Update app.py

app.py

@@ -1,76 +1,68 @@
 from diffusers import StableDiffusionPipeline
 from diffusers import StableDiffusionImg2ImgPipeline
-from diffusers import AutoencoderKL, UNet2DConditionModel
 import gradio as gr
 import torch
+from PIL import Image
+import utils
+
+is_colab = utils.is_google_colab()
+
+max_width = 832
+max_height = 832
+
+class Model:
+    def __init__(self, name, path, prefix):
+        self.name = name
+        self.path = path
+        self.prefix = prefix
 
 models = [
-  "nitrosocke/Arcane-Diffusion",
-  "nitrosocke/archer-diffusion",
-  "nitrosocke/elden-ring-diffusion",
-  "nitrosocke/spider-verse-diffusion",
-  "nitrosocke/modern-disney-diffusion",
-  "hakurei/waifu-diffusion",
-  "lambdalabs/sd-pokemon-diffusers",
-  "yuk/fuyuko-waifu-diffusion",
-  "AstraliteHeart/pony-diffusion",
-  "nousr/robo-diffusion",
-  "DGSpitzer/Cyberpunk-Anime-Diffusion",
-  "sd-dreambooth-library/herge-style"
+     Model("Arcane", "nitrosocke/Arcane-Diffusion", "arcane style "),
+     Model("Archer", "nitrosocke/archer-diffusion", "archer style "),
+     Model("Elden Ring", "nitrosocke/elden-ring-diffusion", "elden ring style "),
+     Model("Spider-Verse", "nitrosocke/spider-verse-diffusion", "spiderverse style "),
+     Model("Modern Disney", "nitrosocke/modern-disney-diffusion", "modern disney style "),
+     Model("Classic Disney", "nitrosocke/classic-anim-diffusion", ""),
+     Model("Waifu", "hakurei/waifu-diffusion", ""),
+     Model("Pokémon", "lambdalabs/sd-pokemon-diffusers", ""),
+     Model("Fuyuko Waifu", "yuk/fuyuko-waifu-diffusion", ""),
+     Model("Pony Diffusion", "AstraliteHeart/pony-diffusion", ""),
+     Model("Robo Diffusion", "nousr/robo-diffusion", ""),
+     Model("Cyberpunk Anime", "DGSpitzer/Cyberpunk-Anime-Diffusion", "dgs illustration style "),
+     Model("Hergé Style", "sd-dreambooth-library/herge-style", "herge_style "),
 ]
 
-prompt_prefixes = {
-  models[0]: "arcane style ",
-  models[1]: "archer style ",
-  models[2]: "elden ring style ",
-  models[3]: "spiderverse style ",
-  models[4]: "modern disney style ",
-  models[5]: "",
-  models[6]: "",
-  models[7]: "",
-  models[8]: "",
-  models[9]: "",
-  models[10]: "dgs illustration style ",
-  models[11]: "herge_style ",
-}
-
 current_model = models[0]
-pipes = []
-vae = AutoencoderKL.from_pretrained(current_model, subfolder="vae", torch_dtype=torch.float16)
-for model in models:
-  unet = UNet2DConditionModel.from_pretrained(model, subfolder="unet", torch_dtype=torch.float16)
-  pipe = StableDiffusionPipeline.from_pretrained(model, unet=unet, vae=vae, torch_dtype=torch.float16)
-  pipe_i2i = StableDiffusionImg2ImgPipeline.from_pretrained(model, unet=unet, vae=vae, torch_dtype=torch.float16)
-  pipes.append({"name":model, "pipeline":pipe, "pipeline_i2i":pipe_i2i})
+pipe = StableDiffusionPipeline.from_pretrained(current_model.path, torch_dtype=torch.float16)
+if torch.cuda.is_available():
+  pipe = pipe.to("cuda")
 
 
 device = "GPU 🔥" if torch.cuda.is_available() else "CPU 🥶"
 
-def inference(model, img, strength, prompt, neg_prompt, guidance, steps, width, height, seed):
+def inference(model_name, prompt, guidance, steps, width=512, height=512, seed=0, img=None, strength=0.5, neg_prompt=""):
 
   generator = torch.Generator('cuda').manual_seed(seed) if seed != 0 else None
   
   if img is not None:
-    return img_to_img(model, prompt, neg_prompt, img, strength, guidance, steps, width, height, generator)
+    return img_to_img(model_name, prompt, neg_prompt, img, strength, guidance, steps, width, height, generator)
   else:
-    return txt_to_img(model, prompt, neg_prompt, guidance, steps, width, height, generator)
+    return txt_to_img(model_name, prompt, neg_prompt, guidance, steps, width, height, generator)
 
