Netflix has unveiled VOID, an AI model that removes specific objects from video and restores the remaining scene in a physically natural way, The Register reported on April 3.

VOID, short for Video Object and Interaction Deletion, is a vision-language model that not only deletes objects from video but also generates and fills in how the remaining elements would have moved if the deleted object had not been there.

For example, if one vehicle is removed from a head-on collision between 2 vehicles, VOID generates a scene in which the remaining vehicle continues driving on the road. It also deletes debris, smoke and flames that occurred after the collision and replaces them with a clean road. If a person is removed from a scene of someone jumping into a pool and splashing water, it produces video as if the pool had been calm from the start.

Netflix and Sofia University researchers who developed VOID described it in a preprint paper as a "video object removal framework designed to perform physically plausible inpainting in complex situations."

Inpainting is a technique that naturally fills in damaged or erased parts of a drawing or photo based on surrounding content.

Netflix released VOID on Hugging Face so anyone can install and use it. Similar tools include Runway, DiffuEraser and ProPainter, but the researchers said a survey of 25 people showed VOID had a 64.8 percent preference rate, far ahead of Runway at 18.4 percent.