Ultrakurzzeitreligionenmachine learning | design research
Year: 2022
Role: Motion Design
Music: Edvard
Tools: Blender, StyleGan3, Meshroom
The bachelor thesis "Ultra-short-term religions" deals with the question of the visuality of text-to-image generators and asks to what extent this aesthetic is sustainable. The title, a term borrowed from the fashion discourse of Norbert Bolz's "Fashions are not Trends", implies the ephemeral, short-lived nature of aesthetic phenomena. The possibility of using different AI models to convert text into images develops a new visuality that is analysed in this work. For this purpose, the neural network StyleGAN3 is used, which makes it possible to synthesise similar images from image data sets. This creates an abstraction of the visuality of the images generated by text. The image base for each dataset was collected from social networks in order to involve the users of the text-to-image generators in the development of visuality. The newly synthesised images are used as the basis for 'digital fashion' objects, creating a new contextualisation of objects and thus a new contextualisation of visuality.