-def txt_to_img(model, prompt, neg_prompt, guidance, steps, width, height, generator=None):
+def txt_to_img(model_name, prompt, neg_prompt, guidance, steps, width, height, generator=None):
 
     global current_model
     global pipe
-    if model != current_model:
-        current_model = model 
-        pipe = pipe.to("cpu")
-        
-    for pipe_dict in pipes:
-          if(pipe_dict["name"] == current_model):
-            pipe = pipe_dict["pipeline"]
-          
-          if torch.cuda.is_available():
-            pipe = pipe.to("cuda")
-
-    prompt = prompt_prefixes[current_model] + prompt
+    if model_name != current_model.name:
+        for model in models:
+          if model.name == model_name:
+            current_model = model
+
+        pipe = StableDiffusionPipeline.from_pretrained(current_model.path, torch_dtype=torch.float16)
+        if torch.cuda.is_available():
+          pipe = pipe.to("cuda")
+
+    prompt = current_model.prefix + prompt
     results = pipe(
       prompt,
       negative_prompt=neg_prompt,
@@ -87,19 +79,18 @@ def img_to_img(model, prompt, neg_prompt, img, strength, guidance, steps, width,
 
     global current_model
     global pipe
-    if model != current_model:
-        current_model = model
-        pipe = pipe.to("cpu")
+    if model_name != current_model.name:
+        for model in models:
+          if model.name == model_name:
+            current_model = model
+
+        pipe = StableDiffusionImg2ImgPipeline.from_pretrained(current_model.path, torch_dtype=torch.float16)
         
-    for pipe_dict in pipes:
-          if(pipe_dict["name"] == current_model):
-            pipe = pipe_dict["pipeline_i2i"]
-          
-          if torch.cuda.is_available():
-            pipe = pipe.to("cuda")
-
-    prompt = prompt_prefixes[current_model] + prompt
-    ratio = min(height / img.height, width / img.width)
+        if torch.cuda.is_available():
+              pipe = pipe.to("cuda")
+
+    prompt = current_model.prefix + prompt
+    ratio = min(max_height / img.height, max_width / img.width)
     img = img.resize((int(img.width * ratio), int(img.height * ratio)))
     results = pipe(
         prompt,
@@ -115,7 +106,6 @@ def img_to_img(model, prompt, neg_prompt, img, strength, guidance, steps, width,
     image = results.images[0] if not results.nsfw_content_detected[0] else Image.open("nsfw.png")
     return image
 
-
 css = """
   <style>
   .finetuned-diffusion-div {
@@ -144,7 +134,7 @@ css = """
 """
 with gr.Blocks(css=css) as demo:
     gr.HTML(
-        """
+        f"""
             <div class="finetuned-diffusion-div">
               <div>
                 <h1>Finetuned Diffusion</h1>
@@ -152,26 +142,29 @@ with gr.Blocks(css=css) as demo:
               <p>
                Demo for multiple fine-tuned Stable Diffusion models, trained on different styles: <br>
                <a href="https://huggingface.co/nitrosocke/Arcane-Diffusion">Arcane</a>, <a href="https://huggingface.co/nitrosocke/archer-diffusion">Archer</a>, <a href="https://huggingface.co/nitrosocke/elden-ring-diffusion">Elden Ring</a>, <a href="https://huggingface.co/nitrosocke/spider-verse-diffusion">Spiderverse</a>, <a href="https://huggingface.co/nitrosocke/modern-disney-diffusion">Modern Disney</a>, <a href="https://huggingface.co/hakurei/waifu-diffusion">Waifu</a>, <a href="https://huggingface.co/lambdalabs/sd-pokemon-diffusers">Pokemon</a>, <a href="https://huggingface.co/yuk/fuyuko-waifu-diffusion">Fuyuko Waifu</a>, <a href="https://huggingface.co/AstraliteHeart/pony-diffusion">Pony</a>, <a href="https://huggingface.co/sd-dreambooth-library/herge-style">Hergé (Tintin)</a>, <a href="https://huggingface.co/nousr/robo-diffusion">Robo</a>, <a href="https://huggingface.co/DGSpitzer/Cyberpunk-Anime-Diffusion">Cyberpunk Anime</a>
+              </p> <br>
+              <p>
+               Running on <b>{device}</b>
               </p>
             </div>
         """
     )
+    # gr.Markdown(f"Running on: {device}", elem_id="markdown_device")
     with gr.Row():
         
