Final Project: _portrAIt
Check out this project here!
For the final project of my Creative Coding course, I decided to work with machine learning and p5.js, specifically using Generative Adversarial Network models available through Runway ML.
I initially became interested in GANs as an art tool after observing some very interesting output from Casey Reas, George Muncey, and an artist under the alias of hoodwinkedfool. Drawing from these inspirations, I sought to create an experience that would allow users to interact with a GAN in a reflective way, offering their own visage to it and, in return, receiving generated interpretation of the input.
I find the images that GANs create to be both intriguing and disturbing, as they remind me of the relationship we have with the technology that we use. Especially under the current circumstances, most of us are showing our faces, as well as sharing every bit of information about ourselves, to some form of smart device almost at all times. These generated images are how I imagine these devices interpret the information that they receive. I think these images illustrate the juxtaposition and dichotomy between the humanity we attribute to modern technology through our intimacy with it and it’s intrinsic lack of humanity or the ability to understand humanity that we share with it. With this project I want to make this message more personal by generating these images in real time with the user as reference.
After experimenting with StyleGAN and Latent Space Walks, I found that training a GAN with user input was too taxing in terms of both time and resources, and discovered BigBiGAN in my search for a more instantly gratifying alternative.
In essence, my final project is a user interface built with p5.js, HTML, and CSS that allows the user to capture images of their face, submit them to a BigBiGAN model hosted by RunwayML, and receive the generated images that the GAN outputs.
The hosted version of the program can be found here. If you are interested in experimenting with it, please contact me for the API key.
The local version of the program, which offers a slightly different experience (in that the user receives real-time video interpretation of their image) can be found here. In order to interact with this, download the Runway ML desktop client and run the BigBiGAN model with the input set to HTTP. The program will handle the rest.
