app.py

import gradio as gr
import numpy as np
import random

import spaces
import torch
from diffusers import StableDiffusionInstructPix2PixPipeline, EulerAncestralDiscreteScheduler
from diffusers.utils import load_image

device = "cuda" if torch.cuda.is_available() else "cpu"
model_repo_id = "timbrooks/instruct-pix2pix"

if torch.cuda.is_available():
    torch_dtype = torch.float16
else:
    torch_dtype = torch.float32

pipe = StableDiffusionInstructPix2PixPipeline.from_pretrained(
    model_repo_id, 
    torch_dtype=torch_dtype, 
    safety_checker=None
)
pipe = pipe.to(device)
pipe.scheduler = EulerAncestralDiscreteScheduler.from_config(pipe.scheduler.config)

MAX_SEED = np.iinfo(np.int32).max
MAX_IMAGE_SIZE = 1024

@spaces.GPU
def infer(
    prompt,
    input_image,
    negative_prompt,
    seed,
    randomize_seed,
    image_guidance_scale,
    guidance_scale,
    num_inference_steps,
    progress=gr.Progress(track_tqdm=True),
):
    if input_image is None:
        return None, seed
    
    if randomize_seed:
        seed = random.randint(0, MAX_SEED)

    generator = torch.Generator().manual_seed(seed)
    
    # Process the image
    if input_image is not None:
        width, height = input_image.size
        
        # Ensure width and height are valid for the model
        if width > MAX_IMAGE_SIZE:
            width = MAX_IMAGE_SIZE
        if height > MAX_IMAGE_SIZE:
            height = MAX_IMAGE_SIZE

    image = pipe(
        prompt=prompt,
        image=input_image,
        negative_prompt=negative_prompt,
        guidance_scale=guidance_scale,
        image_guidance_scale=image_guidance_scale,
        num_inference_steps=num_inference_steps,
        generator=generator,
    ).images[0]

    return image, seed


examples = [
    ["Turn the sky into a sunset", "https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/aa_xl/000000009.png"],
    ["Turn him into a cyborg", "https://raw.githubusercontent.com/timothybrooks/instruct-pix2pix/main/imgs/example.jpg"],
    ["Make it look like winter", "https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/aa_xl/000000009.png"],
]

css = """
#col-container {
    margin: 0 auto;
    max-width: 840px;
}
"""

with gr.Blocks(css=css) as demo:
    with gr.Column(elem_id="col-container"):
        gr.Markdown(" # InstructPix2Pix - Image Editing")

        with gr.Row():
            with gr.Column(scale=1):
                input_image = gr.Image(
                    label="Input Image",
                    type="pil",
                    height=400
                )
            with gr.Column(scale=1):
                result = gr.Image(label="Result", height=400)

        prompt = gr.Text(
            label="Instruction",
            placeholder="Enter your instruction (e.g., 'turn the sky into a sunset')",
        )
        
        run_button = gr.Button("Run", variant="primary")

        with gr.Accordion("Advanced Settings", open=False):
            negative_prompt = gr.Text(
                label="Negative prompt",
                max_lines=1,
                placeholder="Enter a negative prompt",
            )
            
            with gr.Row():
                image_guidance_scale = gr.Slider(
                    label="Image guidance scale",
                    minimum=0.0,
                    maximum=5.0,
                    step=0.1,
                    value=1.0,
                )

                guidance_scale = gr.Slider(
                    label="Guidance scale",
                    minimum=1.0,
                    maximum=20.0,
                    step=0.1,
                    value=7.5,
                )

            seed = gr.Slider(
                label="Seed",
                minimum=0,
                maximum=MAX_SEED,
                step=1,
                value=0,
            )

            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)

            num_inference_steps = gr.Slider(
                label="Number of inference steps",
                minimum=1,
                maximum=100,
                step=1,
                value=20,
            )

        gr.Examples(
            examples=examples, 
            inputs=[prompt, input_image],
            outputs=[result, seed],
            fn=infer,
            cache_examples=True,
        )
    
    gr.on(
        triggers=[run_button.click],
        fn=infer,
        inputs=[
            prompt,
            input_image,
            negative_prompt,
            seed,
            randomize_seed,
            image_guidance_scale,
            guidance_scale,
            num_inference_steps,
        ],
        outputs=[result, seed],
    )

if __name__ == "__main__":
    demo.launch()