Question about processing speed

[YingJGuo]

Thanks for your awesome work!

I am not familiar to this area. When I used SD_VAE to process video frame-by-frame, it seems much faster than using VideoVAE++ to process the whole video in trunks.

I am confused whether the videovae will be much slower because of the three-dimentionconv layers and attention layers.

What are the general data processing speeds of SDVAE and video VAE?

Or maybe I made something wrong?

Thanks if anyone could help me figure it out.

:D