Netflix open sources VOID, an AI model using physics for video removal
Netflix has open-sourced VOID (Video Object and Interaction Deletion), a new AI model for video object removal that simulates physical interactions post-deletion, significantly improving realism in post-production VFX cleanup. The model, built on Alibaba’s CogVideoX and fine-tuned with synthetic data, chains multiple AI systems like Google's Gemini 3 Pro and Meta’s SAM2. VOID requires substantial GPU resources (40GB+ VRAM) but is available under an Apache 2.0 license for commercial use and offers a Hugging Face demo and Google Colab notebook for accessibility.
Key Takeaways
- VOID uses a 'quadmask' system to identify primary deletion targets and areas physically affected by the removal.
- Built on Alibaba’s 5-billion-parameter CogVideoX, the model was fine-tuned with synthetic physics data from Google and Adobe.
- In human evaluation studies, VOID was preferred 64.8% of the time, significantly outperforming competitors like Runway (18.4%).
- Hardware requirements remain high for local deployment, necessitating an A100 GPU or equivalent with at least 40GB of VRAM.
Why It Matters
VOID addresses a critical bottleneck in post-production by automating complex 'interaction-aware' cleanup that previously required manual frame-by-frame correction. By releasing the model under an Apache 2.0 license, Netflix is decentralizing high-end VFX capabilities, potentially lowering the barrier for smaller studios. This multi-model architecture—combining scene reasoning (Gemini) with segmentation (SAM2) and generation (CogVideoX)—establishes a new technical framework for B2B video editing pipelines. Watch for the integration of these physics-aware capabilities into mainstream creative suites like Adobe Premiere Pro and DaVinci Resolve as open-source adoption scales.
Additional Context
The release of VOID coincides with a significant period of AI-driven transformation within the streaming sector's engineering roadmaps. According to Fora Soft reporting in October 2025, AI has shifted from a marketing focal point to a critical line item for technical infrastructure, specifically in areas like per-shot optimization and automated content understanding. This architectural evolution is reflected in the growing market for AI video tools, which reached a valuation of approximately $716.8 million in 2025 and is projected by some industry observers to grow more than threefold by 2032, per industry data from June 2026.
While Netflix open-sources technical components like VOID, the company is simultaneously accelerating the use of machine learning in its proprietary business operations. During the RampUp 2026 event in April, Netflix executives highlighted how the streaming giant uses AI to handle vast datasets for ad-tier campaign planning and personalized member experiences. This follows a broader trend among major streamers to utilize specialized AI models for technical efficiency while maintaining lean production teams. According to Alibaba Cloud Summit data from May 2026, the proliferation of specialized models—such as the CogVideoX-based systems Netflix utilized for VOID—is increasingly being enabled through large-scale cloud partnerships that offer high-performance computing resources to media organizations.
Read full article at zenvanriel.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source