Overview
- Netflix published its VOID research model and paper on Hugging Face for public testing, and Tom's Guide reports the code is under an Apache 2.0 license.
- The tool, built on a vision‑language model, removes people or props and then regenerates the scene to look physically plausible, such as smoothing pool water or rerouting a car so a crash never happens.
- Netflix reports that in a 25-person study, reviewers preferred VOID’s edits about 65% of the time over tools like Runway and ProPainter.
- The shared demos focus on simple, sparsely populated scenes, and coverage notes it is unclear how well the system handles crowded city shots or complex interactions.
- Running the model needs a high-end GPU with roughly 40GB of VRAM, and the release lands as Netflix pursues creative AI tools, including a reported acquisition of Ben Affleck’s startup InterPositive.