Negative Poetry v.2.0: AI Color Theory is an experiment in negation using color-text relations and machine learning. Using OpenAI’s CLIP model – one of the first successful models trained on a large corpus of both text and images – we developed a system where 4096 colors (12-bit RGB) are matched to words of an English dictionary.
Each word and image are transformed into a 512-dimensional point space and then compressed into spherical 2D in a process that maximizes preservation of distances between pairs of points. Word-color relation is established by proximity and color negations are looked up by 180 degree rotation over the polar axis. By this method, we construct a color-word dictionary that facilitates word negations using geometry of language and color complementarity.
Collaborators: Studio RNDR (software & visualisations)
Support: Stimuleringsfonds Creative Industries Fund NL (Digital Culture grant), LAB[au] Brussels
Installation view: P,Oe?!e TOPO? / LAB[au], Brussels (15-31 Oct 2023)
Screenshots: (1) AI Color space mapped onto 2D surface of a sphere, (2) word-color relation index, (3) negation using 180 degree color oppositions
The development of this work and the overarching Negative Poetry v2.0 project was made possible by the Creative Industries Fund NL.
