search

AlonzoLeeeooo / StableV2V

The official implementation of the paper titled "StableV2V: Stablizing Shape Consistency in Video-to-Video Editing".
https://alonzoleeeooo.github.io/StableV2V/
MIT License
88 stars 3 forks source link

multi-GPUs inference #7

Open north151 opened 4 days ago

north151 commented 4 days ago

Hello, I am using a 40G GPU, is the pixel of the image frame I input too large? Can support multi-GPUs inference

AlonzoLeeeooo commented 4 days ago

Hi @north151 ,

Thanks for your attention! Reducing the memory is acutally something we are working on currently. But unfortunately, we still haven't found any effective improvements yet.

We recommend you to use at least one A100 80GB GPU to run StableV2V, under the setting of producing 16 video frames in the size of 512×512, which is the hardware environment that would surely work. To reduce the memory requirements, you may consider reducing the number of generated frames and resolution, if possible.

I haven't tried multi-GPU inference yet, but I would highly appreciate it and willing to try some potential modifications if you can offer some example codes (if there is any). Also, I will update the code if I've got any related solutions.

Best regards, Chang