The images below were generated from a single photograph by
(called "content") and paintings from famous artists (called "style"). These images are combined together using style transfer with a deep neural network that extracts patterns, and another algorithm that recombines these patterns together.
It's a new algorithm—yet to be finalized and published—that generates all
paintings (without new code or retraining), based on original research from the University of Tuebingen
. Also see this publication from the University of Oxford
that was the inception for it all. Follow
for updates and details when we publish the new improved algorithm.