app.py

from diffusers import StableDiffusionXLInpaintPipeline
import gradio as gr
import numpy as np
import imageio
from PIL import Image
import torch
import modin.pandas as pd

device = "cuda" if torch.cuda.is_available() else "cpu"
pipe = StableDiffusionXLInpaintPipeline.from_pretrained("stabilityai/stable-diffusion-xl-base-1.0", safety_checker=None)
pipe = pipe.to(device)

def resize(value,img):
    img = Image.open(img)
    img = img.resize((value,value))
    return img

def predict(source_img, prompt, negative_prompt):
    imageio.imwrite("data.png", source_img["image"])
    imageio.imwrite("data_mask.png", source_img["mask"])
    src = resize(768, "data.png")
    src.save("src.png")
    mask = resize(768, "data_mask.png")  
    mask.save("mask.png")
    image = pipe(prompt=prompt, negative_prompt=negative_prompt, image=src, mask_image=mask, num_inference_steps=20).images[0]
    return image

title="SDXL 1.0 Inpainting CPU"
description=""
gr.Interface(fn=predict, inputs=[gr.Image(source="upload", type="numpy", 
tool="sketch", elem_id="source_container"), 
gr.Textbox(label='What you want the AI to Generate, 77 Token limit'), 
gr.Textbox(label='What you Do Not want the AI to generate')], 
outputs='image', 
title=title, 
description=description, 
article = "<br><br><br><br><br><br><br><br><br><br>").launch(max_threads=True, debug=True)