
What if filmmakers no longer needed to reshoot expensive scenes or even stick to the original storyline once filming is done?
That’s exactly the future Netflix is hinting at, as it quietly steps into the AI race with a new video model that could fundamentally change how movies are made.
Called VOID, short for Video Object and Interaction Deletion, the model can remove objects from a video and then intelligently rebuild the scene as if those objects never existed in the first place.
Think of a crash sequence. A car slams into a truck, explodes, debris flies everywhere, the kind of scene that costs millions to shoot. Now imagine being able to erase the truck entirely and have the AI seamlessly generate a version where the car just drives away, as if the crash never happened.
Unlike traditional editing tools that simply erase objects, VOID goes a step further. It understands how objects interact in a scene and then reimagines how everything else should behave once something is removed. So if a person jumps into a pool and creates a splash, VOID can remove the person and regenerate the water to look perfectly undisturbed.
The model, developed by researchers affiliated with Netflix and academia, is based on a vision-language model (VLM) architecture. It analyses both visual cues and contextual relationships to generate what researchers describe as “physically plausible” outcomes after edits.
Existing tools like Runway, ProPainter and DiffuEraser already allow creators to make changes to video. But Netflix’s team claims VOID significantly outperforms these alternatives. In a human evaluation study involving 25 participants, VOID-generated outputs were preferred nearly 65% of the time, far ahead of competitors.
For studios, it could mean fewer costly reshoots, faster production cycles and greater creative flexibility. For creators outside Hollywood, the barrier to high-quality visual storytelling could drop dramatically.
And Netflix isn’t keeping the tech locked away. The company has made VOID available on Hugging Face, allowing developers and creators to experiment with it.





