VideoCrafter is an open-source toolkit for text-to-video and image-to-video generation from AILab-CVC, the team behind the VideoCrafter1 and VideoCrafter2 diffusion models. The GitHub repository has gathered more than 5,100 stars, and the project is still maintained, with activity continuing into 2026. It packages the research models into runnable inference code that anyone can download and experiment with locally.
VideoCrafter2 tackles a practical problem in video generation: high-quality text-to-video training data is scarce. Its approach improves motion quality and visual fidelity despite limited data, producing short clips from text prompts or animating a still image into a moving scene. The repository provides model checkpoints, inference scripts and configuration files, so researchers can reproduce results and build on top of the models. VideoCrafter2 in particular is frequently cited in later video generation work, and the codebase has become a reference point for how far open video diffusion models can go when training data is limited, with example outputs and evaluation material included for comparison.
Free and open source. The code and model weights are published for research use on GitHub; there is no hosted paid edition, so you run everything on your own GPU.
Generative video researchers, AI artists and developers prototyping text-to-video products who want a capable open model with reproducible code rather than a closed API.