Two architectural systems pass through Memories of a Network 2: an informally developed settlement in the Atacama Desert and a geometrically ordered English manor house.

Nussbaumer reconstructed fragments of both as digital 3D models and used image sequences of these virtual spaces to train a machine-learning system that generates images from visual relations within its source material. Outputs from different stages of the training were fragmented and recomposed into a new image space, which continued to change through the process of drawing.

Architecture remains recognisable without returning to either source intact. Each transfer reorganises what preceded it: architecture becomes model, model becomes dataset, generated images become composition, composition becomes drawing.

What persists through these translations is not an original place, but a set of relations from which another one can emerge.