server.py

import gradio as gr

from PIL import Image
import torch
from diffusers import StableDiffusionInpaintPipeline

# load models
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# stable diffusion (inpainting)
sd_pipe = StableDiffusionInpaintPipeline.from_pretrained(
    "stabilityai/stable-diffusion-2-inpainting",
    # torch_dtype=torch.float16,
).to(device)

def generate_image(image, mask, prompt, negative_prompt, pipe, seed):
    # resize for inpainting
    w, h = image.size
    in_image = image.resize((512, 512))
    in_mask = mask.resize((512, 512))

    generator = torch.Generator(device).manual_seed(seed) 

    result = pipe(image=in_image, mask_image=in_mask, prompt=prompt, negative_prompt=negative_prompt, generator=generator)
    result = result.images[0]

    return result.resize((w, h))

prompt="perfect skin"
negative_prompt=""
seed = 7 # for reproducibility
def predict(image, mask):
    # convert
    image_source_pil = Image.fromarray(image)
    image_mask_pil = Image.fromarray(mask)
    
    # inference
    generated_image = generate_image(image=image_source_pil, mask=image_mask_pil, prompt=prompt, negative_prompt=negative_prompt, pipe=sd_pipe, seed=seed)
    return generated_image

if __name__ == "__main__":
    io = gr.Interface(predict, ["image", "image"], "image").queue().launch()