app.py

import gradio as gr
import numpy as np
import random
import spaces
import torch
from diffusers import DiffusionPipeline, FlowMatchEulerDiscreteScheduler
from PIL import Image
import io
import os

dtype = torch.bfloat16
device = "cuda" if torch.cuda.is_available() else "cpu"

# Set your Hugging Face API token
huggingface_token = os.getenv("HUGGINGFACE_TOKEN")

# Load the diffusion pipeline with the Hugging Face API token
pipe = DiffusionPipeline.from_pretrained("black-forest-labs/FLUX.1-dev", torch_dtype=dtype, token=huggingface_token).to(device)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 2048

@spaces.GPU(duration=200)
def infer(prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=5.0, num_inference_steps=28, progress=gr.Progress(track_tqdm=True)):
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)
    generator = torch.Generator().manual_seed(seed)
    image = pipe(
        prompt=prompt,
        width=width,
        height=height,
        num_inference_steps=num_inference_steps,
        generator=generator,
        guidance_scale=guidance_scale
    ).images[0]
    return image, seed

def download_image(image, file_format):
    img_byte_arr = io.BytesIO()
    image.save(img_byte_arr, format=file_format)
    img_byte_arr = img_byte_arr.getvalue()
    return img_byte_arr

examples = [
    "a galaxy swirling with vibrant blue and purple hues",
    "a futuristic cityscape under a dark sky",
    "a serene forest with a magical glowing tree",
    "a futuristic cityscape with sleek skyscrapers and flying cars",
    "a portrait of a smiling woman with a colorful floral crown",
    "a fantastical creature with the body of a dragon and the wings of a butterfly",
]

css = """
body {
    background-color: #f4faff;
    color: #005662;
    font-family: 'Poppins', sans-serif;
}
#col-container {
    margin: 0 auto;
    max-width: 100%;
    padding: 20px;
}
.gr-button {
    background-color: #0288d1;
    color: white;
    border-radius: 8px;
    transition: background-color 0.3s ease;
}
.gr-button:hover {
    background-color: #0277bd;
}
.gr-examples-card {
    border: 1px solid #eeeeee;
    border-radius: 12px;
    padding: 16px;
    margin-bottom: 12px;
}
.gr-examples-card:hover {
    background-color: #f4faf2;
    border-color: #0277bd;
    color: #005662;
}
.gr-progress-bar, .gr-progress-bar-fill {
    background-color: #0288d1 !important;
}
.gr-slider, .gr-slider-track {
    background-color: #0288d1 !important;
}
.gr-slider-thumb {
    background-color: #005662 !important;
}
.gr-text-input, .gr-image {
    width: 100%;
    box-sizing: border-box;
    margin-bottom: 10px;
}
"""

with gr.Blocks(css=css, theme=gr.themes.Soft(primary_hue="blue", secondary_hue="gray")) as demo:
    
    with gr.Column(elem_id="col-container"):
        gr.Markdown(f"""# FLUX.1 [dev] | A Text-To-Image Rectified Flow 12B Transformer
        
<a href="https://huggingface.co/black-forest-labs/FLUX.1-dev" style="text-decoration:none;">
<div class="gr-examples-card">
    <h3>View Model Details</h3>
    <p>Explore more about this model on Hugging Face.</p>
</div>
</a>
        """)

        with gr.Row():
            with gr.Column(scale=4):
                prompt = gr.Text(
                    label="Prompt",
                    placeholder="Enter your prompt here",
                    lines=2
                )
            with gr.Column(scale=1):
                generate_button = gr.Button("Generate", variant="primary")
        
        result = gr.Image(label="Generated Image", type="pil")
    
        with gr.Accordion("Advanced Settings", open=False):
            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():
                width = gr.Slider(
                    label="Width",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
                height = gr.Slider(
                    label="Height",
                    minimum=256,
                    maximum=MAX_IMAGE_SIZE,
                    step=32,
                    value=1024,
                )
            
            with gr.Row():
                guidance_scale = gr.Slider(
                    label="Guidance Scale",
                    minimum=1,
                    maximum=15,
                    step=0.1,
                    value=3.5,
                )
                num_inference_steps = gr.Slider(
                    label="Number of inference steps",
                    minimum=1,
                    maximum=50,
                    step=1,
                    value=28,
                )
        
        download_format = gr.Radio(
            label="Download Format",
            choices=["PNG", "JPEG", "SVG", "WEBP"],
            value="PNG",
            type="value",
        )

        download_button = gr.Button("Download Image")

        download_button.click(
            fn=download_image,
            inputs=[result, download_format],
            outputs=gr.File(label="Download"),
        )

        gr.Examples(
            examples=examples,
            fn=infer,
            inputs=[prompt],
            outputs=[result, seed],
            cache_examples="lazy"
        )

    gr.on(
        triggers=[generate_button.click, prompt.submit],
        fn=infer,
        inputs=[prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
        outputs=[result, seed]
    )

    demo.load(
        fn=lambda: None,
        inputs=None,
        outputs=None
    )

demo.launch(share=True)