Dutch painter Rembrandt van Rijn died in the 17th century, but he just came out with a new painting. That is, an algorithmic version of Rembrandt did. 

A project called The Next Rembrandt is using machine learning to create new art in the Dutch painter’s style, based on analysis of every single one of his paintings, pixel by pixel. Each painting was analyzed carefully for characteristics about its subjects (by gender, age, and face direction), geometry, composition, materials, and even the depth of the paint layers.

The software came up with this list of characteristics for a prototypical Rembrandt:

Here’s what physical features in a Rembrandt portrait look like, on average:

And here is the final result: the most Rembrandty painting that Rembrandt didn’t paint: 

Pretty impressive. 

[h/t Kottke]

All images via YouTube