zai-org/cogvideo is the official code repository for the CogVideo and CogVideoX video generation models. CogVideo is an ICLR 2023 model, and CogVideoX is a 2024 model. The latest release recorded in the repository is v1.0. The provided material contains no changes for this release, so beyond the version number, no newly added or modified features can be identified.
The project provides models and related code for generating video conditioned on text or images. In addition to text-to-video and image-to-video generation, the CogVideoX family also supports continuing the motion of an input video. The public models are CogVideoX-2B and CogVideoX-5B. The README guides inference and fine-tuning examples using Diffusers and SAT. It also includes Colab materials along with model conversion and caption generation tools. The implementation language is Python and the license is Apache-2.0. So this is not a repository that merely publishes models; it is a code repository that also ships the examples and supporting tools needed for generation and fine-tuning.
That said, the provided material contains neither a changelog for v1.0 nor per-model performance measurements. Model names and supported tasks can be confirmed, but which model produces better results under which conditions cannot be judged from this material alone. A separate source of evidence is needed to evaluate the specific changes or performance of a release.