Of interest to VFX researchers, but running it needs a GPU with 40GB of memory and licence terms are unstated.
Kenning's take
What it does
VOID (Video Object and Interaction Deletion) is a research model from Netflix and INSAIT that removes a chosen object from a video and also corrects what that object did in the scene. If a person holding a guitar is removed, for example, the guitar falls naturally instead of floating. It is built on the CogVideoX video model, with code on GitHub, weights on Hugging Face and an online demo.
- Click an object to remove it from video
- Fixes knock-on physical interactions such as falls and collisions
- Vision-language model finds the affected regions
- Optional second pass to stop object morphing
- Code, weights, Colab notebook and Hugging Face demo
Who it suits. Video researchers and technical VFX teams with access to a high-memory GPU.
Key facts
- Category
- AI models
- Made by
- Netflix (with INSAIT, Sofia University)
- Released
- 2 April 2026
- How to get it
- free research release on GitHub and Hugging Face; needs a GPU with 40GB+ VRAM to run locally
- Works on
- open-weights, Web
- In the UK
- Yes
- Free plan
- free code and weights; licence terms not stated in the README
FAQ
When was Netflix VOID released?
The VOID paper was posted to arXiv on 2 April 2026, and it was accepted at ECCV 2026.
What does VOID do differently from other object removers?
It removes the physical effects an object had on the scene, such as other objects falling or colliding, not just shadows and reflections.
What hardware does VOID need?
The README says it needs a GPU with 40GB or more of VRAM, such as an A100.
Facts from the maker's own pages and the sources listed. Checked 2026-10-11.
Similar tools
ACE-Step 1.5
Open-source music generation model that writes full songs with lyrics and runs on ordinary consumer GPUs.
AlphaGenome Atlas
Google DeepMind database of AI-predicted effects for every possible single-letter change in human DNA.
Amazon Nova 2
Amazon's second family of Nova AI models, covering reasoning, speech conversation and combined text-and-image generation.
Cisco Antares
Cisco's small open-weight models that find where known security vulnerabilities sit in a codebase, running locally.