Twin Peaks
A GAN trained on parsed frames from Twin Peaks, distilling the show's recurring palettes and compositional symmetries into entirely new, invented frames.

Overview
A short generative AI experiment usingTwin Peaksas source material, made in May 2022. I assembled a dataset by parsing frames from clips of the series, then trained a GAN on the resulting imagery to explore what the model could learn, and invent, from that visual world.
Concept
The GAN revealed unexpected patterns in the show's cinematography — recurring colour palettes, compositional symmetries, and atmospheric qualities — which the model distilled and recombined into entirely new frames that never appeared in the source footage.

Reflection
The result is less aboutTwin Peaksspecifically than about what a GAN surfaces when trained on a single, stylistically consistent source: the model doesn't reproduce the show, it reproduces its visual grammar.