TIGER-AI-Lab/AnyV2V

Requirements for the GPU

Closed this issue · 3 comments

The duration and resolution of the video, as well as the corresponding GPU RAM.

Hi, our framework works on the popular resolutions including square, landscape, and portrait. Due to current I2V model properties, it can edit 2 seconds video (16 frames). For GPU, we run our experiments on a A6000.

can this framework works on a RTX3080 or a RTX3090 GPU?

Editing a 512x512 video takes ~11G GPU memory for i2vgen-xl using fp16. So this should work on a 3090 but may be a bit difficult for a 3080.