It's copy-pasting parts of the training images over and over.<p>In figure 8 of the technical report [0], compare the hair in images (0,0), (2,0), (3,0), (3,3), (4,4).<p>The paper suggests their method generates copyright-free images, yet they are very obviously derived from the input images and you can identify the parts of individual input images that are mashed together to form the output.<p>All in all their method seems to be performing "obfuscated memorization," in the sense that the generated images are scrambled up enough to fool their plagiarsim-detector loss function.<p>[0] <a href="http://make.girls.moe/technical_report.pdf" rel="nofollow">http://make.girls.moe/technical_report.pdf</a>