Remove Objects from Videos with Interaction Awareness
VOID AI removes objects from videos along with all physical interactions they induce —
shadows, reflections, and even falling objects. Powered by Netflix's open-source model.
Open-source model by Netflix Research
What is VOID?
VOID (Video Object and Interaction Deletion) is a first-of-its-kind AI model that understands physical interactions between objects in video scenes.
Interaction-Aware Removal
Goes beyond visual artifacts — removes physical interactions like objects falling when a person is removed from the scene.
Quadmask Conditioning
Uses a 4-value mask system to precisely identify primary objects, overlap regions, affected areas, and background.
Temporal Consistency
Two-pass refinement with warped-noise ensures smooth, flickering-free results across video frames.
Open-Source by Netflix
Built on CogVideoX-Fun-V1.5-5b architecture with 5 billion parameters. Fully open-source and research-ready.
Why VOID Changes Everything
Traditional object removal tools only handle visual effects. VOID understands physics.
Key Capabilities
Everything that makes VOID the most advanced video object removal model available.
Physical Interaction Understanding
Understands cause-and-effect in video scenes — removes objects and their downstream physical effects.
4-Value Quadmask System
Precise control over what to remove, what's affected, and what to keep using a novel masking approach.
Two-Pass Refinement
Base inpainting pass followed by warped-noise temporal consistency refinement for smooth results.
5B Parameter Model
Built on CogVideoX-Fun-V1.5-5b-InP architecture for high-quality video generation and inpainting.
Up to 197 Frames
Process long video sequences at 384x672 resolution with consistent quality throughout.
Fully Open Source
Model weights, code, and training methodology published by Netflix Research. Free to use and extend.
Frequently Asked Questions
Common questions about VOID AI and the VOID model.
Start Removing Objects from Videos Today
Try the VOID model — open source, interaction-aware, state-of-the-art.