I’ve spent a lot of time testing every AI video tool on the market, from OpenAI’s Sora to the latest Runway update. The pitch is usually the same: “Enter the prompt to get the movie.”
However, Netflix has secretly released a research model called VOID that is doing something completely different. Rather than building new worlds and scenes from scratch, rewrite what you’ve already shot. The technology is so good that you may never trust “real” videos again.
What is Netflix VOID?
VOID stands for Delete Video Objects and Interactions. At first glance, it looks like a higher-end version of “Magic Eraser” installed in Pixel 8 and Galaxy S24. When you select an object, it disappears.
Article continues below
But here’s the amazing part. VOID understands physics and cause and effect. In other words, while most editing tools simply “patch” the holes left behind in the background texture, VOID actually rewrites the logic of the scene to account for the missing objects.
Some tests on GitHub reveal what AI can do.
- Guitar test: In the research demo, the person with the guitar has been removed. With other tools, the guitar just floats in and out. VOID recognizes that the guitar is not supported and generates a frame where the guitar naturally falls to the ground.
- Collision test: If you remove one car from a head-on collision, VOID will not leave any ghost effects of fire or smoke. It “reimagines” the remaining car’s path as if the accident never happened, turning a wrecked car into a peaceful drive on an empty road.
Why is this the “end of reshoots”?
watch on
For companies like Netflix, this confirms the secret to significant cost reduction in the movie industry. Remember the infamous “Game of Thrones” Starbucks cup moment? Correcting this typically requires expensive frame-by-frame digital surgery.
With VOID, producers can simply remove unnecessary objects and let the AI realistically simulate what happens next. Whether it’s water splashing, dust settling, or nothing happening.
It’s more than just a small fix. Instead of having to bring back 100 crew members for reshoots, AI can now correct mistakes after filming is complete. You can also change the details of your story by removing key objects and recalculating the scene so everything looks natural.
Could you please try it?
The most surprising thing about this release is that Netflix has open sourced it. This model can currently be found at Hugging Face (under the Apache 2.0 license).
But don’t expect to run this on your MacBook Air. VOID is a beast. To run inference comfortably, you need a GPU (think NVIDIA A100 or H100) with at least 40 GB of VRAM. Additionally, it’s built on a 5 billion parameter version of CogVideoX and uses a unique “quad mask” system to tell the AI which parts of the physics it needs to recompute.
Take-out
In the past, a “visual receipt” was the final proof. Now that power is starting to disappear. Netflix has introduced a tool that allows you to seamlessly rewrite real footage so it looks completely authentic.
At the same time, AI “slop” is becoming more convincing than ever, flooding the internet with content that feels real but isn’t. The result is a world where seeing something no longer means you can trust it. The era of editable reality has officially begun.
to follow Tom’s Guide to Google News and Add us as a preferred source Get the latest news, analysis, and reviews in your feed.
