The images generated in the Postmachineism series were trained on a dataset of several thousand images of minimalism, color fields and sculpture by Donald Judd, Sol Lewitt, Dan Flavin, Dan Graham, Carl Andre, Ellsworth Kelly, Eva Hesse, Frank Stella, Mark Rothko, Barnett Newman, and more.
The images in Volume One are the first round of training results.
The images that were created after the initial training seem to have a very heavy bias towards works by Judd and Rothko with a sprinkle of Barnett Newmans “zips”.
Lots and lots of color fields and sculptural bricks and block type shapes. I am not sure where the “face” like images came from since there were no people or animals used in any of the training images.