Collection of generative models in Tensorflow (github.com)

The summary of all different GANs and especially the structure of the plots is very helpful. However I am unsure about the mnist and fashion-mnist results. I can't see any significant difference between the methods. Maybe the CelebA (human faces) database would be more helpful to see the differences.

Curious: Does anyone use GANs for anything of value, or are they just gimmick NNs?