Imagery

Most every famous painting has a model behind it, even if only sketched by the artist. As the painting was created, of course, the image of the model became the image in the painting. Since few of even the greatest artists painted what we would today call a "photorealistic" style, the image on canvas doesn't necessarily tell us what the person really looked like. They're probably close enough that we would recognize them if we met them, but there would still be differences because the painted image comes from the artist and is influenced by him or by her.

Comes now one Denis Shiryaev, an artist who used a neural network to examine the faces found in these paintings and create from them possible faces of the models used. Since the only evidence we have for what these people looked like are the paintings themselves, there's no guarantee that the images the network developed represent what those models "really" looked like. But the process produces images that look like photos of real people, and so we can see what kind of face might have inspired the artists' imaginations.

The network-developed images range from beautiful to ordinary, which helps us understand both the role that the inspiration and talent of the artist plays and the fact that sometimes the ordinary is a step on the way to the extraordinary.