Jasper Zheng (Shuoyang)
CV


Manipulated Network: quirks and oddities from corrupted generative models


Type

Algorithmic Digital Images

Materials

TensorFlow, generative adversarial networks, 29.7 x 21 cm canvas

[pdf]



Mar, 2022

During the re-implementation of Network Bending, I’ve come across plenty of quirks and oddities generated from corrupted models. These models produce impossible, distorted but realistic images, diverted from the original outputs and sometimes lead to aesthetic preoccupations. Although utilising machine autonomous, creating this collection is still like a craft: I marked configurations that cause these semantically meaningful results, tweaked and built the operation template again, and then re-examined the subsequent images.

The decisions can be either arbitrary or intentional, sometimes deliberate, all in order to create potentially aesthetic results.