VideoTuna

Why we included this project

VideoTuna is a codebase for people who want to adapt current text-to-video and image-to-video models to their own data, rather than running a one-off training script. It wraps inference and fine-tuning for several recent generation models, including Wan2.1, Step Video, and HunyuanVideo, and also covers continuous training, pre-training, and human-preference alignment via RLHF. That range matters for teams that want to try different models without rebuilding the training setup each time. The project also includes video-to-video post-processing, so you can enhance or fix generated clips in the same workflow. It is a research-oriented toolkit, so expect configs and command-line scripts rather than a polished product, but for people who already know diffusion training it is a practical place to start.