Nvidia's AI now fakes eye contact on your video calls

Anyone who records video or spends half their day on calls knows the pain of glancing down at notes: the moment your eyes dart away, everyone knows you're reading a script.

Nvidia rolled out Broadcast version 1.4, featuring an AI tool called Eye Contact that solves this problem by essentially deepfaking your eyes in real time. Even if you're looking at a secondary monitor or reading a prepared script, the software replaces your eyes with simulated ones that stay locked onto the camera.

A pair of AI eyes that never look away

Nvidia pitches the new beta feature directly at content creators who want to record themselves without having to memorize a script or look away from the lens.

The tool detects your gaze and synthesizes a set of eyes matching your natural eye color and lighting, continuously pointing them forward. There is a built-in disconnect fail-safe: if you turn your head too far away from the webcam, the effect automatically disables to prevent body horror glitches.

In hands-on testing, however, the uncanny valley is alive and well. While the feature looks impressively natural when holding still, quick movements can cause the digital eyes to dart around at unnatural, superhuman speeds, accompanied by occasional jarring pop-ins. Nvidia openly admits this is a work in progress, citing millions of eye colors and lighting combinations, and is asking users to submit test videos to train the model.

Alongside Eye Contact, the Broadcast 1.4 update adds an Instagram-style vignette effect and improves its existing virtual background blur, replacement, and noise-removal features. The update is free for anyone running an RTX GPU.

From subtle warps to generative replacement

AI gaze correction is not entirely new.

Apple began experimenting with an Attention Correction feature inside FaceTime back in 2018, which still exists in current iOS versions. Microsoft also baked an eye contact setting into Windows 11 for hardware equipped with a dedicated NPU.

However, those earlier systems focused on subtle geometric shifts to offset the awkward angle between a screen and a webcam. Nvidia’s approach is far more aggressive: it uses raw generative horsepower to paint completely simulated eyeballs directly onto your face. It mirrors what the company did with DLSS 3 generating whole frames in games, only applied to your webcam feed.

Why it matters

For solo creators and streamers, this tool practically kills the need for an expensive teleprompter setup. You can place your bullet points anywhere on your screen and maintain perfect direct connection with your audience, completely free as long as you own an RTX card. But we wouldn't recommend using it to fake your way through internal team meetings just yet; when the algorithm misfires, the glitchy cybernetic stare is far more awkward than just admitting you looked away.

Strategically, Nvidia continues to flex its desktop AI dominance. By turning gaming silicon into everyday video-production infrastructure, it creates sticky software lock-in that AMD and Intel still struggle to replicate.

The real catch is social. Normalizing consumer-grade real-time deepfakes in routine communication chips away at authentic interaction. When you can no longer trust whether someone is actually looking at you through a screen, video calls lose the very thing that made them feel personal.