In update 1.4 of its NVIDIA Broadcast software, the manufacturer has added a feature aptly called Eye Contact.
The application powered by AI and accessible to the brand’s RTX-stamped graphics cards now makes it possible to simulate via a deepfake the fact that the user’s gaze is focused on the camera, even if this is not the case in reality.
Follow the gaze of the AI
Since its launch, the NVIDIA Broadcast app has used artificial intelligence to make life easier for content creators, with varying degrees of success. Removal of ambient noise, setting up scenes without green screens, and now deepfake gaze to keep him facing the camera.
In its blog post announcing this new feature, NVIDIA indeed indicates that Eye Contact is aimed primarily at content creators ” looking to film themselves while reading notes or a script “. A way to maintain contact with your audience at all times, in short. The following video further illustrates how it works.
NVIDIA Broadcast update 1.4 received other additions and improvements, such as a vignette effect, coming to improve the effects of blurring, replacing and removing the virtual background. Two features strongly requested by the community are also finally arriving: the ability to change the horizontal orientation of the camera, and to take a screen print of what it records to create kinds of selfies.
Virtual eyes still a little crazy
NVIDIA also specifies that the newly arrived Eye Contact feature is still in beta stage, and hopes that the community will provide feedback in order to improve it. In fact, artificial intelligence still has some difficulty in effectively simulating looking directly at the camera lens.
The animation is still perfectible and greatly lacks stability. If our gaze strays too far from the lens, Eye Contact will no longer be able to follow, and the effect deepfake will disappear. There is no doubt, however, that the next updates will refine this new addition to NVIDIA Broadcast.
Finally, we can’t help but think that such a feature, with laudable intentions at first sight, will end up being used for malicious purposes. It would for example be possible to pretend to follow a videoconference with his colleagues or his grandmother, while being much more focused and interested in what is happening on his phone.
Source : NVIDIA
3