Update app.py

app.py

@@ -1,64 +1,222 @@
-from diffusers import StableDiffusionPipeline
+from diffusers import AutoencoderKL, UNet2DConditionModel, StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, DPMSolverMultistepScheduler
+from diffusers import EulerAncestralDiscreteScheduler
 import gradio as gr
 import torch
 import os
 from PIL import Image
+import datetime
+import time
+import psutil
 
 from share_btn import community_icon_html, loading_icon_html, share_js
 
+def is_google_colab():
+    try:
+        import google.colab
+        return True
+    except:
+        return False
+
+is_colab = is_google_colab()
+        
+start_time = time.time()
+
+
 if torch.cuda.is_available():
   torchfloat = torch.float16
 else:
   torchfloat = torch.float32
 
+class Model:
+    def __init__(self, name, path="", prefix=""):
+        self.name = name
+        self.path = path
+        self.prefix = prefix
+        self.pipe_t2i = None
+        self.pipe_i2i = None
+
 
 models = [
-  "DGSpitzer/Cyberpunk-Anime-Diffusion"
+   Model("Cyberpunk Anime Diffusion", "DGSpitzer/Cyberpunk-Anime-Diffusion", "dgs illustration style "),
+   Model("Guan Yu Diffusion", "DGSpitzer/Guan-Yu-Diffusion", "")
 ]
 
-prompt_prefixes = {
-  models[0]: "dgs illustration style "
-}
+custom_model = None
+if is_colab:
+  models.insert(0, Model("Custom model"))
+  custom_model = models[0]
 
-current_model = models[0]
+last_mode = "txt2img"
+current_model = models[1] if is_colab else models[0]
+current_model_path = current_model.path
 
-#auth_token = os.environ.get("test") or True
-#pipe = StableDiffusionPipeline.from_pretrained(current_model, use_auth_token=auth_token, torch_dtype=torchfloat, revision="fp16")
-pipe = StableDiffusionPipeline.from_pretrained(current_model, torch_dtype=torchfloat)
+if is_colab:
+  pipe = StableDiffusionPipeline.from_pretrained(
+      current_model.path,
+      torch_dtype=torch.float16,
+      scheduler=DPMSolverMultistepScheduler.from_pretrained(current_model.path, subfolder="scheduler"),
+      safety_checker=lambda images, clip_input: (images, False)
+      )
 
+else:
+  eulera = EulerAncestralDiscreteScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000)
+  pipe = StableDiffusionPipeline.from_pretrained(
+      current_model.path,
+      torch_dtype=torch.float16,
+      scheduler=eulera
+      )
+    
 if torch.cuda.is_available():
   pipe = pipe.to("cuda")
-else:
-  pipe = pipe.to("cpu")
+  #pipe.enable_xformers_memory_efficient_attention()
 
 device = "GPU 🔥" if torch.cuda.is_available() else "CPU 🥶"
 
-def on_model_change(model):
+def error_str(error, title="Error"):
+    return f"""#### {title}
+            {error}"""  if error else ""
+
+def custom_model_changed(path):
+  models[0].path = path
+  global current_model
+  current_model = models[0]
+
+def on_model_change(model_name):
+  
+  prefix = "Enter prompt. \"" + next((m.prefix for m in models if m.name == model_name), None) + "\" is prefixed automatically" if model_name != models[0].name else "Don't forget to use the custom model prefix in the prompt!"
+
+  return gr.update(visible = model_name == models[0].name), gr.update(placeholder=prefix)
+
+def inference(model_name, prompt, neg_prompt, guidance, steps, width=512, height=512, seed=0, img=None, strength=0.5, n_images=1):
+  print("Generated image with prompt: " + prompt)
+  if neg_prompt != "":
+    print("Negative prompt: " + neg_prompt)
+      
+  print(psutil.virtual_memory()) # print memory usage
+
+  global current_model
+  for model in models:
+    if model.name == model_name:
+      current_model = model
+      model_path = current_model.path
+
+  generator = torch.Generator('cuda').manual_seed(seed) if seed != 0 else None
+
+  try:
+    if img is not None:
+      return img_to_img(model_path, prompt, neg_prompt, img, strength, guidance, steps, width, height, n_images, generator), None, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
+    else:
+      return txt_to_img(model_path, prompt, neg_prompt, guidance, steps, width, height, n_images, generator), None, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
+  except Exception as e:
+    return None, error_str(e), gr.update(visible=False), gr.update(visible=False), gr.update(visible=False)
+
+def inference_example(model_name, prompt, neg_prompt, guidance, steps, width=512, height=512):
+    _image, _error, _bool1, _bool2, _bool3 = inference(model_name, prompt, neg_prompt, guidance, steps, width, height)
+    return _image
+
+
+def txt_to_img(model_path, prompt, neg_prompt, guidance, steps, width, height, n_images, generator):
 
