If you would like to read about SGAN and using the SGAN project, get in touch with me and I will add you to a private repo.
I would recommend using my SGAN project, since it supports GAN, WGAN, GP-WGAN and SGAN in one project, with just one-line setting. And it is about 15% faster than the code you can find on GitHub. Besides, it has many practical features like plots that can be restored, auto checkpointing, etc.
If you are doing image generating work, GP-WGAN would be the best choice.
If you are doing stochastic modelling work, I recommend trying out our SGAN. As far as we can see, SGAN achieve state-of-the-art in this specific task with improvement of 5 ~ 10 times. Besides, GP-WGAN has 50% probability of failing in this stochastic modelling task. We did not observe such failures in our SGAN. (GAN and WGAN are even worse than GP-WGAN on this task)