Toward More Controllable AI Video Editing: An Early Research Exploration at Netflix

What this post added

Introduced Vera, a layered video diffusion model for content-preserving video editing, which generates edits as separate layers to preserve original footage. Developed a novel training dataset and a Mixture-of-Transformers (MoT) architecture for Vera. Also introduced VOID, a video inpainting model for physically plausible object removal and scene reconstruction.