-    global current_model
+    print(f"{datetime.datetime.now()} txt_to_img, model: {current_model.name}")
+
+    global last_mode
     global pipe
-    if model != current_model:
-        current_model = model
-        pipe = StableDiffusionPipeline.from_pretrained(current_model, torch_dtype=torchfloat)
+    global current_model_path
+    if model_path != current_model_path or last_mode != "txt2img":
+        current_model_path = model_path
+
+        if is_colab or current_model == custom_model:
+          pipe = StableDiffusionPipeline.from_pretrained(
+              current_model_path,
+              torch_dtype=torch.float16,
+              scheduler=DPMSolverMultistepScheduler.from_pretrained(current_model.path, subfolder="scheduler"),
+              safety_checker=lambda images, clip_input: (images, False)
+              )
+        else:
+          pipe = StableDiffusionPipeline.from_pretrained(
+              current_model_path,
+              torch_dtype=torch.float16,
+              scheduler=DPMSolverMultistepScheduler.from_pretrained(current_model.path, subfolder="scheduler")
+              )
+          # pipe = pipe.to("cpu")
+          # pipe = current_model.pipe_t2i
+
         if torch.cuda.is_available():
-            pipe = pipe.to("cuda")
-
-def inference(prompt, negPrompt, guidance, steps, width_input, height_input):
-    print("Generated image with prompt: " + prompt)
-    if negPrompt != "":
-        print("Negative prompt: " + negPrompt)
-    promptPrev = prompt
-    prompt = prompt_prefixes[current_model] + prompt
-    results = pipe(prompt, negative_prompt = negPrompt, num_inference_steps=int(steps), guidance_scale=guidance, width=width_input, height=height_input)
-    image = results.images[0] if not results.nsfw_content_detected[0] else Image.open("nsfw_placeholder.jpg")
-    return image, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
-
-def inference_example(prompt, negPrompt, guidance, steps, width_input, height_input):
+          pipe = pipe.to("cuda")
+          pipe.enable_xformers_memory_efficient_attention()
+        last_mode = "txt2img"
+
+    prompt = current_model.prefix + prompt  
+    result = pipe(
+      prompt,
+      negative_prompt = neg_prompt,
+      num_images_per_prompt=n_images,
+      num_inference_steps = int(steps),
+      guidance_scale = guidance,
+      width = width,
+      height = height,
+      generator = generator)
     
-    prompt = prompt_prefixes[current_model] + prompt
-    results = pipe(prompt, negative_prompt = negPrompt, num_inference_steps=int(steps), guidance_scale=guidance, width=width_input, height=height_input)
-    image = results.images[0] if not results.nsfw_content_detected[0] else Image.open("nsfw_placeholder.jpg")
-    return image
+    return replace_nsfw_images(result)
+
+def img_to_img(model_path, prompt, neg_prompt, img, strength, guidance, steps, width, height, n_images, generator):
+
+    print(f"{datetime.datetime.now()} img_to_img, model: {model_path}")
+
+    global last_mode
+    global pipe
+    global current_model_path
+    if model_path != current_model_path or last_mode != "img2img":
+        current_model_path = model_path
+
+        if is_colab or current_model == custom_model:
+          pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
+              current_model_path,
+              torch_dtype=torch.float16,
+              scheduler=DPMSolverMultistepScheduler.from_pretrained(current_model.path, subfolder="scheduler"),
+              safety_checker=lambda images, clip_input: (images, False)
+              )
+        else:
+          pipe = StableDiffusionImg2ImgPipeline.from_pretrained(
+              current_model_path,
+              torch_dtype=torch.float16,
+              scheduler=DPMSolverMultistepScheduler.from_pretrained(current_model.path, subfolder="scheduler")
+              )
+          # pipe = pipe.to("cpu")
+          # pipe = current_model.pipe_i2i
+        
+        if torch.cuda.is_available():
+          pipe = pipe.to("cuda")
+          pipe.enable_xformers_memory_efficient_attention()
+        last_mode = "img2img"
+
+    prompt = current_model.prefix + prompt
+    ratio = min(height / img.height, width / img.width)
+    img = img.resize((int(img.width * ratio), int(img.height * ratio)), Image.LANCZOS)
+    result = pipe(
+        prompt,
+        negative_prompt = neg_prompt,
+        num_images_per_prompt=n_images,
+        image = img,
+        num_inference_steps = int(steps),
+        strength = strength,
+        guidance_scale = guidance,
+        # width = width,
+        # height = height,
+        generator = generator)
+        
+    return replace_nsfw_images(result)
+
+def replace_nsfw_images(results):
+    #Only return 1 image for community sharing
+    if is_colab:
+      return results.images[0]
+      
+    for i in range(len(results.images)):
+      if results.nsfw_content_detected[i]:
+        results.images[i] = Image.open("nsfw_placeholder.jpg")
+    return results.images[0]
+
 
