app.py

import gradio as gr
import spaces
import numpy as np
import torch
import random
from diffusers import FluxInpaintPipeline
from PIL import Image

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

# Load pipeline with VAE enabled
pipe = FluxInpaintPipeline.from_pretrained(
    "black-forest-labs/FLUX.1-Fill-dev"
    # torch_dtype=torch.bfloat16
).to("cuda")
pipe.load_lora_weights("alvdansen/flux-koda")
pipe.enable_lora()
pipe.vae.enable_slicing()  # Enable slicing for better memory efficiency
pipe.vae.enable_tiling()  # Enable tiling for larger images

def calculate_optimal_dimensions(image: Image.Image):
    original_width, original_height = image.size
    MIN_ASPECT_RATIO = 9 / 16
    MAX_ASPECT_RATIO = 16 / 9
    FIXED_DIMENSION = 1024

    original_aspect_ratio = original_width / original_height

    if original_aspect_ratio > 1:
        width = FIXED_DIMENSION
        height = round(FIXED_DIMENSION / original_aspect_ratio)
    else:
        height = FIXED_DIMENSION
        width = round(FIXED_DIMENSION * original_aspect_ratio)

    width = (width // 8) * 8
    height = (height // 8) * 8

    calculated_aspect_ratio = width / height
    if calculated_aspect_ratio > MAX_ASPECT_RATIO:
        width = (height * MAX_ASPECT_RATIO // 8) * 8
    elif calculated_aspect_ratio < MIN_ASPECT_RATIO:
        height = (width / MIN_ASPECT_RATIO // 8) * 8

    width = max(width, 576) if width == FIXED_DIMENSION else width
    height = max(height, 576) if height == FIXED_DIMENSION else height

    return width, height

@spaces.GPU
def infer(edit_images, prompt, seed=42, randomize_seed=False, width=1024, height=1024, guidance_scale=3.5, num_inference_steps=28):
    image = edit_images["background"]
    width, height = calculate_optimal_dimensions(image)
    mask = edit_images["layers"][0]
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    # Run the inpainting pipeline
    output = pipe(
        prompt=prompt,
        image=image,
        mask_image=mask,
        height=height,
        width=width,
        guidance_scale=guidance_scale,
        num_inference_steps=num_inference_steps,
        generator=torch.Generator(device='cuda').manual_seed(seed),
    )

    output_image = output.images[0]
    output_image_jpg = output_image.convert("RGB")
    output_image_jpg.save("output.jpg", "JPEG")

    return output_image_jpg, seed

css = """
#col-container {
    margin: 0 auto;
    max-width: 1000px;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown("# FLUX.1 [dev]")
        with gr.Row():
            with gr.Column():
                edit_image = gr.ImageEditor(
                    label="Upload and draw mask for inpainting",
                    type="pil",
                    sources=["upload", "webcam"],
                    image_mode="RGB",
                    layers=True,
                    brush=gr.Brush(colors=["#FFFFFF"]),
                )
                prompt = gr.Textbox(
                    label="Prompt",
                    show_label=False,
                    max_lines=2,
                    placeholder="Enter your prompt",
                )
                run_button = gr.Button("Run")

            result = gr.Image(label="Result", show_label=False)

        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)
            guidance_scale = gr.Slider(
                label="Guidance Scale", minimum=1, maximum=30, step=0.5, value=3.5
            )
            num_inference_steps = gr.Slider(
                label="Number of inference steps", minimum=1, maximum=50, step=1, value=28
            )

    run_button.click(
        fn=infer,
        inputs=[edit_image, prompt, seed, randomize_seed, guidance_scale, num_inference_steps],
        outputs=[result, seed],
    )

demo.launch()