         with gr.Column():
-            model = gr.Dropdown(label="Model", choices=models, value=models[0])
+            model_name = gr.Dropdown(label="Model", choices=[m.name for m in models], value=models[0].name)
             prompt = gr.Textbox(label="Prompt", placeholder="Style prefix is applied automatically")
             run = gr.Button(value="Run")
-            gr.Markdown(f"Running on: {device}")
-            
-            with gr.Tab("Options"):
 
+            with gr.Tab("Options"):
                 neg_prompt = gr.Textbox(label="Negative prompt", placeholder="What to exclude from the image")
                 guidance = gr.Slider(label="Guidance scale", value=7.5, maximum=15)
-                steps = gr.Slider(label="Steps", value=50, maximum=100, minimum=2)
-                width = gr.Slider(label="Width", value=512, maximum=1024, minimum=64)
-                height = gr.Slider(label="Height", value=512, maximum=1024, minimum=64)
+                steps = gr.Slider(label="Steps", value=50, maximum=100, minimum=2, step=1)
+                width = gr.Slider(label="Width", value=512, maximum=max_width, minimum=64, step=8)
+                height = gr.Slider(label="Height", value=512, maximum=max_height, minimum=64, step=8)
                 seed = gr.Slider(0, 2147483647, label='Seed (0 = random)', value=0, step=1)
+                
             with gr.Tab("Image to image"):
                 image = gr.Image(label="Image", height=256, tool="editor", type="pil")
                 strength = gr.Slider(label="Transformation strength", minimum=0, maximum=1, step=0.01, value=0.5)
@@ -179,16 +172,18 @@ with gr.Blocks(css=css) as demo:
         with gr.Column():
             image_out = gr.Image(height=512)
 
-    inputs = [model, image, strength, prompt, neg_prompt, guidance, steps, width, height, seed]
+    inputs = [model_name, prompt, guidance, steps, width, height, seed, image, strength, neg_prompt]
     prompt.submit(inference, inputs=inputs, outputs=image_out)
     run.click(inference, inputs=inputs, outputs=image_out)
+  
     gr.Examples([
-        [models[0], "jason bateman disassembling the demon core", 7.5, 50],
-        [models[3], "portrait of dwayne johnson", 7.0, 75],
-        [models[4], "portrait of a beautiful alyx vance half life", 10, 50],
-        [models[5], "Aloy from Horizon: Zero Dawn, half body portrait, smooth, detailed armor, beautiful face, illustration", 7, 45],
-        [models[4], "fantasy portrait painting, digital art", 4, 30],
-    ], [model, prompt, guidance, steps], image_out, img_to_img, cache_examples=False)
+        [models[0].name, "jason bateman disassembling the demon core", 7.5, 50],
+        [models[3].name, "portrait of dwayne johnson", 7.0, 75],
+        [models[4].name, "portrait of a beautiful alyx vance half life", 10, 50],
+        [models[5].name, "Aloy from Horizon: Zero Dawn, half body portrait, smooth, detailed armor, beautiful face, illustration", 7.0, 45],
+        [models[4].name, "fantasy portrait painting, digital art", 4.0, 30],
+    ], [model_name, prompt, guidance, steps, seed], image_out, inference, cache_examples=not is_colab and torch.cuda.is_available())
+  
     gr.Markdown('''
       Models by [@nitrosocke](https://huggingface.co/nitrosocke), [@Helixngc7293](https://twitter.com/DGSpitzer) and others. ❤️<br>
       Space by: [![Twitter Follow](https://img.shields.io/twitter/follow/hahahahohohe?label=%40anzorq&style=social)](https://twitter.com/hahahahohohe)
@@ -196,5 +191,6 @@ with gr.Blocks(css=css) as demo:
       ![visitors](https://visitor-badge.glitch.me/badge?page_id=anzorq.finetuned_diffusion)
     ''')
 
-demo.queue()
-demo.launch()
+if not is_colab:
+  demo.queue()
+demo.launch(debug=is_colab)