Audio-Reactive, Machine-Learning-Generated typography generated along a 512-dimensional axis.

With a new interest in Machine-Learning and the eternal love for typography, I combined the two: I trained a database of 4,000 fonts installed on my machine using a Stylegan 2 model, then exported using Michael Arafiz's open-sourced Lucid Sonic Dreams notebook. The results were absolutely astonishing — type could suddenly morph in ways I had not previously believed possible — from upper to lowercase, serif to sans, bold to italic, expanded to condensed, hairline to ultrablack — all transitioning smoothly without any awkward in-between states. The type comes to life — something meant to be read is dancing instead.

Through this project I realised the potential of artificial Intelligence as a tool in the field of design. It could generate 10s of videos in a second, each absolutely unique, each beyond the scope of any human animator. As opposed to our experienced world of 3 — or 4 — dimensions, the AI model operates in a space of 512 of them, well outside the constraints of our imagination. Each category is mapped onto a point and the Lucid Dreams notebook traverses this latent space; interpolating through space and resulting in, say, the condensed double-storied lowercase 'g' smoothly transforming into an italic flared seriffed upper.

All the while expressing the depth, harmony and emotion of beautiful music.