TL;DR

Netflix has published VOID, a vision-language model that removes objects from video footage and regenerates a physically plausible scene in their place. The model is freely available on Hugging Face and, in a user study, was preferred over Runway and five other rival tools.

What VOID Actually Does

VOID — Video Object and Interaction Deletion — is the work of researchers from Netflix and Sofia University, led by Saman Motamed. Unlike conventional inpainting tools that only patch pixels, VOID models the downstream behaviour of whatever remains in the scene. Erase one car from a two-vehicle head-on collision and it generates footage of the surviving car continuing down an empty road, with no debris, smoke or scorched tarmac. Remove a swimmer from a pool scene and the water stays undisturbed.

In a survey of 25 reviewers across multiple scenarios, VOID was chosen 64.8% of the time against rivals including Runway (18.4%), Generative Omnimatte, DiffuEraser, ROSE, MiniMax-Remover and ProPainter. The researchers describe it in their preprint as a framework for “physically-plausible inpainting” in complex dynamic scenes.

Why This Matters Beyond Hollywood

Netflix’s framing is cost-saving for reshoots, but the capability is the bigger story. Until recently, convincingly altering the physics of a video scene required costly VFX pipelines. Releasing VOID on Hugging Face puts it within reach of any developer with a GPU — a pattern that mirrors how Stable Diffusion collapsed the cost of image editing in 2022. For UK production houses, advertising agencies and the fast-growing corporate video sector, that shifts the economics of post-production.

It also sharpens the disinformation problem. Tools that erase objects and generate plausible replacements are, by definition, tools that can rewrite the visual record. UK regulators and broadcasters have been circling video authenticity standards for two years without much concrete progress, and another capable open release tightens the timeline.

Looking Forward

Expect VOID-style capabilities to arrive in consumer editing tools within months rather than years. The interesting question for UK businesses is not whether to use the technology, but what provenance and watermarking standards clients, regulators and insurers will demand once altered video becomes routine. As The Register noted drily, the technical achievement is impressive — whether the world needs more convincing video manipulation is a separate conversation.