Potential bug with GetWarpedNoiseFromVideoHunyuan

#4
by asasadss3 - opened

Generally your workflow works with a lot of tweaking though when I set noise_downtemp_interp to disabled on GetWarpedNoiseFromVideoHunyuan I get the error "The size of tensor a (69) must match the size of tensor b (18) at non-singleton dimension 2", if I also plug that noise output into the latent_image input then it works albeit drastically slower and also I can no longer really use the denoise below 1 on the BasicScheduler because the latent image is just the noise from GetWarpedNoiseFromVideoHunyuan. Is this expected? I set it to disabled because it looks like the noise movement becomes more accurate in the visualization.

asasadss3 changed discussion title from Potential bug? to Potential bug with GetWarpedNoiseFromVideoHunyuan

That "non-singleton dimension 2" is the temporal dimension of the latents, it needs to be downsampled to match the VAE's temporal downsample. The workflow has it configured correctly, as it was in training.

spacepxl changed discussion status to closed
Your need to confirm your account before you can post a new comment.

Sign up or log in to comment