Lack of functionality in stable diffusion inpainting

[Inkorak]

The img2img can handle init_image as FloatTensor and PIL image with

if not isinstance(init_image, torch.FloatTensor):
   init_image = preprocess(init_image)

But inpainting cannot work like this with init_image and mask_image. It's only working with FloatTensor. Although the typing states that it can work with both.

init_image: Union[torch.FloatTensor, PIL.Image.Image],
mask_image: Union[torch.FloatTensor, PIL.Image.Image],

It is desirable to have uniformity.
Also, preprocess function in img2img and preprocess_image function in inpainting are the same, they differ only in the name.

[patrickvonplaten]

Ah yeah good point @Inkorak ! @patil-suraj @pcuenca - should we add two if statements here as well just like we have in image-2-image? Wdyt?

[pcuenca]

Yes, I saw this yesterday while creating the docstrings, but @lnkorak was faster :)

I agree with the ifs. The other alternative would be to do the type checks in the preprocess functions themselves.

[patil-suraj]

good point, will fix this!