NVIDIA Broadcast uses AI to correct streamer gaze

Those who do lives on the internet are well aware of the importance of always looking at the camera – one of the most important ways to show an audience that it is the focus of communication. However, for obvious reasons, it's not so simple to always keep an eye on the webcam – but NVIDIA is working on a way to mitigate this issue through AI within NVIDIA Broadcast, its tool for streamers.

NVIDIA Broadcast's AI-powered solution for streamers to always look at the camera

The AI ​​solution was implemented on NVIDIA Broadcast, the company's solution focused on helping streamers live. The tool, called Eye Contact (eye contact, in free translation) is available from version 1.4 of the program, and replaces the eyes in the image with simulations that are always aligned with the camera.

In the announcement of the novelty, NVIDIA explains that the focus of the function is to help content creators who need to record something with a script, for example, and instead of decorating or showing the look coming out of the camera, the tool always leaves everything aligned – with up to even the color of the simulated eyes, according to the company, copying the real ones, and the video disconnecting if the user leaves the area of ​​operation of the modification.

O The Verge posted a short video showing how the tool works, which we shared above. In the content, a part of the screen is missing the eye contact active, while the other is already testing the new function – with it even seeming to work well, although a certain tone of Strangeness Valley in content is inescapable, at least in the current application.

How to use Eye Contact in NVIDIA Broadcast

Using Eye Contact is easy — with the first step being downloading the NVIDIA Broadcast on the solution's official website.

After downloading and installing NVIDIA Broadcast, just open the application, and then in its main menu, which is next to the preview of the camera image, choose the Visual Contact (Beta) option in the last dropdown menu and start using the solution.

Dilemmas upon dilemmas with NVIDIA Broadcast and AI

At the same time that the NVIDIA sells the function as something that should be used for recording and the like, it's undeniable that the eye contact it will also be used by users for several other situations – and these can create certain dilemmas and add more layers to discussions about technology applications in our daily lives.

For example, an employee can pretend to be paying attention in a virtual work meeting through the tool, with his eyes at the other participants always appearing to be focused on the screen while he is actually reading or doing something else on the computer – and if so situation of correcting the gaze is discovered, what are the employment consequences?

However, at the current level of technology, perhaps this dilemma does not exist yet - as we can see in the demonstration of the The Verge above, as much as the result is impressive and the fact that the possibility of using this tool is something that shows advances in the field of AI, there is still a certain strangeness in the result that shows the viewers of the video that it is not natural.

In addition, still using the demonstration as a basis, it is possible to notice that the detection of the user's face is still not 100% accurate - at several times, the "filter" disappeared due to a more sudden movement of the reporter's head, so it still it seems far away to think that the eye contact do NVIDIA broadcast have a lot of adherence in its initial state.

But the future, as always, is unpredictable, so who doesn't guarantee that in future updates the NVIDIA don't improve the tool further, to the point where it's hard to tell the natural eyes from the eyes created by the tool's AI – for now, we can only sit back while the technology continues to evolve and watch ways of staring.

Source: , The Verge

See also other features

What to expect from the creator economy in 2023