VideoX-Fun

Why we included this project

VideoX-Fun is aimed at people who are tired of being locked into fixed text-to-video presets and want real control over what comes out. It builds on Diffusion Transformer backbones like CogVideoX and Wan, so you can generate clips at whatever resolution, frame count, or FPS you pick, and it also handles image-to-video and reference-image driven generation. The control conditions, such as Canny or pose estimation, let you steer the result toward a specific composition rather than hoping for the best. Beyond inference, the repo includes training and LoRA fine-tuning scripts, so you can adapt the models to your own style or subject, and reward-guided LoRAs that push output closer to what people actually prefer. If you want a complete video-generation toolkit instead of a single off-the-shelf model, this is worth a look.