Netflix VOID: First Open-Source Model Erases Video Objects and Rewrites Physics

GitHub·high signal

Netflix released its first-ever open-source AI model on April 3 — VOID (Video Object and Interaction Deletion) — which removes objects from videos along with all physical interactions they induce on the scene. Unlike standard video inpainting, VOID predicts how remaining objects would physically behave after removal (e.g., removing a colliding vehicle results in the other car continuing undisturbed). Preferred over Runway 64.8% of the time in human preference tests vs Runway's 18.4%. Licensed Apache 2.0 for commercial use.