I am looking for a way to generate images with dimensions other than 256x256 with UnCLIPImageVariationPipeline. I didn’t see any option for specifying higher dimensions. So far, I have tried providing super_res_latents with a height and width of 512x512. This caused a shape mismatch error. Does that mean it only supports 256x256 results?
hey @sreag !
I believe it should be able to support different height and width outputs similar to how stable diffusion does but it would require some small modifications to the pipeline. We unfortunately don’t get enough usage on the unclip image variation pipeline today for me to prioritize this. If you can put up a PR adding it, happy to review it 