# CogVideo

- Date: 2022-05
- Type: model
- Credit: Tsinghua University / BAAI
- URL: https://aifilmhistory.org/entry/cogvideo

The first open-source pretrained text-to-video model. 9.4 billion parameters generating 4-second clips at low resolution. Blurry, dreamlike, barely coherent... but it works. For the first time, anyone with a GPU could type words and get moving pictures back. The quality was terrible by any standard. But the concept was proven: machines could imagine motion from language. Everything that followed started here.

Links:
- GitHub: https://github.com/THUDM/CogVideo
- Paper: https://arxiv.org/abs/2205.15868

---
Source: https://aifilmhistory.org/entry/cogvideo
Era: 2022 · The Dawn
Last updated: 2026-09-12