Merge branch #aipicasso/cool-japan-diffusion-latest-demo' into 'ReiPlush64/cool-japan-diffusion-latest-demo'

app.py

@@ -5,6 +5,7 @@ import gradio as gr
 import torch
 from PIL import Image
 import random
+import spaces
 
 import requests
 
@@ -14,15 +15,15 @@ import requests
  
 model_id = 'aipicasso/cool-japan-diffusion-2-1-2'
 
-scheduler = EulerAncestralDiscreteScheduler(prediction_type="v_prediction") # .from_pretrained(model_id) # , subfolder="scheduler"
-feature_extractor = CLIPImageProcessor() # .from_pretrained(model_id)
+scheduler = EulerAncestralDiscreteScheduler.from_pretrained(model_id, subfolder="scheduler")
+feature_extractor = CLIPImageProcessor.from_pretrained(model_id)
 
 pipe = StableDiffusionPipeline.from_pretrained(
   model_id,
   torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
   scheduler=scheduler)
-pipe.enable_xformers_memory_efficient_attention()
-pipe.enable_freeu(s1=0.9, s2=0.2, b1=1.4, b2=1.6)
+#pipe.enable_xformers_memory_efficient_attention()
+#pipe.enable_freeu(s1=0.9, s2=0.2, b1=1.4, b2=1.6)
 # b1: 1.4, b2: 1.6, s1: 0.9, s2: 0.2
 pipe_i2i = StableDiffusionImg2ImgPipeline.from_pretrained(
   model_id,
@@ -32,21 +33,22 @@ pipe_i2i = StableDiffusionImg2ImgPipeline.from_pretrained(
   safety_checker=None,
   feature_extractor=feature_extractor
 )
-pipe_i2i.enable_xformers_memory_efficient_attention()
+#pipe_i2i.enable_xformers_memory_efficient_attention()
 
 upscaler = StableDiffusionLatentUpscalePipeline.from_pretrained("alfredplpl/x2-latent-upscaler-for-anime", torch_dtype=torch.float16)
-upscaler.enable_xformers_memory_efficient_attention()
+#upscaler.enable_xformers_memory_efficient_attention()
 
 
 if torch.cuda.is_available():
   pipe = pipe.to("cuda")
   pipe_i2i = pipe_i2i.to("cuda")
+  upscaler=upscaler.to("cuda")
 
 def error_str(error, title="Error"):
     return f"""#### {title}
             {error}"""  if error else ""
 
-
+@spaces.GPU
 def inference(prompt, guidance, steps, image_size="Square", seed=0, img=None, strength=0.5, neg_prompt="", cool_japan_type="Anime", disable_auto_prompt_correction=False):
 
   generator = torch.Generator('cuda').manual_seed(seed) if seed != 0 else None
@@ -142,7 +144,7 @@ def auto_prompt_correction(prompt_ui,neg_prompt_ui,cool_japan_type_ui,disable_au
             neg_prompt=f"(((deformed))), {neg_prompt}, girl, boy, photo, people, low quality, ui, error, lowres, jpeg artifacts, 2d, 3d, cg, text"
 
     return prompt,neg_prompt
-    
+
 def txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator,superreso=False):
     global pipe, upscaler
     if(superreso):
@@ -155,8 +157,6 @@ def txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator,sup
         height = height,
         output_type="latent",
         generator = generator).images
-      pipe=pipe.to("cpu")
-      upscaler=upscaler.to("cuda")
       result = upscaler(
         prompt=prompt,
         negative_prompt = neg_prompt,
@@ -165,8 +165,6 @@ def txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator,sup
         guidance_scale=0,
         generator=generator,
       )
-      pipe=pipe.to("cuda")
-      upscaler=upscaler.to("cpu")
     else:  
       result = pipe(
         prompt,
@@ -194,8 +192,6 @@ def img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height
         #height = height,
         output_type="latent",
         generator = generator).images
-      pipe=pipe.to("cpu")
-      upscaler=upscaler.to("cuda")
       result = upscaler(
         prompt=prompt,
         negative_prompt = neg_prompt,
@@ -204,8 +200,6 @@ def img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height
         guidance_scale=0,
         generator=generator,
       )
-      pipe=pipe.to("cuda")
-      upscaler=upscaler.to("cpu")
     else:  
       result = pipe_i2i(
         prompt,

requirements.txt

@@ -1,9 +1,3 @@
-torch
-diffusers>=0.26.3
-transformers>=4.38.2
 accelerate
-ftfy
-triton
-xformers
-requests
-huggingface_hub>=0.21.4
+transformers
+diffusers