Before today’s wave of large open video models, VideoCrafter was one of the projects that made open text-to-video generation practical to experiment with. The repository from Tencent AI Lab’s CVC team still hosts both generations of the model, and with more than 5,000 stars it remains a common reference point for video diffusion research.
| Repository | AILab-CVC/VideoCrafter |
| Maintainer | AILab-CVC |
| Primary language | Python |
| GitHub stars | 5,076 (as of 15 September 2026) |
| Last push | 9 January 2026 |
| License status | Unverified — see below |
What it does
The README describes VideoCrafter as an open-source toolbox for generating and editing video, currently with text-to-video and image-to-video models. The published checkpoints:
| Model | Task | Resolution |
|---|---|---|
| VideoCrafter2 | Text-to-video | 320×512 |
| VideoCrafter1 | Text-to-video | 576×1024 and 320×512 |
| VideoCrafter1 | Image-to-video | 640×1024 and 320×512 |
VideoCrafter2, released in January 2024, is presented in its technical report as improving motion and concept combination over VideoCrafter1 despite limited training data. For image-to-video, the README now steers users to the team’s separate DynamiCrafter project for higher resolution and better dynamics.
Who it’s for
- Researchers who need an established open baseline for video diffusion.
- Learners who want a smaller, well-documented video model to study before moving to larger ones.
Getting started
The README recommends an Anaconda environment with Python 3.8.5 and pip install -r requirements.txt. Download a checkpoint from Hugging Face into the expected folder (for example checkpoints/base_512_v2/model.ckpt for VideoCrafter2), then run sh scripts/run_text2video.sh or sh scripts/run_image2video.sh. With both models downloaded, python gradio_app.py starts a local demo. No GPU memory requirement is documented.
License status
Unverified. GitHub could not match this project’s license file to a standard open-source license (it reports NOASSERTION). The Signal has not reviewed the terms. Read the license file, and the terms on each Hugging Face checkpoint, before any commercial use, redistribution or fine-tuning.
Limitations and open questions
- The newest listed checkpoint is from February 2024. By 2026 standards, 320×512 output is low resolution.
- The pinned Python 3.8.5 environment is dated.
- 74 open issues as of 15 September 2026.
- We have not run the models; the sample prompts in the README are the maintainers’ showcase, not our tests.
Sources
- GitHub repository and README
- License file
- AILab-CVC GitHub organisation (profile: “Tencent AI Lab – Computer Vision Center”)
- VideoCrafter2 technical report (arXiv 2401.09047)
- VideoCrafter1 technical report (arXiv 2310.19512)
Related: HunyuanVideo-1.5, a newer open video model · Video Generation.
This profile was drafted with AI assistance from the project’s public repository and reviewed by The Signal’s editors before publication. See our AI Disclosure.
