further improve

run_kandinsky.py

@@ -8,41 +8,24 @@ import os
 from huggingface_hub import HfApi
 from pathlib import Path
 
-api = HfApi()
-prev_time = time.time()
-
+from diffusers import DiffusionPipeline
+import torch
 
-prompt = "a picture of elon musk next to a rocket"
-negative_prompt = "low quality, ugly"
+api = HfApi()
 
-pipe_prior = DiffusionPipeline.from_pretrained(
-    "kandinsky-community/kandinsky-2-1-prior", torch_dtype=torch.float16
-)
+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
+prompt = "A alien cheeseburger creature eating itself, claymation, cinematic, moody lighting"
+negative_prompt = "low quality, bad quality"
 
 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
+image_embeds, negative_image_embeds = pipe_prior(prompt, negative_prompt, guidance_scale=1.0, generator=generator).to_tuple()
 
-next_time = time.time()
-print("Text-to-image", next_time - prev_time)
-prev_time = next_time
+images = t2i_pipe(prompt, num_images_per_prompt=4, image_embeds=image_embeds, negative_image_embeds=negative_image_embeds, negative_prompt=negative_prompt).images
 
 for i, image in enumerate(images):
     path = os.path.join(Path.home(), "images", f"aa_{i}.png")

run_local.py

@@ -1,8 +1,9 @@
 #!/usr/bin/env python3
-from diffusers import StableDiffusionPipeline, DDIMScheduler
+from diffusers import StableDiffusionPipeline, KDPM2DiscreteScheduler, StableDiffusionImg2ImgPipeline, HeunDiscreteScheduler, KDPM2AncestralDiscreteScheduler, DDIMScheduler
 import time
 import os
 from huggingface_hub import HfApi
+# from compel import Compel
 import torch
 import sys
 from pathlib import Path
@@ -10,34 +11,53 @@ import requests
 from PIL import Image
 from io import BytesIO
 
-begin = ["a picture of <rickmann>", "a photo of <rickmann>", "The <rickmann>", "an image of <rickmann>"]
-mid = ["", " on a bike", " with sunglasses", " at the beach", " in front of a mountain", " in the water", " on a boat", " at a fashion show", " as a superstar model", " while it snows", " in a forest", " with a nice landscape"]
-end = ["", " , disco light style", ", minecraft style", " , picasso style", " as a lego person", ""]
+# path = sys.argv[1]
+path = "runwayml/stable-diffusion-v1-5"
+# path = "stabilityai/stable-diffusion-2-1"
 
 api = HfApi()
 start_time = time.time()
-path = "patrickvonplaten/papa_out_5"
-pipe = StableDiffusionPipeline.from_pretrained(path, safety_checker=None, torch_dtype=torch.float16)
-pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config)
+pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
+# pipe = StableDiffusionImg2ImgPipeline.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)
+
+
 pipe = pipe.to("cuda")
-counter = 1000
 
-for b in begin:
-    for m in mid:
-        for e in end:
-            prompt = b + m + e + ", highly realistic, super resolution, high quality photography, beautiful"
+prompt = "A lion in galaxies, spirals, nebulae, stars, smoke, iridescent, intricate detail, octane render, 8k"
+
+# 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))
+#
+
+for TIMESTEP_TYPE in ["trailing", "leading"]:
+    for RESCALE_BETAS_ZEROS_SNR in [True, False]:
+        for GUIDANCE_RESCALE in [0,0, 0.7]:
 
-            images = pipe(prompt=prompt, num_images_per_prompt=2, eta=1.0, negative_prompt="ugly, bad quality, deformed", num_inference_steps=50).images
+            pipe.scheduler = DDIMScheduler.from_config(pipe.scheduler.config, timestep_type=TIMESTEP_TYPE, rescale_betas_zero_snr=RESCALE_BETAS_ZEROS_SNR)
+            generator = torch.Generator(device="cpu").manual_seed(0)
+            images = pipe(prompt=prompt, generator=generator, num_images_per_prompt=4, num_inference_steps=40, guidance_rescale=GUIDANCE_RESCALE).images
 
             for i, image in enumerate(images):
-                path = os.path.join(Path.home(), "papa", f"{counter}.png")
+                file_name = f"bb_{i}_{TIMESTEP_TYPE}_{str(int(RESCALE_BETAS_ZEROS_SNR))}_{GUIDANCE_RESCALE}"
+                path = os.path.join(Path.home(), "images", f"{file_name}.png")
                 image.save(path)
 
                 api.upload_file(
                     path_or_fileobj=path,
                     path_in_repo=path.split("/")[-1],
-                    repo_id="patrickvonplaten/papa",
+                    repo_id="patrickvonplaten/images",
                     repo_type="dataset",
                 )
-                print(f"https://huggingface.co/datasets/patrickvonplaten/papa/blob/main/{counter}.png")
-                counter += 1
+                print(f"https://huggingface.co/datasets/patrickvonplaten/images/blob/main/{file_name}.png")