main_app.py

import os
import random
import uuid
import json

import gradio as gr
import numpy as np
from PIL import Image
import spaces
import torch
from diffusers import DiffusionPipeline
from typing import Tuple

# Check for the Model Base

bad_words = json.loads(os.getenv('BAD_WORDS', "[]"))
bad_words_negative = json.loads(os.getenv('BAD_WORDS_NEGATIVE', "[]"))
default_negative = os.getenv("default_negative","")

def check_text(prompt, negative=""):
    for i in bad_words:
        if i in prompt:
            return True
    for i in bad_words_negative:
        if i in negative:
            return True
    return False

style_list = [
    {
        "name": "3840 x 2160",
        "prompt": "hyper-realistic 8K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic",
        "negative_prompt": "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly",
    },
    {
        "name": "2560 x 1440",
        "prompt": "hyper-realistic 4K image of {prompt}. ultra-detailed, lifelike, high-resolution, sharp, vibrant colors, photorealistic",
        "negative_prompt": "cartoonish, low resolution, blurry, simplistic, abstract, deformed, ugly",
    },
    {
        "name": "Photo",
        "prompt": "cinematic photo {prompt}. 35mm photograph, film, bokeh, professional, 4k, highly detailed",
        "negative_prompt": "drawing, painting, crayon, sketch, graphite, impressionist, noisy, blurry, soft, deformed, ugly",
    },
    {
        "name": "Cinematic",
        "prompt": "cinematic still {prompt}. emotional, harmonious, vignette, highly detailed, high budget, bokeh, cinemascope, moody, epic, gorgeous, film grain, grainy",
        "negative_prompt": "anime, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured",
    },
    {
        "name": "Anime",
        "prompt": "anime artwork {prompt}. anime style, key visual, vibrant, studio anime, highly detailed",
        "negative_prompt": "photo, deformed, black and white, realism, disfigured, low contrast",
    },
    {
        "name": "3D Model",
        "prompt": "professional 3d model {prompt}. octane render, highly detailed, volumetric, dramatic lighting",
        "negative_prompt": "ugly, deformed, noisy, low poly, blurry, painting",
    },
    {
        "name": "(No style)",
        "prompt": "{prompt}",
        "negative_prompt": "",
    },
]

styles = {k["name"]: (k["prompt"], k["negative_prompt"]) for k in style_list}
STYLE_NAMES = list(styles.keys())
DEFAULT_STYLE_NAME = "3840 x 2160"

def apply_style(style_name: str, positive: str, negative: str = "") -> Tuple[str, str]:
    p, n = styles.get(style_name, styles[DEFAULT_STYLE_NAME])
    if not negative:
        negative = ""
    return p.replace("{prompt}", positive), n + negative

DESCRIPTIONs = """ㅤㅤㅤ """

DESCRIPTION = """ㅤㅤㅤ """

if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>⚠️Running on CPU, This may not work on CPU.</p>"

MAX_SEED = np.iinfo(np.int32).max
CACHE_EXAMPLES = torch.cuda.is_available() and os.getenv("CACHE_EXAMPLES", "0") == "1"
MAX_IMAGE_SIZE = int(os.getenv("MAX_IMAGE_SIZE", "2048"))
USE_TORCH_COMPILE = os.getenv("USE_TORCH_COMPILE", "0") == "1"
ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1"

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

NUM_IMAGES_PER_PROMPT = 1

if torch.cuda.is_available():
    pipe = DiffusionPipeline.from_pretrained(
        "SG161222/RealVisXL_V4.0",
        torch_dtype=torch.float16,
        use_safetensors=True,
        add_watermarker=False,
        variant="fp16"
    )
    pipe2 = DiffusionPipeline.from_pretrained(
        "SG161222/RealVisXL_V3.0",
        torch_dtype=torch.float16,
        use_safetensors=True,
        add_watermarker=False,
        variant="fp16"
    )
    if ENABLE_CPU_OFFLOAD:
        pipe.enable_model_cpu_offload()
        pipe2.enable_model_cpu_offload()
    else:
        pipe.to(device)    
        pipe2.to(device)    
        print("Loaded on Device!")
    
    if USE_TORCH_COMPILE:
        pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True)
        pipe2.unet = torch.compile(pipe2.unet, mode="reduce-overhead", fullgraph=True)
        print("Model Compiled!")
else:
    pipe = DiffusionPipeline.from_pretrained(
        "SG161222/RealVisXL_V4.0",
        torch_dtype=torch.float32,
        use_safetensors=True,
        add_watermarker=False
    )
    pipe2 = DiffusionPipeline.from_pretrained(
        "SG161222/RealVisXL_V3.0",
        torch_dtype=torch.float32,
        use_safetensors=True,
        add_watermarker=False
    )
    pipe.to(device)
    pipe2.to(device)
    print("Loaded on Device!")

