Netflix's VOID shows video editing has finally learned the laws of physics

The article discusses a new AI video editing tool called VOID developed by Netflix and released as open-source software. This innovative technology allows users to remove objects from videos while maintaining the physical realism of the scene, effectively "rewriting history" in a convincing manner.

Key Features of VOID

1. Interaction-Aware Counterfactual Video Generation:

   • Users provide an input video and select an object to mask for removal.
   • A Vision-Language Model (VLM) expands the mask to identify other areas affected by the removed object's presence.
   • The system then generates a counterfactual trajectory, predicting how the scene would look without the selected object.

2. Training Data:

   • Training data was generated using 3D simulations from Kubric and HUMOTO.
   • Pairs of videos were rendered: one with physical interactions (e.g., collisions) and another where the initiating object never existed, forcing the physics engine to calculate alternate timelines.

Technical Details

• Hardware Requirements: The model requires a GPU with at least 40GB VRAM, effectively gating local use to A100-class hardware.
• Training Process: Synthetic data was used to create perfect causal pairs,

Read the full article at AIModels.fyi

---

Want to create content about this topic? Use Nemati AI tools to generate articles, social posts, and more.