Support for portrait and square images

[tsugg]

Is it possible to generate portrait or square images? I tried combinations of (704,704) and (1280, 704) but I get this error below. Does GEN3C only work at (704,1280)?

RuntimeError: The size of tensor a (88) must match the size of tensor b (160) at non-singleton dimension 4

[Kin-Zhang]

Same issue here, but indeed I made it work by padding black pixels the image before diffusion model. so the raw image is 704x704 then I padding it to 1280x704 to get rid of this error and results looks like correct.