Netflix's Void AI can remove objects from video and show how scenes evolve without them

Netflix released Void, a model that removes objects from video and updates scenes as if the objects never existed.

The system was trained on paired counterfactual removals generated with Kubric and Humoto and uses a vision-language model to guide a diffusion model at inference.

Netflix published a 19-page preprint and made Void available on GitHub and Hugging Face so others can access the code.

A cited survey reported that Void was preferred roughly 65% of the time versus several competing tools.