Explorations in machine learning and latent space type design

Technology has always been a driver of creative exploration and innovations. As designers, we are currently living in one of the most exciting eras where technology machine learning and artificial intelligence are making enormous leaps forward opening up new exciting avenues. In this project, we wanted to explore the latent space between common typographic defined styles to see what was hidden in the in-between. The hidden space has a structure that can be examined, such as by interpolating between points and performing vector arithmetic between points in latent space which have meaningful and targeted effects on the generated letterforms. We used a Generative Adversarial Networks, or GANs, trained on raster renderings of glyphs input into an architecture for training neural networks for generating images.

Trained on over one thousand steps the project has led to Enterprise Ireland funding to explore the creation of variable fonts based on the outcomes of the project. It has also had a massive impact since being posted online garnering over forty thousand impressions. While type designers denote abstractions in the stroke weight similar to that of a calligraphic pen, the GAN looks at the spaces in-between to create new moments and explorations of the type, that are both hypnotising to watch and groundbreaking in terms of form.