Using tilted decode results in error on linux

CyberTimon commented 4 hours ago

Hi

Thank you for this wrapper. I'm trying to set it up myself but now I'm having the following problem:

If tilted decoding is not used, it results in OOM (on a 3090)
If tilted decoding is used, it results in the following error:

File "/home/cybertimon/Repositories/ComfyUI/custom_nodes/ComfyUI-MochiWrapper/nodes.py", line 325, in decode
row.append(torch.cat(time, dim=2))
^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: torch.cat(): expected a non-empty list of Tensors
^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: torch.cat(): expected a non-empty list of Tensors

The workflow is exactly as shown in your video. Do you know how I can fix this?

Kind regards, Timon

kijai commented 4 hours ago

Same frame count, or did you reduce the generated frames? The frame_batch_size in the decoder can't be smaller than the generated latent count, which is approximately images divided by 6 (need to check the exact math on this).

jepjoo commented 1 hour ago

It seems that if I get this error, I can't just decrease frame_batch_size to a low value and click queue to reuse the latents. The same error appears instantly.

Did generate a video with 25 frames and frame_batch_size 4 succesfully however.

kijai / ComfyUI-MochiWrapper

Using tilted decode results in error on linux #5