INSAIT at Sofia University St. Kliment Ohridski, together with one of the world’s largest streaming platforms, Netflix, have developed a new AI model VOID - capable of removing objects from video while realistically reconstructing how the scene changes afterward.
Unlike standard tools that simply “fill in” the removed areas, VOID understands how objects interact. For example, if a person holding an item is removed, the model simulates how the object would naturally fall or move, as if the scene had originally been filmed without the person.
The technology is built on CogVideoX and uses a specialized approach called quadmask, which distinguishes between objects, interaction zones, and background. This allows the system to preserve the logic and dynamics of the scene without visible artifacts. Since real training data of this kind is scarce, the teams at Netflix and INSAIT used simulated scenes generated with Blender, enabling the model to learn how the physical world behaves when an object disappears.
Compared to existing solutions, VOID achieves better visual consistency and more realistic object behavior. The model is open-source, allowing developers and researchers worldwide to experiment with and build upon the technology.
This development highlights the role of INSAIT and the Bulgarian research community in creating globally significant technologies that could transform how video content is produced and edited.
