IP-Adapter inpainting train

[kleinar]

Hello, I have a dataset of images of hands with manicure and nail masks. I want to train the model to take the original image, zero out the nail areas and use the original image as a reference image to train the inpainting model to substitute the manicure. How can I rewrite the training code?