaeEncoderStep(SequentialPipelineBlocks):
"""
FLF2V Vae Image Encoder step that resize and encode and encode the first and last frame images to generate the
latent conditions
Components:
vae (`AutoencoderKLWan`) video_processor (`VideoProcessor`)
Inputs:
image (`Image`):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
last_image (`Image`):
The last frameimage
num_frames (`int`, *optional*, defaults to 81):
TODO: Add description.
generator (`None`, *optional*):
TODO: Add description.
Outputs:
resized_image (`Image`):
TODO: Add description.
resized_last_image (`Image`):
TODO: Add description.
first_last_frame_latents (`Tensor`):
video latent representation with the first and last frame images condition
image_condition_latents (`Tensor | NoneType`):
TODO: Add description.
"""
model_name = "wan-i2v"
block_classes = [
WanImageResizeStep,
WanImageCropResizeStep,
WanFirstLastFrameVaeEncoderStep,
WanPrepareFirstLastFrameLatentsStep,
]
block_names = ["image_resize", "last_image_resize", "vae_encoder", "prepare_first_last_frame_latents"]
@property
def description(self):
return "FLF2V Vae Image Encoder step that resize and encode and encode the first and last frame images to generate the latent conditions"
# wan2.1 Auto Vae Encoder
# auto_docstring
class WanAutoVaeEncoderStep(AutoPipelineBlocks):
"""
Vae Image Encoder step that encode the image to generate the image latentsThis is an auto pipeline block that works
for image2video tasks. - `WanFLF2VVaeEncoderStep` (flf2v) is used when `last_image` is provided. -
`WanImage2VideoVaeEncoderStep` (image2video) is used when `image` is provided. - if `last_image` or `image` is not
provided, step will be skipped.
Components:
vae (`AutoencoderKLWan`) video_processor (`VideoProcessor`)
Inputs:
image (`Image`, *optional*):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
last_image (`Image`, *optional*):
The last frameimage
num_frames (`int`, *optional*, defaults to 81):
TODO: Add description.
generator (`None`, *optional*):
TODO: Add description.
Outputs:
resized_image (`Image`):
TODO: Add description.
resized_last_image (`Image`):
TODO: Add description.
first_last_frame_latents (`Tensor`):
video latent representation with the first and last frame images condition
image_condition_latents (`Tensor | NoneType`):
TODO: Add description.
first_frame_latents (`Tensor`):
video latent representation with the first frame image condition
"""
model_name = "wan-i2v"
block_classes = [WanFLF2VVaeEncoderStep, WanImage2VideoVaeEncoderStep]
block_names = ["flf2v_vae_encoder", "image2video_vae_encoder"]
block_trigger_inputs = ["last_image", "image"]
@property
def description(self):
return (
"Vae Image Encoder step that encode the image to generate the image latents"
+ "This is an auto pipeline block that works for image2video tasks."
+ " - `WanFLF2VVaeEncoderStep` (flf2v) is used when `last_image` is provided."
+ " - `WanImage2VideoVaeEncoderStep` (image2video) is used when `image` is provided."
+ " - if `last_image` or `image` is not provided, step will be skipped."
)
# ====================
# 3. DENOISE (inputs -> set_timesteps -> prepare_latents -> denoise)
# ====================
# wan2.1 I2V core denoise (support both I2V and FLF2V)
# inputs (text + image_condition_latents) -> set_timesteps -> prepare_latents -> denoise (latents)
# auto_docstring
class WanImage2VideoCoreDenoiseStep(SequentialPipelineBlocks):
"""
denoise block that takes encoded text and image latent conditions and runs the denoising process.
