Using StyleGANs, I created an AI self portrait of myself from 1,000 selfies. This was an experimental project aimed at understanding how a computer would "see" me. It further acts as a metaphor for how the range of photos of me across the internet can be seen as an alternate digital representation of myself.
In this increasingly digital world, there are hundreds of photos of each person living online, whether posted by them, their friends, or just being in the background of other's photos, these photos soon start to form an alternate virtual identity of someone.
Facebook announced that it bought an AR startup, Scrape Technologies, which is aiming to build a 3D map/replica of the entire world. These announcements raise the question of who will inhabit these digital spaces and if we are building digital replicas of our environment, what is to say we won't start building digital replicas of ourselves? This project is an experimental exploration into digital representations of oneself using AI.
Output 1 - Trained off of 450 photos
I took 450 photos of myself and trained it using a StyleGAN.
The outputs look creepy and I can see some of the details of my features coming through, although not entirely. I especially love the one of me in a fancy pink hat.
Training in progress:
Output 2 - Trained off of 1000 photos
I increased the amount of data to 1,000 photos for training and saw drastically better results:
In the future, I would be curious about playing around with different facial expressions or different accessories to see the variety in outputs.