further improve

run_kandinsky.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python3
+from diffusers import DiffusionPipeline
+import torch
+from diffusers.models.attention_processor import AttnAddedKVProcessor2_0, XFormersAttnAddedKVProcessor, AttnAddedKVProcessor
+
+import time
+import os
+from huggingface_hub import HfApi
+from pathlib import Path
+
+api = HfApi()
+prev_time = time.time()
+
+
+prompt = "a picture of elon musk next to a rocket"
+negative_prompt = "low quality, ugly"
+
+pipe_prior = DiffusionPipeline.from_pretrained(
+    "kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16
+)
+pipe_prior.to("cuda")
+t2i_pipe = DiffusionPipeline.from_pretrained("kandinsky-community/kandinsky-2-1", torch_dtype=torch.float16)
+t2i_pipe.to("cuda")
+
+t2i_pipe.unet.set_attn_processor(AttnAddedKVProcessor2_0())
+t2i_pipe.unet.to(memory_format=torch.channels_last)
+t2i_pipe.unet = torch.compile(t2i_pipe.unet, mode="reduce-overhead", fullgraph=True)
+
+next_time = time.time()
+print("Loading", next_time - prev_time)
+prev_time = next_time
+
+generator = torch.Generator(device="cuda").manual_seed(12)
+image_embeds, negative_image_embeds = pipe_prior(prompt, negative_prompt=negative_prompt, generator=generator).to_tuple()
+
+next_time = time.time()
+print("Prior", next_time - prev_time)
+prev_time = next_time
+
+for _ in range(3):
+    images = t2i_pipe(prompt, image_embeds=image_embeds, negative_image_embeds=negative_image_embeds, negative_prompt=negative_prompt, num_inference_steps=50, generator=generator).images
+
+next_time = time.time()
+print("Text-to-image", next_time - prev_time)
+prev_time = next_time
+
+for i, image in enumerate(images):
+    path = os.path.join(Path.home(), "images", f"aa_{i}.png")
+    image.save(path)
+
+    api.upload_file(
+        path_or_fileobj=path,
+        path_in_repo=path.split("/")[-1],
+        repo_id="patrickvonplaten/images",
+        repo_type="dataset",
+    )
+    print(f"https://huggingface.co/datasets/patrickvonplaten/images/blob/main/aa_{i}.png")

run_local.py

@@ -1,9 +1,8 @@
 #!/usr/bin/env python3
-from diffusers import StableDiffusionControlNetPipeline
+from diffusers import StableDiffusionPipeline
 import time
 import os
 from huggingface_hub import HfApi
-# from compel import Compel
 import torch
 import sys
 from pathlib import Path
@@ -11,47 +10,31 @@ import requests
 from PIL import Image
 from io import BytesIO
 
-path = sys.argv[1]
+begin = ["a picture of <rickmann>", "a photo of <rickmann>", "<rickmann>", "an image of <rickmann>"]
+end = ["", " smiling", " with sunglasses", " at the beach", " in front of a mountain", " in beautiful sunshine", " in avatar style", " as a picasso painting", " as an oil painting", " as oil art", " while it snows", " in a forest", " with a nice landscape"]
 
 api = HfApi()
 start_time = time.time()
-#pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16, device_map="auto")
-#pipe.scheduler = HeunDiscreteScheduler.from_config(pipe.scheduler.config)
-pipe = StableDiffusionControlNetPipeline.from_pretrained(path, torch_dtype=torch.float16, safety_checker=None)
-pipe.scheduler = KDPM2AncestralDiscreteScheduler.from_config(pipe.scheduler.config)
-
-# compel = Compel(tokenizer=pipe.tokenizer, text_encoder=pipe.text_encoder)
-
-
+path = "patrickvonplaten/papa_out_5"
+pipe = StableDiffusionPipeline.from_pretrained(path, safety_checker=None, torch_dtype=torch.float16)
 pipe = pipe.to("cuda")
-
-prompt = "ghibli style, a fantasy landscape with castles"
-
-# rompts = ["a cat playing with a ball++ in the forest", "a cat playing with a ball++ in the forest", "a cat playing with a ball-- in the forest"]
-
-# prompt_embeds = torch.cat([compel.build_conditioning_tensor(prompt) for prompt in prompts])
-
-# generator = [torch.Generator(device="cuda").manual_seed(0) for _ in range(prompt_embeds.shape[0])]
-#
-url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
-
-response = requests.get(url)
-image = Image.open(BytesIO(response.content)).convert("RGB")
-image.thumbnail((768, 768))
-
-generator = torch.Generator(device="cpu").manual_seed(0)
-images = pipe(prompt=prompt, image=image, generator=generator, strength=0.75, num_inference_steps=30).images
-
-print("Time", time.time() - start_time)
-
-for i, image in enumerate(images):
-    path = os.path.join(Path.home(), "images", f"aa_{i}.png")
-    image.save(path)
-
-    api.upload_file(
-        path_or_fileobj=path,
-        path_in_repo=path.split("/")[-1],
-        repo_id="patrickvonplaten/images",
-        repo_type="dataset",
-    )
-    print(f"https://huggingface.co/datasets/patrickvonplaten/images/blob/main/aa_{i}.png")
+counter = 0
+
+for b in begin:
+    for e in end:
+        prompt = b + e + ", highly realistic, super resolution, high quality photography, beautiful"
+
+        images = pipe(prompt=prompt, num_images_per_prompt=4, negative_prompt="ugly, bad quality, deformed", num_inference_steps=50).images
+
+        for i, image in enumerate(images):
+            path = os.path.join(Path.home(), "papa", f"{counter}.png")
+            image.save(path)
+
+            api.upload_file(
+                path_or_fileobj=path,
+                path_in_repo=path.split("/")[-1],
+                repo_id="patrickvonplaten/papa",
+                repo_type="dataset",
+            )
+            print(f"https://huggingface.co/datasets/patrickvonplaten/papa/blob/main/{counter}.png")
+            counter += 1