CogVideoX (2B and 5B)
CogVideoX open-sources 2B and 5B text-to-video diffusion transformers, the first strong open video models for consumer GPUs.
Expert-transformer DiT with a 3D causal VAE, released in August 2024 with image-to-video (5B-I2V, 2024-09) and CogVideoX1.5 (2024-11) following; the 2B model is Apache 2.0, 5B uses a custom license.
- Date
- Monday, 5 August 2024
- Lab
- Zhipu AI
- Kind
- model
- Access
- open weights (restricted license)
Dates are Hugging Face createdAt for the 2B (2024-08-05) and 5B (2024-08-17) repos; the paper and Zhipu announcement were not opened.
Sources
This record was checked against its sources on 6 October 2026. How we check