 css = """
 #col-container {max-width: 700px; margin-left: auto; margin-right: auto;}
@@ -90,6 +248,7 @@ a {text-decoration-line: underline; font-weight: 600;}
 #share-btn-container .wrap {
     display: none !important;
 }
+#gallery{min-height:20rem}
 """
 
 with gr.Blocks(css=css) as demo:
@@ -105,7 +264,7 @@ with gr.Blocks(css=css) as demo:
                 "
               >
                 <h1 style="font-weight: 900; margin-bottom: 7px;">
-                  DGS Diffusion Space
+                  DGSpitzer Diffusion Space
                 </h1>
               </div>
               <p style="margin-bottom: 10px; font-size: 94%">
@@ -129,37 +288,68 @@ with gr.Blocks(css=css) as demo:
     with gr.Row():
         
         with gr.Column():
-            model = gr.Dropdown(label="Model", choices=models, value=models[0])
-            prompt = gr.Textbox(label="Prompt", placeholder="{} is added automatically".format(prompt_prefixes[current_model]), elem_id="input-prompt")
-            negPrompt = gr.Textbox(label="Negative Prompt", placeholder="Enter what you don't want to generate", elem_id="input-negPrompt")
-            guidance = gr.Slider(label="Guidance scale", value=7, maximum=8)
-            steps = gr.Slider(label="Steps", value=20, maximum=30, minimum=2)
+            model_name = gr.Dropdown(label="Model", choices=[m.name for m in models], value=current_model.name)
+            #prompt = gr.Textbox(label="Prompt", show_label=False, max_lines=2, placeholder="Enter your prompt~", elem_id="input-prompt").style(container=False)
+            prompt = gr.Textbox(label="Prompt", show_label=False, max_lines=2, placeholder="Enter your prompt~", elem_id="input-prompt")
+            neg_prompt = gr.Textbox(label="Negative Prompt", placeholder="Enter what you don't want to generate", elem_id="input-neg-prompt")
             width_input = gr.Slider(label="Width", value=576, maximum=768, minimum=384, step=64)
             height_input = gr.Slider(label="Height", value=576, maximum=768, minimum=384, step=64)
+            #n_images = gr.Slider(label="Images", value=1, minimum=1, maximum=4, step=1)
+            guidance = gr.Slider(label="Guidance scale", value=7, maximum=10)
+            steps = gr.Slider(label="Steps", value=20, maximum=50, minimum=2)
+            seed = gr.Slider(0, 2147483647, label='Seed (0 = random)', value=0, step=1)
             run = gr.Button(value="Run")
             gr.Markdown(f"Running on: {device}")
+            
+            # with gr.Column(scale=45):
+            #     with gr.Tab("Text to Image"):
+            #         with gr.Group():
+            #             gr.Markdown(f"Current in text to Image Mode, ")
+            #             #seed = gr.Slider(0, 2147483647, label='Seed (0 = random)', value=0, step=1)
+                
+            #     with gr.Tab("Image to Image"):
+            #         with gr.Group():
+            #             gr.Markdown(f"Image to Image Mode")
+            #             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)
+                      
         with gr.Column():
             image_out = gr.Image(height=512, type="filepath", elem_id="output-img")
+            error_output = gr.Markdown()
+            #gallery = gr.Gallery(label="Generated images", show_label=False, elem_id="gallery").style(grid=[2], height="auto")
+        
 
     with gr.Column(elem_id="col-container"):
         with gr.Group(elem_id="share-btn-container"):
           community_icon = gr.HTML(community_icon_html, visible=False)
           loading_icon = gr.HTML(loading_icon_html, visible=False)
           share_button = gr.Button("Share to community", elem_id="share-btn", visible=False)
-          
-    model.change(on_model_change, inputs=model, outputs=[])
-    run.click(inference, inputs=[prompt, negPrompt, guidance, steps, width_input, height_input], outputs=[image_out, share_button, community_icon, loading_icon])
+    
+    if is_colab:
+      model_name.change(on_model_change, inputs=model_name, outputs=[custom_model_group, prompt], queue=False)
+      custom_model_path.change(custom_model_changed, inputs=custom_model_path, outputs=None)
+    
+    #outputs = [gallery, error_output]
+    run.click(inference, inputs=[model_name, prompt, neg_prompt, guidance, steps, width_input, height_input, seed], outputs=[image_out, error_output, share_button, community_icon, loading_icon])
     
