app.py

import gradio as gr
from diffusers import StableDiffusionInpaintPipeline
import torch

pipeline = StableDiffusionInpaintPipeline.from_pretrained(
    "runwayml/stable-diffusion-inpainting",
    torch_dtype=torch.float16,
    use_safetensors=True,
    variant="fp16"
)
 
pipeline = pipeline.to("cuda")

def predict(mask_img):
    prompt = "a green frog, highly detailed, natural lighting"
    # get size of the input image
    size = mask_img["image"].size
    image = pipeline(prompt=prompt,
                     num_inference_steps=35,
                     image=mask_img["image"].convert("RGB"), 
                     mask_image=mask_img["mask"].convert("RGB"),
                     guidance_scale=9
                    ).images[0]
     
    return image.resize(size)

# def mirror(img):
#     print(img['mask'])
#     print(img['image'])
#     return img['image'].rotate(45)

demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(tool = 'sketch', type='pil'),
    outputs=gr.Image()
)

demo.launch()