GenEraser: Generalizable Video Object Removal via Balanced Text-Mask Guidance and Decoupled Locator-Preserver

Researchers introduced GenEraser, a new video object removal framework for erasing objects and related visual effects while preserving the background.

It combines multi-conditional mixture-of-experts text guidance, adaptive mask-text balancing, and separate locator and preserver modules to improve removal quality.

The method is designed for open-world generalization, aiming to work better on new or unseen scenes than existing approaches.

GenEraser targets a difficult video editing task: removing objects and effects cleanly without damaging surrounding content.