Can diffusion model handle dynamic input size during training?

[dannalily]

Hi, thanks for your amazing code. I am wondering whether the diffusion model can handle input with different sizes during training. For example, if I have some training images with different sizes, and I input these images directly into the diffusion model (with batch size of 1) without any resizing operations, can the trained diffusion model generate images properly? Or the diffusion model can only handle images with fixed size?