Product reviews, deals and the latest tech news

AI-Driven Eye Contact is now available in NVIDIA Broadcast 1.4

The “eye contact” functionality has been made available to the general public via an upgrade to NVIDIA Broadcast. If you’re using an NVIDIA RTX graphics card, you may fool onlookers into thinking you’re staring at them straight on by using this feature.

Ordinary people don’t have access to teleprompter-style equipment, which is required to stare squarely at the camera during video filming. NVIDIA Broadcast is a piece of software that use AI (Artificial Intelligence) to manipulate the video feed and redirect your sight.

This feature has great potential since it is both practical and simple to implement, and because eye contact with the camera increases viewer engagement (you just need to turn it ON).

Surprisingly, it works rather well if your gaze is around 35 degrees in front of or behind the camera’s lens. It’s recommended that you provide your best effort to stare directly into the camera for the most professional results.

Currently, there are a few small bugs. If you blink, for instance, the AI may lose track of the direction in which your eyes are looking for a split second, causing them to snap in a new direction. If you tilt your head too far away and then return, the same thing might happen.

NVIDIA may be able to resolve this issue in the future by storing the position of your gazing eye just before to the blink. Besides that little flaw, I haven’t had anybody realise I’m using it on calls, and it does its job enough. At least not enough to make a remark about it.

To a large extent, the work of the NVIDIA AI Research team has led to the development of NVIDIA Broadcast and its features. It’s great to see regular updates to the freely accessible software, especially since several webcam-related capabilities were exhibited as research projects a few years ago. Also, it’s accessible to programmers.

Having both this and NVIDIA’s upcoming AI Face Alignment capability would be fantastic for making the effect appear more realistic. Realigning only the face is one thing, but the face and body together are another. It seems like another hurdle that has to be cleared before that function can be made available to the public.

Workloads like these demonstrate that GPUs are finding usage by a far wider audience than just those interested in 3D graphics for gaming or CAD.