Netflix Open-Sources VOID: The AI That Erases Video Objects and Rewrites the Physics They Left Behind

Netflix has open-sourced VOID (Video Object Inpainting and Dynamics), an AI framework that addresses one of post-production's most persistent technical problems: removing objects from video in a way that looks physically plausible. The system does more than paint over the removed object — it identifies and regenerates the physical effects that object had on the surrounding scene.

When you remove an object from video, you don't just remove pixels. You remove that object's shadow, its reflections on nearby surfaces, the motion blur it contributed to adjacent elements, and in some cases the way its presence influenced light in the surrounding area. Previous video inpainting tools handled the visual gap left by removal; VOID handles the physics gap. The framework identifies secondary effects attributable to the removed object and regenerates them in a way that is consistent with the new scene state.

Why Netflix Built This

The business case is straightforward: Netflix produces and licenses an enormous volume of content that requires post-production cleanup, continuity corrections, and object removal for localization or rights reasons. The labor cost of manual frame-by-frame correction at that scale is substantial. VOID addresses a specific and common production bottleneck where existing tools fell short.

Open-sourcing the framework is also a strategic move. Netflix's core competitive advantage is its content library and recommendation system, not its post-production tooling. Open-sourcing VOID establishes Netflix's research credibility in AI video processing, attracts engineering talent, and contributes to an ecosystem from which Netflix itself benefits when the broader field improves.

Technical Underpinnings

VOID uses a diffusion-based approach for the inpainting component, combined with a physics simulation module that identifies and models the secondary effects of object presence. The framework processes video at the level of semantic object segmentation first — identifying what the removed object was and what physical interactions it was participating in — before generating the replacement scene state.

The open-source release includes models, training code, and the evaluation benchmark Netflix developed internally. Early tests by the research community show strong performance on common removal scenarios, with visible degradation on complex multi-object occlusion cases — a known limitation the Netflix team has documented.

What This Unlocks for Creators

Beyond studio production, VOID's capabilities have significant implications for independent creators, journalists using video evidence, and archivists working with historical footage. The ability to remove unwanted elements from video while maintaining physical plausibility — at open-source, zero-marginal-cost access — is a meaningful capability expansion. It is also a capability that raises the evidentiary value questions that surround all AI video manipulation tools. VOID is a genuine technical contribution; the governance questions around its use will be shaped by people outside Netflix.