A team of scientists from the UK, Germany, France and the US created a neural network to develop a very particular project, using portrait videos.
The AI they created can transfer facial expressions from one person, displayed on video to another person in a different video.
That is, you can transfer from one video to another facial movement, implementing them in a second person, without losing the realism, as we see in the examples:
One detail to highlight is that you only need a few minutes to do this work…. analyze the video to be copied, recording the facial movements, taking into account critical points of the face, such as the mouth or eyes. It also considers other details, for example, the effects of shading or lighting.
That way, you can transfer the entire head position along with all facial movements and expressions, achieving a realistic effect. We can see some examples in the video they share:
The possibilities offered by this dynamic are many, such as film dubbing, virtual reality, simulations, etc. Although the creators also recognize that it could be used for other unethical purposes, they seek to be transparent about the dynamics and potential of this neural network.
We can see more details of this project in this document.