1 / 5
(un)stable equilibrium 2:2 (2026) is the second piece in the second series of (un)stable equilibrium, which is an ongoing series of experimental artworks that are borne out of an artistic practice developed around the training of generative neural networks without any training data.
In this series of works, the process of training is borne out in the video pieces — with each work in the series documenting a separate training run. This process shows the generative network attempting to converge to a fixed point that is undefined, caught in an endless, unresolvable search for equilibrium.
In (un)stable equilibrium 2:2, three points in latent space are shown in parallel over the course of training.
This work is created using a bespoke algorithm created for training generative neural networks without data, based on the architecture and an adaptation of the training code for the original StyleGAN.
A detailed technical description of the implementation is given in: Broad, Terence. "Still searching for an (un) stable equilibrium: visualising the process of training generative neural networks without data." XAIxArts: 4th international workshop on eXplainable AI for the Arts at the ACM Creativity and Cognition Conference, 2026.
Terence Broad is an artist and researcher working in London. He is a Senior Lecturer at the UAL Creative Computing Institute and has a PhD from Goldsmiths, University of London. His research-led practice takes a hacking approach to working with generative neural networks that treats them as artistic materials. He has built frameworks that allow for the expressive manipulation of generative neural networks and developed data-free approaches to training and configuring neural networks that open up new possibilities beyond the conventional imitation-based learning.
