app.py

#!/usr/bin/env python
import os
import random
import uuid
import gradio as gr
import numpy as np
from PIL import Image
import spaces
import torch
from diffusers import StableDiffusionXLPipeline, EulerAncestralDiscreteScheduler

DESCRIPTIONx = """


## TEXT 2 IMAGE PLAYGROUND 🥠


"""

css = '''
.gradio-container{max-width: 690px !important}
h1{text-align:center}
footer {
    visibility: hidden
}
'''

js_func = """
function refresh() {
    const url = new URL(window.location);
    if (url.searchParams.get('__theme') !== 'dark') {
        url.searchParams.set('__theme', 'dark');
        window.location.href = url.href;
    }
}
"""

examples = [
    "3d image, cute girl, in the style of Pixar --ar 1:2 --stylize 750, 4K resolution highlights, Sharp focus, octane render, ray tracing, Ultra-High-Definition, 8k, UHD, HDR, (Masterpiece:1.5), (best quality:1.5)",
    "1girl, makima \(chainsaw man\), chainsaw man, black jacket, black necktie, black pants, braid, business suit, fingernails, formal, hand on own chin, jacket on shoulders, light smile, long sleeves, looking at viewer, looking up, medium breasts, office lady, smile, solo, suit, upper body, white shirt, outdoors",
    "1boy, male focus, gojou satoru, jujutsu kaisen, black jacket, blindfold lift, blue eyes, glowing, glowing eyes, high collar, jacket, jujutsu tech uniform, solo, grin, white hair",
    "Man in brown leather jacket posing for camera, in the style of sleek and stylized, clockpunk, subtle shades, exacting precision, ferrania p30  --ar 67:101 --v 5",
    "Commercial photography, giant burger, white lighting, studio light, 8k octane rendering, high resolution photography, insanely detailed, fine details, on white isolated plain, 8k, commercial photography, stock photo, professional color grading, --v 4 --ar 9:16 "
]

MODEL_OPTIONS = {
    "Realism : V4.0_Lightning": "SG161222/RealVisXL_V4.0_Lightning",
    "Anime : Cagliostrolab": "cagliostrolab/animagine-xl-3.1"
}

MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "4096"))
USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "0") == "1"
ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"
BATCH_SIZE = int(os.getenv("BATCH_SIZE", "1"))

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

def load_and_prepare_model(model_id):
    pipe = StableDiffusionXLPipeline.from_pretrained(
        model_id,
        torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
        use_safetensors=True,
        add_watermarker=False,
    ).to(device)
    pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)
    
    if USE_TORCH_COMPILE:
        pipe.compile()
    
    if ENABLE_CPU_OFFLOAD:
        pipe.enable_model_cpu_offload()
    
    return pipe

# Preload and compile both models
models = {key: load_and_prepare_model(value) for key, value in MODEL_OPTIONS.items()}

MAX_SEED = np.iinfo(np.int32).max

def save_image(img):
    unique_name = str(uuid.uuid4()) + ".png"
    img.save(unique_name)
    return unique_name

def randomize_seed_fn(seed: int, randomize_seed: bool) -> int:
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    return seed

@spaces.GPU(duration=60, enable_queue=True)
def generate(
    model_choice: str,
    prompt: str,
    negative_prompt: str = "",
    use_negative_prompt: bool = False,
    seed: int = 1,
    width: int = 1024,
    height: int = 1024,
    guidance_scale: float = 3,
    num_inference_steps: int = 25,
    randomize_seed: bool = False,
    use_resolution_binning: bool = True, 
    num_images: int = 1,  
    progress=gr.Progress(track_tqdm=True),
):
    global models
    pipe = models[model_choice]
    
    seed = int(randomize_seed_fn(seed, randomize_seed))
    generator = torch.Generator(device=device).manual_seed(seed)

    options = {
        "prompt": [prompt] * num_images,
        "negative_prompt": [negative_prompt] * num_images if use_negative_prompt else None,
        "width": width,
        "height": height,
        "guidance_scale": guidance_scale,
        "num_inference_steps": num_inference_steps,
        "generator": generator,
        "output_type": "pil",
    }

