Unofficial implementation of StyleGan2 paper: https://arxiv.org/pdf/1912.04958.pdf
This small GAN was trained for about 250k updates with batch size of 8 on 15k images from WikiArt dataset.
Random set of outputs:
A few cherry-picked better set of examples: