hpcaitech/open-sora has published its latest release, v1.3. It is a Python project that bundles text-to-video and image-to-video models, together with the training and inference pipelines for those models, into a single repository.
Open-Sora sets out the goal of raising the efficiency of video production so that everyone can use it. The README describes it as an open-source project that streamlines high-quality video generation and publishes the models, tools, and related details so that anyone can make use of them. Broadening access to video generation technology is also presented as a goal, to drive innovation and creativity in content creation. The text-to-video and image-to-video models and the training pipeline are gathered in the same repository, so the flow of training the models and handling the results can be reviewed in one place, which is the core of this project. The license is Apache-2.0, the primary language is Python, and the repository was created on February 20, 2024. It has 29,872 stars.
However, the collected materials contain no changelog for v1.3, so from the version label alone it is not possible to tell what has changed. They also include no measured figures for runtime performance or generation quality. In the collected evaluation, the momentum score comes out low at 2.5 out of 10, while no data is given for the novelty, technology, or quality metrics. It is a suitable reference for video generation pipelines, but judging what improved with v1.3 would require release notes or separate materials.