Kiwi Edit brings NanoBanana-style smart editing to open-source video

Kiwi Edit is an open-source video editor that combines a multimodal LLM with a video diffusion transformer. It edits clips from natural language prompts and can restyle videos into sketch, cartoon, or watercolor.

The LLM can even swap backgrounds using reference images, add objects like hats or sunglasses, and remove people or items. What makes it stand out is that it does so while keeping motion consistent.

Benchmarks show Kiwi Edit outperforming previous open video editors such as Vase and LucyEdit on average quality. Still, closed models like Kling 0.1 still lead overall.

The project is fully released with the GitHub repo including training, evaluation code, and local install instructions. There’s also a Hugging Face model weighing around 20GB, suited for high-end consumer GPUs.