     share_button.click(None, [], [], _js=share_js)
     
     gr.Examples([
-        ["perfect face portrait of beautiful smile girl,  clean face, wears hoody, half body, soldier working in a cyberpunk city, cleavage, intricate, 8k, highly detailed, digital painting, intense, sharp focus", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 512, 704],
-        ["portrait of a beautiful fancy gorgeous anime girl, intricate details", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 448, 640],
-        ["a beautiful perfect face girl, Anime fine details portrait of school girl in front of modern tokyo city landscape on the background deep bokeh, anime masterpiece by studio ghibli, 8k, sharp high quality anime, artstation", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 704, 704],
-        ["city landscape with fancy car, racing on the road, gopro, intricate details, 4k, cyberpunk", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 704, 704],
-        ["portrait of liu yifei girl, soldier working in a cyberpunk city, cleavage, intricate, 8k, highly detailed, digital painting, intense, sharp focus", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 704, 704],
-        ["portrait of a muscular beard male in dgs illustration style, half-body, holding robot arms, strong chest", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 512, 640],
-    ], [prompt, negPrompt, guidance, steps, width_input, height_input], image_out, inference_example, cache_examples=torch.cuda.is_available())
+        [models[0].name, "perfect face portrait of beautiful smile girl,  clean face, wears hoody, half body, soldier working in a cyberpunk city, cleavage, intricate, 8k, highly detailed, digital painting, intense, sharp focus", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 512, 704],
+        [models[0].name, "portrait of a beautiful fancy gorgeous anime girl, intricate details", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 448, 640],
+        [models[0].name, "a beautiful perfect face girl, Anime fine details portrait of school girl in front of modern tokyo city landscape on the background deep bokeh, anime masterpiece by studio ghibli, 8k, sharp high quality anime, artstation", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 704, 704],
+        [models[0].name, "city landscape with fancy car, racing on the road, gopro, intricate details, 4k, cyberpunk", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 704, 704],
+        [models[0].name, "portrait of liu yifei girl, soldier working in a cyberpunk city, cleavage, intricate, 8k, highly detailed, digital painting, intense, sharp focus", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 704, 704],
+        [models[0].name, "portrait of a muscular beard male in dgs illustration style, half-body, holding robot arms, strong chest", "out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 512, 640],
+        [models[1].name, "Portrait of guanyu walking in ancient battlefield, close up shot", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 576, 576],
+        [models[1].name, "Portrait of a man holding very cute panda plushie, Guanyu", "", 7, 30, 576, 576],
+        [models[1].name, "taking selfie on ancient battlefield of China, guanyu, gopro, sharp focus, old scratched photo, three kingdoms warriors everywhere, masterpiece", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 576, 576],
+        [models[1].name, "portrait of a superman stands in a cyberpunk city,  guanyu", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 512, 704],
+        [models[1].name, "portrait of batman guanyu with armor, close up shot, city night background", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 576, 576],
+        [models[1].name, "a gorgeous wuxia girl standing in the palace", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 576, 576],
+        [models[1].name, "a handsome wuxia man", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 30, 576, 576],
+        [models[1].name, "a wuxia girl diving underwater, surrounded by a shark", "blurry, out of focus, scary, creepy, evil, disfigured, missing limbs, ugly, gross, missing fingers", 7, 20, 576, 576],
+    ], [model_name, prompt, neg_prompt, guidance, steps, width_input, height_input], outputs=image_out, fn=inference_example, cache_examples=torch.cuda.is_available())
     gr.Markdown('''
       Models and Space by [@DGSpitzer](https://www.youtube.com/channel/UCzzsYBF4qwtMwJaPJZ5SuPg)❤️ [@大谷的游戏创作小屋](https://space.bilibili.com/176003)
       [![Twitter Follow](https://img.shields.io/twitter/follow/DGSpitzer?label=%40DGSpitzer&style=social)](https://twitter.com/DGSpitzer)
@@ -169,5 +359,6 @@ with gr.Blocks(css=css) as demo:
       
     ''')
 
-demo.queue()
-demo.launch()
+if not is_colab:
+  demo.queue(concurrency_count=1)
+demo.launch(debug=is_colab, share=is_colab)