Generating anime faces with deep convolution neural networks.
Generating with a GAN network
Montage of 120 faces generated by a GAN. The model was trained for 71 epochs and the results are pretty cute(-ish)
An interactive web demo lies here.
Generating with a VAE network
Montage of 90 faces generated by a VAEGAN. It's a lot blurrier than GAN, possibly due to the high weight of the L2 reconstruction loss. Due to performance and memory constraints I wasn't able to test this for long, however, feel free to try tuning the hyperparameters.