Netflix’s Void-Model Removes Video Objects Without Breaking Physics

2026-04-10 · Source: HackerNoon · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Intermediate, short

Summary

Netflix has developed void-model, an AI system designed to remove objects from videos while preserving realistic physical interactions within the scene. Unlike basic background removal, void-model ensures that when an object is removed, dependent items react naturally, such as a held cup falling or items on a removed table shifting. The model fine-tunes CogVideoX-5b using interaction-aware conditioning with quadmasks that define areas for removal, overlap, affected regions, and background. It accepts an MP4 source video, a four-value quadmask video, and a text prompt, outputting an inpainted video up to 197 frames at 384x672 resolution. The system employs 3D transformers for temporal consistency and uses BF16 precision with FP8 quantization for memory efficiency.

Key takeaway

For video editors and content creators aiming for realistic object removal, void-model offers a significant upgrade over traditional inpainting. You should consider integrating this model to automate complex scene alterations, especially when physical interactions are critical, reducing manual retouching and enhancing production value for film, advertising, and research applications.

Key insights

void-model performs physics-aware object removal from videos, maintaining realistic environmental interactions.

Principles

• Counterfactual video generation requires interaction awareness.
• Temporal consistency is crucial for longer video sequences.

Method

The model fine-tunes CogVideoX-5b with interaction-aware conditioning via quadmasks, using 3D transformers and an optional two-pass refinement for temporal coherence.

In practice

• Use quadmasks to define removal, overlap, and affected regions.
• Apply two-pass refinement for longer video sequences.
• Experiment with text prompts to guide inpainting.

Topics

• void-model
• Physics-aware Object Removal
• Video Inpainting
• CogVideoX-5b
• Quadmasks

Best for: Machine Learning Engineer, Computer Vision Engineer, AI Scientist, Creative Technologist, AI Engineer, Research Scientist

Related on AIssential

• 🍎Video Object Deletion🍎 👉Void by Netflix is a novel video object removal framework... — Void by Netflix offers physically-plausible video object removal for complex scenarios.
• NVIDIA’s New AI Just Made Video Editing Look Easy — Omnimatte Zero enables real-time, zero-shot video object removal by copying existing background information across frames.
• Rein3D: Reinforced 3D Indoor Scene Generation with Panoramic Video Diffusion Models — Rein3D integrates 3D Gaussian Splatting with panoramic video diffusion to restore and refine 3D indoor scenes from sparse inputs.
• Netflix's VOID shows video editing has finally learned the laws of physics — VOID introduces causal reasoning to video object removal, predicting counterfactual physics rather than just inpainting pixels.
• NTIRE 2026: A Multi-Track Study in Image Restoration — NTIRE workshops benchmark image restoration models against real-world degradations using rigorous evaluation protocols.
• OR-Action: Multi-Role Video Understanding with Fine-Grained Actions — Fine-grained OR action understanding requires explicit temporal modeling and multi-view feature alignment beyond scene graphs.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by HackerNoon.