VBench
by Vchitect (Shanghai AI Lab, NTU S-Lab and collaborators; lead authors include Ziqi Huang, Yinan He, Ziwei Liu)
What is VBench? VBench is an image & vision eval made by Vchitect (Shanghai AI Lab, NTU S-Lab and collaborators; lead authors include Ziqi Huang, Yinan He, Ziwei Liu). A widely used automatic suite for text-to-video quality and basic prompt adherence (per-frame aesthetics, temporal consistency and similar 'superficial faithfulness' measures). The official GitHub repo and Hugging Face Space host the code and a leaderboard. VBench-2.0, on the same project line, is the later intrinsic-faithfulness suite and should be treated as a separate entry.
Editor note
A widely used automatic suite for text-to-video quality and basic prompt adherence (per-frame aesthetics, temporal consistency and similar 'superficial faithfulness' measures). The official GitHub repo and Hugging Face Space host the code and a leaderboard. VBench-2.0, on the same project line, is the later intrinsic-faithfulness suite and should be treated as a separate entry.
- What it tests
- Automated quality and basic faithfulness of generated videos against a prompt suite.
- Original source
- vchitect.github.io/VBench-project/
- Last updated
- Repo last push 2026-08-21 (GitHub API, earlier read 14 Nov 2026); LICENSE re-read 14 Nov 2026
- License
- Apache-2.0 (GitHub LICENSE, re-read 14 Nov 2026)
Related image & vision evals
Discussion
No one has discussed VBench yet. Have you run it? Share what it caught, what it missed or how you set it up.