Components:
transformer (`WanTransformer3DModel`) scheduler (`UniPCMultistepScheduler`) guider (`ClassifierFreeGuidance`)
Inputs:
num_videos_per_prompt (`None`, *optional*, defaults to 1):
TODO: Add description.
prompt_embeds (`Tensor`):
Pre-generated text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
Pre-generated negative text embeddings. Can be generated from text_encoder step.
height (`None`, *optional*):
TODO: Add description.
width (`None`, *optional*):
TODO: Add description.
num_frames (`None`, *optional*):
TODO: Add description.
image_condition_latents (`None`, *optional*):
TODO: Add description.
num_inference_steps (`None`, *optional*, defaults to 50):
TODO: Add description.
timesteps (`None`, *optional*):
TODO: Add description.
sigmas (`None`, *optional*):
TODO: Add description.
latents (`Tensor | NoneType`, *optional*):
TODO: Add description.
generator (`None`, *optional*):
TODO: Add description.
attention_kwargs (`None`, *optional*):
TODO: Add description.
image_embeds (`Tensor`):
TODO: Add description.
Outputs:
batch_size (`int`):
Number of prompts, the final batch size of model inputs should be batch_size * num_videos_per_prompt
dtype (`dtype`):
Data type of model tensor inputs (determined by `transformer.dtype`)
latents (`Tensor`):
The initial latents to use for the denoising process
"""
model_name = "wan-i2v"
block_classes = [
WanTextInputStep,
WanAdditionalInputsStep(image_latent_inputs=["image_condition_latents"]),
WanSetTimestepsStep,
WanPrepareLatentsStep,
WanImage2VideoDenoiseStep,
]
block_names = [
"input",
"additional_inputs",
"set_timesteps",
"prepare_latents",
"denoise",
]
@property
def description(self):
return "denoise block that takes encoded text and image latent conditions and runs the denoising process."
# ====================
# 4. BLOCKS (Wan2.1 Image2Video)
# ====================
# wan2.1 Image2Video Auto Blocks
# auto_docstring
class WanImage2VideoAutoBlocks(SequentialPipelineBlocks):
"""
Auto Modular pipeline for image-to-video using Wan.
Supported workflows:
- `image2video`: requires `image`, `prompt`
- `flf2v`: requires `last_image`, `image`, `prompt`
Components:
text_encoder (`UMT5EncoderModel`) tokenizer (`AutoTokenizer`) guider (`ClassifierFreeGuidance`)
image_processor (`CLIPImageProcessor`) image_encoder (`CLIPVisionModel`) vae (`AutoencoderKLWan`)
video_processor (`VideoProcessor`) transformer (`WanTransformer3DModel`) scheduler
(`UniPCMultistepScheduler`)
Inputs:
prompt (`None`, *optional*):
TODO: Add description.
negative_prompt (`None`, *optional*):
TODO: Add description.
max_sequence_length (`None`, *optional*, defaults to 512):
TODO: Add description.
image (`Image`, *optional*):
TODO: Add description.
height (`int`, *optional*, defaults to 480):
TODO: Add description.
width (`int`, *optional*, defaults to 832):
TODO: Add description.
last_image (`Image`, *optional*):
The last frameimage
num_frames (`int`, *optional*, defaults to 81):
TODO: Add description.
generator (`None`, *optional*):
TODO: Add description.
num_videos_per_prompt (`None`, *optional*, defaults to 1):
TODO: Add description.
image_condition_latents (`None`, *optional*):
TODO: Add description.
num_inference_steps (`None`, *optional*, defaults to 50):
TODO: Add description.
timesteps (`None`, *optional*):
TODO: Add description.
sigmas (`None`, *optional*):
TODO: Add description.
latents (`Tensor | NoneType`, *optional*):
TODO: Add description.
attention_kwargs (`None`, *optional*):
TODO: Add description.
image_embeds (`Tensor`):
TODO: Add description.
output_type (`str`, *optional*, defaults to np):
The output type of the decoded videos
Outputs:
videos (`list`):
The generated videos.
"""
model_name = "wan-i2v"
block_classes = [
WanTextEncoderStep,
WanAutoImageEncoderStep,
WanAutoVaeEncoderStep,
WanImage2VideoCoreDenoiseStep,
WanVaeDecoderStep,
]
block_names = [
"text_encoder",
"image_encoder",
"vae_encoder",
"denoise",
"decode",
]
_workflow_map = {
"image2video": {"image": True, "prompt": True},
"flf2v": {"last_image": True, "image": True, "prompt": True},
}
@property
def description(self):
return "Auto Modular pipeline for image-to-video using Wan."
@property
def outputs(self):
return [OutputParam.template("videos")]
|