Shiryaev told My Modern Met that he was thinking about creating such a neural network for years and that he had only now found the time to explore the concept: “I finally had the time, skills, and desire to pursue it. Basically, I am an artistic machine-learning person with a soul.”
His neural network was able to analyze seven famous paintings and generate realistic faces from them, approximating the way these women would look in real life.
Some results are incredibly accurate, like Frida Kahlo or Botticelli’s Venus, while others sport similar features but are not identical.
Either way, all are fascinating to study in detail:
Denis explains his process and showcases more of this project in the video below: