Samsung’s AI Center in Moscow has used a new trend in artificial intelligence technology, adversarial learning, to make still images move. Adversarial learning involves training two AI networks, one to generate the fake videos and one to spot the fakes from among real ones. The systems essentially force each other to get better and better at their jobs, thus generating a better end result.
The system generates these videos by looking at a still image or images of a person’s face and identifying the position of landmarks like eyes and nose. It looks for the same things in a video of someone's, anyone’s, face and tracks how these landmarks move. It then applies these movements to the landmarks in the still images to make them move.