    if use_resolution_binning:
        options["use_resolution_binning"] = True

    images = []
    for i in range(0, num_images, BATCH_SIZE):
        batch_options = options.copy()
        batch_options["prompt"] = options["prompt"][i:i+BATCH_SIZE]
        if "negative_prompt" in batch_options:
            batch_options["negative_prompt"] = options["negative_prompt"][i:i+BATCH_SIZE]
        images.extend(pipe(**batch_options).images)

    image_paths = [save_image(img) for img in images]
    return image_paths, seed

def load_predefined_images():
    predefined_images = [
        "assets/1.png",
        "assets/2.png",
        "assets/3.png",
        "assets/4.png",
        "assets/5.png",
        "assets/6.png",
        "assets/7.png",
        "assets/8.png",
        "assets/9.png",
        "assets/10.png",
        "assets/11.png",
        "assets/12.png",
    ]
    return predefined_images

with gr.Blocks(css=css, theme="bethecloud/storj_theme", js=js_func) as demo:
    gr.Markdown(DESCRIPTIONx)    
    with gr.Row():
        prompt = gr.Text(
            label="Prompt",
            show_label=False,
            max_lines=1,
            placeholder="Enter your prompt",
            container=False,
        )
        run_button = gr.Button("Run🚀", scale=0)
    result = gr.Gallery(label="Result", columns=1, show_label=False) 

    with gr.Row():
        model_choice = gr.Dropdown(
            label="Model Selection 🥠",
            choices=list(MODEL_OPTIONS.keys()),
            value="Realism : V4.0_Lightning"
        )

    with gr.Accordion("Advanced options", open=True):
        num_images = gr.Slider(
            label="Number of Images",
            minimum=1,
            maximum=4,
            step=1,
            value=1,
        )
        with gr.Row():
            with gr.Column(scale=1):
                use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=True)
                negative_prompt = gr.Text(
                    label="Negative prompt",
                    max_lines=5,
                    lines=4,
                    placeholder="Enter a negative prompt",
                    value="(deformed, distorted, disfigured:1.3), poorly drawn, bad anatomy, wrong anatomy, extra limb, missing limb, floating limbs, (mutated hands and fingers:1.4), disconnected limbs, mutation, mutated, ugly, disgusting, blurry, amputation",
                    visible=True,
                )
        seed = gr.Slider(
            label="Seed",
            minimum=0,
            maximum=MAX_SEED,
            step=1,
            value=0,
        )
        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
        with gr.Row(visible=True):
            width = gr.Slider(
                label="Width",
                minimum=512,
                maximum=MAX_IMAGE_SIZE,
                step=64,
                value=1024,
            )
            height = gr.Slider(
                label="Height",
                minimum=512,
                maximum=MAX_IMAGE_SIZE,
                step=64,
                value=1024,
            )
        with gr.Row():
            guidance_scale = gr.Slider(
                label="Guidance Scale",
                minimum=0.1,
                maximum=6,
                step=0.1,
                value=3.0,
            )
            num_inference_steps = gr.Slider(
                label="Number of inference steps",
                minimum=1,
                maximum=25,
                step=1,
                value=20,
            )

    gr.Examples(
        examples=examples,
        inputs=prompt,
        cache_examples=False
    )

    use_negative_prompt.change(
        fn=lambda x: gr.update(visible=x),
        inputs=use_negative_prompt,
        outputs=negative_prompt,
        api_name=False,
    )
    
    gr.on(
        triggers=[
            prompt.submit,
            negative_prompt.submit,
            run_button.click,
        ],
        fn=generate,
        inputs=[
            model_choice,
            prompt,
            negative_prompt,
            use_negative_prompt,
            seed,
            width,
            height,
            guidance_scale,
            num_inference_steps,
            randomize_seed,
            num_images
        ],
        outputs=[result, seed],
        api_name="run",
    )

    with gr.Column(scale=3):
        gr.Markdown("### Image Gallery")
        predefined_gallery = gr.Gallery(label="Image Gallery", columns=4, show_label=False, value=load_predefined_images())
           
if __name__ == "__main__":
    demo.queue(max_size=40).launch(show_api=False)