Terence Broad
(un)stable equilibrium
August 14, 2025
SOLOS, in collaboration with Luba Elliott, presents (un)stable equilibrium, a generative series by Terence Broad that advances a rigorously data-abstinent approach to image synthesis. Drawing on the GAN paradigm, Broad’s method replaces external data with a second generator, creating a self-conditioning pair that imitates one another.
Inspired by the generative adversarial networks (GAN) framework, Broad developed an approach made up of two generative neural networks producing images to imitate each other, whilst competing to have more colour diversity. The title of the piece reflects the experimental process of finding a balance of randomness and stability in the training process.
"There’s a long tradition of artists working with film as material. Terry’s work is part of that lineage, but viewed through a neural lens."
Luba Elliot
Contrary to frequent assumptions that the system was trained on colour-field painting, no image dataset underpins this work. The images arise from a self-referential configuration in which a neural network is placed in perpetual feedback, iteratively predicting and refining its own outputs. Conceived as an explicitly non-derivative practice, the series stems from Broad’s early professional experience managing large, utilitarian datasets from traffic cameras in Milton Keynes, labour that exposed the conceptual limits of endlessly curating and reusing existing images. He resolved to step away from data collection and to refuse the appropriation that often shadows contemporary generative workflows.
"Most artists were obsessed with datasets and realism. I wanted to go the other way – toward abstraction."
Terence Broad
Instead of training a Generative Adversarial Network on a bank of pictures, Broad replaced the dataset with a second generator and closed the system. The resulting apparatus is tuned to resist collapse and maintain variation, allowing the paired networks to pursue each other’s signals in a state that tends toward equilibrium while continually deviating from it. While the surfaces may register as calm chromatic fields, the series articulates a position on machine authorship that is not built on borrowed material. (un)stable equilibrium operates as artwork and argument: a demonstration that generativity can be inventive without being extractive.
Bio
Terence Broad is an artist and researcher working in London. His research-led practice takes a hacking approach to working with generative AI systems that treats them as artistic materials. Through practice, he interrogates and makes visible the complex web of computational contingencies that underlie contemporary generative AI systems.