This demonstrates a binary GAN pre-trained on CelebA, where the prior p(z) is a Bernoulli(0.5) scaled so that its values are either -1 or +1.
Download the model file from here and place it in the models/ folder. The run an IPython notebook server to run widgets.ipynb.
Discussion: https://twitter.com/chris_j_beckham/status/1102607608131784710
https://github.com/christiancosgrove/pytorch-spectral-normalization-gan