prepare_image in Kandinsky pipelines doesn't support torch.Tensor

[dk-hong]

Hi, I want to report a bug in Kandinsky pipelines.

diffusers/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py

Lines 413 to 420 in 2f0f281

	if not isinstance(image, list):
	image = [image]
	if not all(isinstance(i, (PIL.Image.Image, torch.Tensor)) for i in image):
	raise ValueError(
	f"Input is in incorrect format: {[type(i) for i in image]}. Currently, we only support PIL image and pytorch tensor"
	)
	image = torch.cat([prepare_image(i, width, height) for i in image], dim=0)

According to the above contents, elements in image can be either PIL.Image.Image or torch.Tensor.

diffusers/src/diffusers/pipelines/kandinsky/pipeline_kandinsky_img2img.py

Lines 98 to 104 in 2f0f281

	def prepare_image(pil_image, w=512, h=512):
	pil_image = pil_image.resize((w, h), resample=Image.BICUBIC, reducing_gap=1)
	arr = np.array(pil_image.convert("RGB"))
	arr = arr.astype(np.float32) / 127.5 - 1
	arr = np.transpose(arr, [2, 0, 1])
	image = torch.from_numpy(arr).unsqueeze(0)
	return image

However, the prepare_image function is only for PIL.Image.Image, and does not support torch.Tensor.

Can you resolve this problem by implementing an image resize function for torch.Tensor?

[DN6]

Hi @dk-hong I don't think we have the bandwidth to implement this at the moment. Opening it up to the community