Experiments in generative adversarial networks.
It's been quite difficult to find the right balance between getting MiróBot to create images with realistic looking shapes and blocks of colour, and just copying originals.
These examples were trained on a subset of Miro paintings that are predominantly red, yellow and black, with rounded forms present in the image.
MiróBot seems pretty good at generating Rothko paintings though.
