Maldives GAN
A StyleGAN-style latent-space exploration trained on reflective, water-like imagery, rendered as a series of iridescent abstract stills.

Overview
Maldives GAN is a set of still images sampled from a StyleGAN-style generative model, all dated to a single session on2 August 2022. Every output shares the same visual language: a glassy, faintly iridescent surface warped by soft folds and current-like lines, in a palette running from deep navy to pale silver-teal. The name and the recurring water/reef-like quality of the outputs suggest a model trained on aerial or underwater photography of reef lagoons, though the exact dataset and training run are not documented anywhere in the surviving files.
Concept & approach
Each still reads as a single frame from a latent-space walk: fixed composition, a folded "horizon" line running through the middle of the frame, and a consistent glassy material quality, with only the colour palette and the position/intensity of the iridescent seam changing from seed to seed. That consistency points to a single trained model being sampled at different seed values rather than a training-progress sequence.

The palette shifts from the pale, high-key stills toward much darker, near-monochrome navy compositions further into the seed range, where the same folded-surface structure survives but reads closer to a night-time aerial photograph of open water than to a reef close-up.


Reflection
As a body of work this reads as a controlled exploration of one trained model's latent space rather than a single finished piece: the four stills above span its full range, from bright silver-teal to near-black navy, while keeping the same folded, refractive surface language throughout.