Reveries by ORKHAN

Rêveries by ORKHAN, explores how the latent structures of artificial neural networks can be repurposed into generative painting engines: leveraging convolutional neural networks (CNNs) designed for object recognition, Mammadov reorients the model from classification to creation, coaxing it to synthesize intricate textures by capturing correlations among feature maps across its layered hierarchy. As these textures evolve, the network distills the statistical essence of natural imagery while abstracting away familiar forms, revealing a dual function—both a creative instrument and a lens through which to glimpse the deep representational spaces that underlie artificial perception. Visit orkhan.art/reveries for more info.