I've never read any benchmarks around render time on sophisticated GANs, so maybe the answer to this is obvious, but: Is this showing a random selection from a set of offline-generated images, or is the GAN actually generating these on each request?
This work is impressive, I don't mean to take anything away from it, but if the author had to filter through 1000 images to select the 5 I saw, that's ... disappointing?