Rêveries: Joseph's dream

Rêveries, 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.

Controls
[ 1 ] • Save Image
[ 2 ] • Start / Stop Video Render
[ 3 ] • Toggle 3D Model
[ 9 ] • Save Neural Painting
[ 0 ] • Start / Stop Simulation
[ – ] • Decrease Simulation size
[ + ] • Increase Simulation size
[ Click ] • Toggle Interaction
[ Double - click / Enter ] • Toggle full Screen
[ Space ] • Play / Pause Simulation

Visit orkhan.art/reveries for more info.