Lost Memories from Latent Space is a generative art project that explores the hidden layers of how neural networks interpret and reimagine visual culture, as well as the latent space and internal mechanisms of neural networks. Drawing on a vast dataset of Renaissance paintings depicting mythological scenes, the series invites us to peer beneath the surface of machine vision.
The project begins with a CLIP-based semantic clustering of the corpus, dividing it into 150 conceptual categories. Each category informs the training of a dedicated conditional GAN model. From these, 150 unique generative loops emerge, each a traversal through latent space, visually unfolding the internal logic and aesthetic tendencies of the model.
At the core of each piece is a dynamic journey through the intermediate layers of the GAN architecture, revealing the evolution of internal neural representations as the final image takes shape. Feature activations - local ‘blobs’ of information - are tracked across multiple layers and channels to visualize their emergence, transformation, and interconnection. These clusters are analyzed and highlighted using color palettes sampled from the statistical distribution of the generated outputs, producing responsive heatmaps and abstract diagrams that echo the tones of the source material.
Occasionally full-layer visualizations reveal the hallucinatory inner logic of the network, glimpses of a visual world inaccessible to the human eye. These inner states shift from chaotic abstraction to ordered representation, allowing us to witness the AI’s act of visual understanding in real time. Sound is generated alongside the visuals by translating the shifting ‘blob’ structures into waveforms. As visual patterns grow or contract, the accompanying sound dynamically responds, creating a synesthetic experience that binds the technical and the aesthetic into a coherent flow.
Lost Memories from Latent Space charts a collective imagination built from the shared visual language of an entire cultural dataset. It invites viewers to reflect on the layered and often opaque nature of machine perception, and on the rich space that exists between human visual understanding and artificial synthesis.
This series is part of Orkhan's ongoing studio research.