Interesting, especially the performance on bigger images, and it looks like a low-effort modification of many standard GAN losses. Seriously, I want to give this a try right now. What do GAN researchers think about this paper?

It should also be appreciated that it comes with code and a short blog post.