app.py

import os
import spaces
import torch
from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler
import gradio as gr
import random
import tqdm

# Enable TQDM progress tracking
tqdm.monitor_interval = 0

# Load the diffusion pipeline
pipe = StableDiffusionXLPipeline.from_pretrained(
    "kayfahaarukku/UrangDiffusion-1.0", 
    torch_dtype=torch.float16, 
    custom_pipeline="lpw_stable_diffusion_xl",
)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)

# Function to generate an image
@spaces.GPU  # Adjust the duration as needed
def generate_image(prompt, negative_prompt, use_defaults, resolution, guidance_scale, num_inference_steps, seed, randomize_seed, progress=gr.Progress()):
    pipe.to('cuda')  # Move the model to GPU when the function is called
    
    if randomize_seed:
        seed = random.randint(0, 99999999)
    if use_defaults:
        prompt = f"{prompt}, masterpiece, best quality"
        negative_prompt = f"nsfw, lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name, {negative_prompt}"
    generator = torch.manual_seed(seed)
    
    def callback(step, timestep, latents):
        progress(step / num_inference_steps)
        return
    
    width, height = map(int, resolution.split('x'))
    image = pipe(
        prompt, 
        negative_prompt=negative_prompt,
        width=width,
        height=height, 
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        generator=generator,
        callback=callback,
        callback_steps=1
    ).images[0]

    torch.cuda.empty_cache()

    return image, seed, prompt, negative_prompt, guidance_scale, num_inference_steps, resolution

# Define Gradio interface
def interface_fn(prompt, negative_prompt, use_defaults, resolution, guidance_scale, num_inference_steps, seed, randomize_seed, progress=gr.Progress()):
    image, seed, prompt, negative_prompt, guidance_scale, num_inference_steps, resolution = generate_image(
        prompt, negative_prompt, use_defaults, resolution, guidance_scale, num_inference_steps, seed, randomize_seed, progress
    )
    
    generation_details = (
        f"{prompt}\n"
        f"Negative Prompt: {negative_prompt}\n"
        f"Steps: {num_inference_steps}\n"
        f"Sampler: Euler a\n"
        f"CFG scale: {guidance_scale}\n"
        f"Seed: {seed}\n"
        f"Size: {resolution}"
    )
    
    return image, seed, gr.update(value=seed), gr.update(visible=True), generation_details

def reset_inputs():
    return (
        gr.update(value=''), 
        gr.update(value=''), 
        gr.update(value=True), 
        gr.update(value='832x1216'), 
        gr.update(value=7), 
        gr.update(value=28), 
        gr.update(value=0), 
        gr.update(value=True),
        gr.update(visible=False),
        ""
    )

with gr.Blocks(title="UrangDiffusion 1.0 Demo", theme="NoCrypt/[email protected]") as demo:
    gr.HTML(
        "<h1>UrangDiffusion 1.0 Demo</h1>"
        "This demo is intended to showcase what the model is capable of and is not intended to be the main generation platform. Results produced with Diffusers are not the best, and it's highly recommended for you to get the model running inside Stable Diffusion WebUI or ComfyUI."
        )
    with gr.Row():
        with gr.Column():
            prompt_input = gr.Textbox(lines=2, placeholder="Enter prompt here", label="Prompt")
            negative_prompt_input = gr.Textbox(lines=2, placeholder="Enter negative prompt here", label="Negative Prompt")
            use_defaults_input = gr.Checkbox(label="Use Default Quality Tags and Negative Prompt", value=True)
            resolution_input = gr.Radio(
                choices=[
                    "1024x1024", "1152x896", "896x1152", "1216x832", "832x1216",
                    "1344x768", "768x1344", "1536x640", "640x1536"
                ],
                label="Resolution",
                value="832x1216"
            )
            guidance_scale_input = gr.Slider(minimum=1, maximum=20, step=0.5, label="Guidance Scale", value=7)
            num_inference_steps_input = gr.Slider(minimum=1, maximum=100, step=1, label="Number of Inference Steps", value=28)
            seed_input = gr.Slider(minimum=0, maximum=99999999, step=1, label="Seed", value=0, interactive=True)
            randomize_seed_input = gr.Checkbox(label="Randomize Seed", value=True)
            generate_button = gr.Button("Generate")
            reset_button = gr.Button("Reset")

        with gr.Column():
            output_image = gr.Image(type="pil", label="Generated Image")
            generation_details_output = gr.Markdown("", visible=False)
            gr.Markdown(
                """
                ### Recommended prompt formatting:
                `1girl/1boy, character name, from what series, everything else in any order, masterpiece, best quality`

                **PS:** `masterpiece, best quality` is automatically added when "Use Default Quality Tags and Negative Prompt" is enabled

                ### Recommended settings:
                - Steps: 25-30
                - CFG: 5-7
                """
            )

    generate_button.click(
        interface_fn,
        inputs=[
            prompt_input, negative_prompt_input, use_defaults_input, resolution_input, guidance_scale_input, num_inference_steps_input, seed_input, randomize_seed_input
        ],
        outputs=[output_image, seed_input, gr.update(value=seed), generation_details_output]
    )
    
    reset_button.click(
        reset_inputs,
        inputs=[],
        outputs=[
            prompt_input, negative_prompt_input, use_defaults_input, resolution_input, guidance_scale_input, num_inference_steps_input, seed_input, randomize_seed_input, generation_details_output, gr.update(value="")
        ]
    )

demo.queue(max_size=20).launch(share=False)