Microsoft researchers presented their latest innovation based on artificial intelligence: the generation, from a simple photograph, of a video of an individual speaking in front of a camera. These faces appear incredibly realistic in terms of the reproduction of emotions and expressions once animated. Modeled on a voice-over, the lips, eyebrows and other elements move so realistically that we understand the concerns about the phenomenon of deep fakes that such technology can generate.
Animate portraits realistically
Called VASA-1, the framework developed as part of this project is capable of generating high definition videos with a frame rate of 40 frames per second. Several complex technologies associated with deep learning were used by the Mountain View firm’s teams to develop VASA-1. The latter works to animate portraits: by providing it with a photo and an audio track, the AI is able to produce a video.
This example, taken from the presentation page on the Microsoft site, illustrates how powerful (and worrying) the technology being developed is. To view all the videos, we invite you to consult the dedicated page on the Microsoft website.