Vadim Epstein’s “Pixie” is a piece from an ongoing series using art imagery to explore the emergence of complexity. The neural network in this piece has been used here recurrently, reprocessing its own output without any additional inputs, as if jammed between broken mirrors. The models have been trained on a dataset that encompasses various figurative imagery — from photography to rough drawings — to enrich and intensify the visual experience. Part of the training data was synthetic itself: some of the source datasets were generated, adding another layer of mediation and thus distancing it even further from the real. Eventually, we observe an ever-changing, shape-shifting, loosely-controlled abstract flux, which appears more lifelike on its own, rather than the flashing snapshots of stiff flesh, stuck in convolutions.