app.py

# Thank AK. https://huggingface.co/spaces/akhaliq/cool-japan-diffusion-2-1-0/blob/main/app.py
from diffusers import StableDiffusionPipeline, StableDiffusionImg2ImgPipeline, EulerAncestralDiscreteScheduler,StableDiffusionLatentUpscalePipeline
from transformers import CLIPImageProcessor
import gradio as gr
import torch
from PIL import Image
import random

import requests

# 得られたIPアドレスを使用してリクエストを行う
response = requests.get('https://huggingface.co/')
print(response.text)

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)

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)
# b1: 1.4, b2: 1.6, s1: 0.9, s2: 0.2
pipe_i2i = StableDiffusionImg2ImgPipeline.from_pretrained(
  model_id,
  torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
  scheduler=scheduler,
  requires_safety_checker=False,
  safety_checker=None,
  feature_extractor=feature_extractor
)
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()


if torch.cuda.is_available():
  pipe = pipe.to("cuda")
  pipe_i2i = pipe_i2i.to("cuda")

def error_str(error, title="Error"):
    return f"""#### {title}
            {error}"""  if error else ""


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

  prompt,neg_prompt=auto_prompt_correction(prompt,neg_prompt,cool_japan_type,disable_auto_prompt_correction)

  if(image_size=="Portrait"):
      height=1024
      width=768
      superreso=False
      #pipe.enable_attention_slicing()
  elif(image_size=="Landscape"):
      height=768
      width=1024  
      superreso=False
      #pipe.enable_attention_slicing()
  elif(image_size=="Highreso."):
      height=1024
      width=1024
      superreso=False
      #pipe.enable_attention_slicing()
  elif(image_size=="Superreso."):
      height=1024
      width=1024
      superreso=True
      #pipe.enable_attention_slicing()
  else:
      height=768
      width=768
      superreso=False
      #pipe.enable_attention_slicing()

  print(prompt,neg_prompt)
    
  try:
    if img is not None:
      return img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height, generator,superreso), None
    else:
      return txt_to_img(prompt, neg_prompt, guidance, steps, width, height, generator,superreso), None
  except Exception as e:
    return None, error_str(e)
def auto_prompt_correction(prompt_ui,neg_prompt_ui,cool_japan_type_ui,disable_auto_prompt_correction):
    # auto prompt correction
    cool_japan_type=str(cool_japan_type_ui)
    if(cool_japan_type=="Manga"):
        cool_japan_type="manga, monochrome, white and black manga"
    elif(cool_japan_type=="Game"):
        cool_japan_type="game"
    else:
        cool_japan_type="anime"
        
    prompt=str(prompt_ui)
    neg_prompt=str(neg_prompt_ui)
    prompt=prompt.lower()
    neg_prompt=neg_prompt.lower()
    if(disable_auto_prompt_correction):
        prompt=f"{cool_japan_type}, {prompt}"
        return prompt, neg_prompt

    if(prompt=="" and neg_prompt==""):
        #prefix=["masterpiece","evangelion, mika pikazo", "konosuba, mika pikazo","steins; gate, ilya kuvshinov",
        #        "ghibli, shinkai makoto", "evangelion, madoka magica"]
        #suffix=["","ayanami rei, asuka langrey", "megumin, aqua from konosuba","mayuri shiina from steins gate, kurisu makise steins gate anime",
        #        "hakurei reimu","kirisame marisa", "kaname madoka, megumin"]
        #prefix_index=random.randrange(len(prefix))
        #suffix_index=random.randrange(len(suffix))
        #prompt=f"{cool_japan_type}, {prefix[prefix_index]}, portrait, a good girl, {suffix[suffix_index]}, good pupil, 4k, detailed"
        prompt=f"{cool_japan_type}, masterpiece, upper body, a girl, good pupil, 4k, detailed"
        neg_prompt=f"(((deformed))), blurry, ((((bad anatomy)))), bad pupil, disfigured, poorly drawn face, mutation, mutated, (extra limb), (ugly), (poorly drawn hands), bad hands, fused fingers, messy drawing, broken legs censor, low quality, (mutated hands and fingers:1.5), (long body :1.3), (mutation, poorly drawn :1.2), ((bad eyes)), ui, error, missing fingers, fused fingers, one hand with more than 5 fingers, one hand with less than 5 fingers, one hand with more than 5 digit, one hand with less than 5 digit, extra digit, fewer digits, fused digit, missing digit, bad digit, liquid digit, long body, uncoordinated body, unnatural body, lowres, jpeg artifacts, 3d, cg, text"
        return prompt, neg_prompt
        
