Ghosts

This is the extended opus from the ongoing series, exploring complexity emergence, based on the feedback loops.

Multidomain image-to-image transforming neural network StarGAN2 has been used here recurrently, reprocessing its own output without additional inputs. The models have been trained on both figurative imagery and abstract art, to enrich and intensify visual & semantic experience. Moreover, part of the training data was synthetic itself: few source datasets were generated with custom StyleGAN2 models, adding another layer of mediation to distance farther from the real.

What we eventually get is an ever-changing shape-shifting loosely controlled abstract flux, which appears more lifelike and expressive on its own, than obscure resemblance of the origin flesh, stuck in neural convolutions.