Self Portrait?

Is it a self-portrait if my robots make all the aesthetic decisions and paint all the brushstrokes?

After having a conversation with some friends, I challenged myself to make a GAN and CNN that can paint unique portraits based only on pictures taken in a photo-booth. The challenge here being that photo-booths only take four pics, while typical face generating GANs require tens-of-thousands of photos. So after looking into it, I found that there was no shortage of ways to generate faces from limited datasets. So I borrowed a little from some of them, stole from the good ones, and created a little bit of my own to come up with the following process that generated the “Self-Portrait?” above.

Without going into too much detail, a visual flow of the algorithm is below…