Gilles Dubuc is an artist living in the south of France. He learned to program in the late 1990s to join the demoscene, and his first programs were 2D and 3D intros, short real-time audiovisual pieces, written in Pascal and assembly. He has made products and platforms for artists for nearly twenty years: inspi.re, a photography community he founded and ran from 2008 to 2013, then deviantART from 2009 to 2013. Founded in 2000, it became the largest online community of artists, and for a generation of digital artists it was the place where they first showed their work and found an audience. From 2023 to 2026 he was the CTO of Layer and co-created the Layer Canvas, a display made for living generative art. He co-designed its hardware with DEMO, oversaw its manufacturing, and is a named inventor on its patent application.
Latent: press kit
Texts, images and the film for press and exhibition use. Images may be reproduced with the credit line below. For anything else, write to gilles@dubuc.fr.
Press PDF
All the texts and a selection of images in one document.
Download the press kit (PDF)Texts
Short description
A real neural network, untrained and given nothing to look at, left running in the dark. Everything on the screen comes from it: the forms lie where its answer falls to zero, the colours show the activity of its hidden neurons, and the bands of light are thoughts crossing them. The same dream never comes back.
Long description
Latent shows what a machine sees with its eyes closed. Inside the piece runs an actual neural network, small, a few hundred weights in two layers of sine neurons, never trained and never shown the world. Its only input is a slow drift of numbers from inside itself, the way a sleeping mind is fed by its own currents.
Nothing in the image is drawn by hand. The forms are the network’s own: every membrane, shell and fold of coral is the place in space where its answer falls to zero, and a million grains of light condense out of the dark to find it. To find it they ask the network which way is down, and they ask it the way machines learn, by backpropagation, here put to work for dreaming rather than learning.
The colours are the activity of one hidden neuron; the bands of light that sweep across the forms are the level sets of another. Every minute or two the weights drift toward another mind, and one dream turns slowly into the next. What you see is the dream itself, shown without translation, and not a picture of a machine dreaming.
Details
- Artist
- Gilles Dubuc (France)
- Title
- Latent
- Year
- 2026
- Medium
- Real-time generative artwork: software, an untrained neural network
- Duration
- Infinite; the sequence never repeats
- Display
- Any screen and aspect ratio, rendered live at full resolution
- Variations
- Set by a seed, a palette, the pace of the dreaming, the distance of the view and the fineness of the forms
- Credit line
- Gilles Dubuc, Latent, 2026. Real-time generative artwork.
Images
Rendered by the work itself at full resolution: 7680 × 4320 for the wide images, 4320 × 4320 for the square ones.
Portraits of the artist
The film
Latent, the film. 2026. 7 minutes, colour, sound. Narrated in English, with subtitles. A companion to the work, narrated over the work itself. The film follows the network from its first random connections to the forms it dreams, and shows each step in diagrams computed from the network as it runs. Watch the film. Broadcast masters (4K, 60 frames per second, SDR and HDR10) are available on request.