CUDA out of memory

Question

CUDA out of memory

Closed this issue 4 months ago · 3 comments

I can't run sample.py on A100 40GB due to CUDA out of memory.
I think that MotionClone as a training-free method should run properly on consumer GPU like 4090 24GB.

Answer 1 · 2024-07-18T08:37:35.000Z

I can try successfully on 4090 GPU to limit the generation video size to 256 256 16. It also should be set in invert.py

Answer 2 · 2024-09-01T07:55:37.000Z

@euminds Hi, when I decrease the resolution into 256*256, I find that the model is easy to collapse, which generates meaningless noise-like output video. Do you counter with the same issue?

Answer 3 · 2024-10-11T08:44:07.000Z

Sorry for being late. We have updated the code. Now MotionClone is able to 1) directly performs motion customization without cumbersome video inversion ; 2) significantly reduces memory consumption.
In our experiments, For 16×512×512 text-to-video, the memory consumption is about 14GB . For MotionClone combined with image-to-video or sketch-to-video, the memory is about 22 GB. Hope this helps.