
Earthly Delights 3.2
Custom generative software, 2024
Private collection
Earthly Delights 3.2 honors experimental filmmaker Stan Brakhage and his 1981 film Garden of Earthly Delights, which alludes to 15th-century artist Hieronymous Bosch’s iconic painting of the same name. REAS references Brakhage through rapid cuts between abstract images and the use of plants as subject and source material.
REAS collected and made high-resolution color scans of native vegetation in the Rocky Mountains of Colorado—the same area where Brakhage worked. From 50 scans, several hundred thousand pictures were cropped and edited to create a data set used to train a generative adversarial network (GAN). The new botanical patterns and forms generated by the GAN have the organic feel of plants but don’t represent any species in existence.
The work unfolds as a nonlinear cinematic sequence. The generative software continuously rearranges fixed frames so that Earthly Delights exists in perpetual evolution. You never see the same arrangement twice.
A type of machine learning system where two AI programs compete to improve each other. Both programs are trained on a dataset of text, images, or sound. One program (the generator) tries to create convincing artificial data, while the other (the discriminator) judges its authenticity. Through this contest, the generator learns to produce increasingly realistic outputs.