Landscape Forms


Landscape Forms uses 3D models of trees produced using a voxel-based Generative Adversarial Network (GAN). Using a poetic audio-visual animation, it cycles through the training epochs of the GAN to animate the process of the system learning to produce the three-dimensional form of trees. This artwork is formalist in nature and explores how the GANs learning three-dimensional forms can be considered within broader art historical questions of figuration and abstraction.

Expressed using a variety of animation techniques, including particle morphing, vertex shape keys, and latent space walks, this work explores forms produced by the GAN that appear reminiscent of modernist sculptors such as Henry Moore or Constantin Brâncuși. It structures these forms within compositions inspired by Romantic landscapes by Phillip Otto Runge, Theodore Rousseau, and Caspar David Friedrich and literati compositions by Ni Zan [倪瓚] and Ma Lin [馬麟]. By situating this exploration of machine learning as a pathway to form and figuration within traditions concerned with the poetic function of landscape, this animation seeks to contribute to a broader consideration of how new technologies inform classical formal questions in visual culture. This work includes a musical accompaniment by Roberto Alonso Trillo, that uses a GAN system to generate a reworking of Romantic music.

About the artist

Peter Nelson is a visual artist and academic working at the intersection of landscape theory, computer games, and computer graphics. Nelson is engaged in a prolonged consideration of the history of landscape images, how they are remediated by technological shifts, and how these shifts absorb and reflect changes in our relationships with the physical environment.