Deep Learning Recreates a Rembrandt
This project extracted data from various Rembrandt paintings to find the core features typical in one of his portraits. With that in place, a new painting was created in the style of the Dutch master.
Susanne Ramharter, you might find this interesting.
An additional note from a comment I just left on Kristin Drysdale’s share of this post:
I sense and understand the unease and disappointment people are expressing with this work. It’s worth clarifying a few things, just in case they’re not completely obvious. The first is that, technically speaking, this isn’t just a merger or morph of already existing Rembrandt paintings. The resulting painting – or more precisely, the resulting image – was generated from scratch. It would be analogous to someone studying a whole bunch of certain types of Rembrandt paintings, and then being asked to create a new painting, based on that training. Do that’s what we’re seeing here: a mimic, an echo, if the original creativity that once stuffed in the soul of Rembrandt.
There are plenty of examples of images where a technique like this could, and most certainly will, be used quite effectively without leaving us with this uneasy sense of loss. In this case though, what’s missing is that, like other great artists, Rembrandt was capable of using light in order to capture something of the essence, or soul, of a person. That’s because the man was likely capable of seeing that soul in the first place, and then capable of translating what he saw into pigments that he left behind for us on a canvas.
What I love about this project is that it was able to help us draw out this distinction.