See how Samsung's AI animates Mona Lisa - or any other picture

The neural network is based on the principle of adversarial learning. It uses one or more images

of a person - the more, the better the result - and passes it through a “facial landmark tracker”, determining the position of the eyes, eyebrows, nose, lips and jaw.

The neural network is divided into three parts. The first, “lining up,” analyzes the original image and makes up a reference system based on it. The second, “generator”, based on this scheme, creates short videos, and the “discriminator” selects the best of the created segments.

Previously, physicists from the University of Queensland createdreproductions of famous works of art, such as Da Vinci's Mona Lisa and Van Gogh's Starry Night, on a quantum canvas as thick as human hair. It was possible to do this by projecting light onto Bose-Einstein condensates (BEC).