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()
def generate(
    prompt: str,
    negative_prompt: str = "",
    use_negative_prompt: bool = False,
    style: str = DEFAULT_STYLE_NAME,
    seed: int = 0,
    width: int = 1024,
    height: int = 1024,
    guidance_scale: float = 3,
    randomize_seed: bool = False,
    use_resolution_binning: bool = True,
    progress=gr.Progress(track_tqdm=True),
):
    if check_text(prompt, negative_prompt):
        raise ValueError("Prompt contains restricted words.")
    
    prompt, negative_prompt = apply_style(style, prompt, negative_prompt)
    seed = int(randomize_seed_fn(seed, randomize_seed))
    generator = torch.Generator().manual_seed(seed)

    if not use_negative_prompt:
        negative_prompt = ""  # type: ignore
    negative_prompt += default_negative

    options = {
        "prompt": prompt,
        "negative_prompt": negative_prompt,
        "width": width,
        "height": height,
        "guidance_scale": guidance_scale,
        "num_inference_steps": 25,
        "generator": generator,
        "num_images_per_prompt": NUM_IMAGES_PER_PROMPT,
        "use_resolution_binning": use_resolution_binning,
        "output_type": "pil",
    }
    
    images = pipe(**options).images + pipe2(**options).images

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

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)",
    "Silhouette of Hamburger standing in front of a, dark blue sky, a little saturated orange in the background sunset, night time, dark background, dark black hair, cinematic photography, cinematic lighting, dark theme, shattered camera lens, digital photography, 70mm, f2.8, lens aberration, grain, boke, double exposure, shaterred, color negative ",
    "A photograph of the front view portrait of an Cat in a full body dynamic pose on a red background in the style of high fashion moment, with rich colors, dramatic light, in a fantasy art style, with surrealism, elegant details, a golden ratio composition, and detailed texture",
    "Closeup of blonde woman depth of field, bokeh, shallow focus, minimalism, fujifilm xh2s with Canon EF lens, cinematic --ar 85:128 --v 6.0 --style raw"
]

css = '''
.gradio-container {
    max-width: 800px !important;
    margin: auto;
}
h1 {
    text-align: center;
    font-size: 2.5em;
    color: #4CAF50;
}
#duplicate-button {
    display: block;
    margin: 10px auto;
    padding: 10px 20px;
    background-color: #4CAF50;
    color: white;
    border: none;
    border-radius: 5px;
    cursor: pointer;
}
#duplicate-button:hover {
    background-color: #45a049;
}
.gradio-button {
    display: block;
    margin: 20px auto;
    padding: 10px 20px;
    background-color: #2196F3;
    color: white;
    border: none;
    border-radius: 5px;
    cursor: pointer;
}
.gradio-button:hover {
    background-color: #0b7dda;
}
.gradio-gallery {
    width: 100%;
    display: flex;
    flex-wrap: wrap;
    gap: 10px;
}
.gradio-gallery img {
    width: calc(50% - 10px);
    height: auto;
    border-radius: 5px;
}
.gradio-accordion, .gradio-slider, .gradio-radio {
    margin-bottom: 20px;
}
.gradio-checkbox {
    margin-top: 10px;
    margin-bottom: 10px;
}
.gradio-slider label, .gradio-checkbox label, .gradio-radio label {
    color: #3E4C59;
}
.gradio-examples {
    display: flex;
    flex-wrap: wrap;
    justify-content: space-between;
}
.gradio-examples > div {
    flex: 1;
    margin: 5px;
    padding: 10px;
    background-color: #f9f9f9;
    border: 1px solid #ddd;
    border-radius: 5px;
    cursor: pointer;
}
.gradio-examples > div:hover {
    background-color: #f1f1f1;
}
'''

with gr.Blocks(css=css, theme="default") as demo:
    gr.Markdown(DESCRIPTIONs)
    with gr.Group():
        with gr.Row():
            prompt = gr.Textbox(
                label="Prompt",
                show_label=True,
                max_lines=1,
                placeholder="Enter your prompt",
                container=True
            )
            run_button = gr.Button("Run")
        result = gr.Gallery(label="Result", columns=2, preview=True)
    with gr.Accordion("Advanced options", open=False):
        use_negative_prompt = gr.Checkbox(label="Use negative prompt", value=True, visible=True)
        negative_prompt = gr.Textbox(
            label="Negative prompt",
            max_lines=1,
            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
        )
        with gr.Row():
            num_inference_steps = gr.Slider(
                label="Steps",
                minimum=10,
                maximum=60,
                step=1,
                value=30
            )
        with gr.Row():
            num_images_per_prompt = gr.Slider(
                label="Images",
                minimum=1,
                maximum=5,
                step=1,
                value=2
            )
        seed = gr.Slider(
            label="Seed",
            minimum=0,
            maximum=MAX_SEED,
            step=1,
            value=0,
            visible=True
        )
        randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
        with gr.Row(visible=True):
            width = gr.Slider(
                label="Width",
                minimum=512,
                maximum=2048,
                step=8,
                value=1024
            )
            height = gr.Slider(
                label="Height",
                minimum=512,
                maximum=2048,
                step=8,
                value=1024
            )
        with gr.Row():
            guidance_scale = gr.Slider(
                label="Guidance Scale",
                minimum=0.1,
                maximum=20.0,
                step=0.1,
                value=6
            )
    with gr.Row(visible=True):
        style_selection = gr.Radio(
            show_label=True,
            container=True,
            interactive=True,
            choices=STYLE_NAMES,
            value=DEFAULT_STYLE_NAME,
            label="Image Style"
        )
    gr.Examples(
        examples=examples,
        inputs=prompt,
        outputs=[result, seed],
        fn=generate,
        cache_examples=CACHE_EXAMPLES
    )

    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=[
            prompt,
            negative_prompt,
            use_negative_prompt,
            style_selection,
            seed,
            width,
            height,
            guidance_scale,
            randomize_seed,
        ],
        outputs=[result, seed],
        api_name="run",
    )
    gr.Markdown(DESCRIPTION)

if __name__ == "__main__":
    demo.queue(max_size=20).launch()