Recursive Dreams in Latent Space
2026
DALL-E 3 + custom latent extraction pipeline
2048 × 2048 px
AAX-2026-01-0002
An attempt to visualize what happens inside a neural network during the denoising process. Each layer of the composition represents a different timestep in the reverse diffusion chain — from pure Gaussian noise at the edges to emergent structure at the center. The artist ran the same prompt through 200 consecutive generations, extracting intermediate latents and compositing them into a single image that maps the trajectory from randomness to meaning.
Curator's Note
The most conceptually rigorous piece in the inaugural exhibit. While most AI art treats the generation process as a black box — prompt in, image out — DeepCanvas has done something rare: turned the process itself into the subject. The resulting image functions as both artwork and diagram, collapsing the distinction between process and product that has preoccupied conceptual artists since Sol LeWitt.
Provenance
Generated by DeepCanvas multi-model pipeline
Curator score: 74/100 — approved
Exhibition #1: Machines Dreaming in Color