    splited_prompt=prompt.replace(","," ").replace("_"," ").split(" ")
    splited_prompt=["a person" if p=="solo" else p for p in splited_prompt]
    splited_prompt=["girl" if p=="1girl" else p for p in splited_prompt]
    splited_prompt=["a couple of girls" if p=="2girls" else p for p in splited_prompt]
    splited_prompt=["a couple of boys" if p=="2boys" else p for p in splited_prompt]
    human_words=["girl","maid","maids","female","woman","girls","a couple of girls","women","boy","boys","a couple of boys","male","man","men","guy","guys"]
    for word in human_words:
        if( word in splited_prompt):
            prompt=f"{cool_japan_type}, masterpiece, {prompt}, good pupil, 4k, detailed"
            neg_prompt=f"(((deformed))), blurry, ((((bad anatomy)))), {neg_prompt}, bad pupil, disfigured, poorly drawn face, mutation, mutated, (extra limb), (ugly), (poorly drawn hands), bad hands, fused fingers, messy drawing, broken legs censor, low quality, (mutated hands and fingers:1.5), (long body :1.3), (mutation, poorly drawn :1.2), ((bad eyes)), ui, error, missing fingers, fused fingers, one hand with more than 5 fingers, one hand with less than 5 fingers, one hand with more than 5 digit, one hand with less than 5 digit, extra digit, fewer digits, fused digit, missing digit, bad digit, liquid digit, long body, uncoordinated body, unnatural body, lowres, jpeg artifacts, 3d, cg, text"

    animal_words=["cat","dog","bird","horse","pigeon"]
    for word in animal_words:
        if( word in splited_prompt):
            prompt=f"{cool_japan_type}, a {word}, 4k, detailed"
            neg_prompt=f"(((deformed))), blurry, ((((bad anatomy)))), {neg_prompt}, bad pupil, disfigured, poorly drawn face, mutation, mutated, (extra limb), (ugly), (poorly drawn hands), bad hands, fused fingers, messy drawing, broken legs censor, low quality, (mutated hands and fingers:1.5), (long body :1.3), (mutation, poorly drawn :1.2), ((bad eyes)), ui, error, missing fingers, fused fingers, one hand with more than 5 fingers, one hand with less than 5 fingers, one hand with more than 5 digit, one hand with less than 5 digit, extra digit, fewer digits, fused digit, missing digit, bad digit, liquid digit, long body, uncoordinated body, unnatural body, lowres, jpeg artifacts, 3d, cg, text"

    background_words=["mount fuji","mt. fuji","building", "buildings", "tokyo", "kyoto", "nara", "shibuya", "shinjuku"]
    for word in background_words:
        if( word in splited_prompt):
            prompt=f"{cool_japan_type}, shinkai makoto, {word}, 4k, 8k, highly detailed"
            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):
      low_res_latents = pipe(
        prompt,
        negative_prompt = neg_prompt,
        num_inference_steps = int(steps),
        guidance_scale = guidance,
        width = width,
        height = height,
        output_type="latent",
        generator = generator).images
      pipe=pipe.to("cpu")
      upscaler=upscaler.to("cuda")
      result = upscaler(
        prompt=prompt,
        negative_prompt = neg_prompt,
        image=low_res_latents,
        num_inference_steps=20,
        guidance_scale=0,
        generator=generator,
      )
      pipe=pipe.to("cuda")
      upscaler=upscaler.to("cpu")
    else:  
      result = pipe(
        prompt,
        negative_prompt = neg_prompt,
        num_inference_steps = int(steps),
        guidance_scale = guidance,
        width = width,
        height = height,
        generator = generator)
    
    return result.images[0]

