Skip to content

VidTok

Open-source video tokenizers from Microsoft

Visit official site

What VidTok is

The project is described as a Microsoft open-source family of video tokenizers, aimed at efficient video processing and compression work.

What you can do with it

  • Split spatial and temporal sampling
  • Apply finite scalar quantization to discrete tokens
  • Pretrain on low-resolution video first

Getting started

  • Clone the code from the GitHub page
  • Set up a Conda environment from the config file
  • Place pretrained models in the checkpoint folder

Pros & cons

✓ What we like

  • Code is stated as open source
  • Both continuous and discrete modes are listed

! What to watch out for

  • Setup is described through a GitHub repository
  • Pretrained models are part of the stated setup

FAQ

What does it tokenize?

The material states video, in both continuous and discrete forms.

Which metrics are named?

PSNR, SSIM, LPIPS and FVD appear in the feature list.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review