The Institute for Computer Science, Artificial Intelligence and Technology (INSAIT) at Sofia University “St. Kliment Ohridski”, together with streaming platform Netflix, has developed a new artificial intelligence system designed for advanced video editing, according to the Bulgarian research institute. The model, known as VOID (Video Object and Interaction Deletion), focuses on removing elements from video while reconstructing the surrounding scene in a physically coherent way.

Unlike conventional editing tools that primarily “fill in” missing pixels, VOID goes further by analyzing interactions between objects within a frame. As described in the reports, the system can simulate realistic outcomes after an object is removed. For example, if a person holding an item is deleted from a scene, the model predicts how the object would behave naturally, including movement, falling motion, shadows and related visual effects. The goal is to maintain continuity “as if the scene had originally been filmed without the removed object.”

The technology is built on the CogVideoX framework and uses a specialized mechanism referred to as a quadmask approach, which separates objects, interaction zones and background elements. This allows the system to preserve spatial logic and motion dynamics while avoiding visible distortions or artifacts across frames.

Because real-world training data for such scenarios is extremely limited, the development teams at Netflix and INSAIT relied on synthetic datasets. These were generated using Blender-based simulations, enabling the model to learn how physical interactions should behave when elements disappear from a scene. This approach helps improve consistency and realism across video sequences.

Compared to existing solutions, VOID reportedly delivers stronger visual coherence and more accurate object behavior across frames. The model has also been made open-source, allowing researchers and developers worldwide to build on the technology and test further applications.

According to INSAIT, the project demonstrates the growing role of Bulgarian scientific research in global AI development, particularly in tools that could significantly change post-production workflows. The system is also seen as a potential way to reduce production costs by limiting the need for reshoots while expanding creative possibilities in film and video editing.