def img_to_img(prompt, neg_prompt, img, strength, guidance, steps, width, height, generator,superreso=False):
    ratio = min(height / img.height, width / img.width)
    img = img.resize((int(img.width * ratio), int(img.height * ratio)), Image.LANCZOS)
    if(superreso):
      low_res_latents = pipe_i2i(
        prompt,
        negative_prompt = neg_prompt,
        image = img,
        num_inference_steps = int(steps),
        strength = strength,
        guidance_scale = guidance,
        #width = width,
        #height = height,
        output_type="latent",
        generator = generator).images
      pipe=pipe.to("cpu")
      upscaler=upscaler.to("cuda")
      result = upscaler(
        prompt=prompt,
        negative_prompt = neg_prompt,
        image=low_res_latents,
        num_inference_steps=20,
        guidance_scale=0,
        generator=generator,
      )
      pipe=pipe.to("cuda")
      upscaler=upscaler.to("cpu")
    else:  
      result = pipe_i2i(
        prompt,
        negative_prompt = neg_prompt,
        image = img,
        num_inference_steps = int(steps),
        strength = strength,
        guidance_scale = guidance,
        #width = width,
        #height = height,
        generator = generator)

        
    return result.images[0]

css = """.main-div div{display:inline-flex;align-items:center;gap:.8rem;font-size:1.75rem}.main-div div h1{font-weight:900;margin-bottom:7px}.main-div p{margin-bottom:10px;font-size:94%}a{text-decoration:underline}.tabs{margin-top:0;margin-bottom:0}#gallery{min-height:20rem}
"""
with gr.Blocks(css=css) as demo:
    gr.HTML(
        f"""
            <div class="main-div">
              <div>
                <h1>Cool Japan Diffusion 2.1.2</h1>
              </div>
              <p>
               Demo for <a href="https://huggingface.co/aipicasso/cool-japan-diffusion-2-1-2">Cool Japan Diffusion 2.1.2</a> .<br>
              </p>
              <p>
              sample : Click "Generate" button without any prompts.
              </p>
              <p>
              sample prompt1 : girl, kimono
              </p>
              <p>
              sample prompt2 : boy, school uniform
              </p>
              Running on {"<b>GPU 🔥</b>" if torch.cuda.is_available() else f"<b>CPU 🥶</b>. For faster inference it is recommended to <b>upgrade to GPU in <a href='https://huggingface.co/spaces/akhaliq/cool-japan-diffusion-2-1-0/settings'>Settings</a></b>"} <br>
              <a style="display:inline-block" href="https://huggingface.co/spaces/aipicasso/cool-japan-diffusion-latest-demo?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate Space"></a> to say goodbye from waiting for the generating.
            </div>
        """
    )
    with gr.Row():
        
        with gr.Column(scale=55):
          with gr.Group():
              with gr.Row():
                cool_japan_type=gr.Radio(["Anime", "Manga", "Game"])
                cool_japan_type.show_label=False
                cool_japan_type.value="Anime"
                
              with gr.Row():
                prompt = gr.Textbox(label="Prompt", show_label=False, max_lines=2,placeholder="[your prompt]") # .style(container=False)
                generate = gr.Button(value="Generate") # .style(rounded=(False, True, True, False))

              image_out = gr.Image(height=768,width=768)
          error_output = gr.Markdown()

        with gr.Column(scale=45):
          with gr.Tab("Options"):
            with gr.Group():
              neg_prompt = gr.Textbox(label="Negative prompt", placeholder="What to exclude from the image")
              disable_auto_prompt_correction = gr.Checkbox(label="Disable auto prompt corretion.")
              with gr.Row():
                image_size=gr.Radio(["Portrait","Landscape","Square","Highreso.","Superreso."])
                image_size.show_label=False
                image_size.value="Square"
                
              with gr.Row():
                guidance = gr.Slider(label="Guidance scale", value=7.5, maximum=15)
                steps = gr.Slider(label="Steps", value=20, minimum=2, maximum=75, step=1)

              seed = gr.Slider(0, 2147483647, label='Seed (0 = random)', value=0, step=1)

          with gr.Tab("Image to image"):
              with gr.Group():
                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)
                  
    inputs = [prompt, guidance, steps, image_size, seed, image, strength, neg_prompt, cool_japan_type, disable_auto_prompt_correction]

    outputs = [image_out, error_output]
    prompt.submit(inference, inputs=inputs, outputs=outputs)
    generate.click(inference, inputs=inputs, outputs=outputs,api_name="generate")

demo.queue(concurrency_count=1)
demo.launch()