These are the results of taking the existing StyleGAN model for bedrooms produced by Nvidia as part of their StyleGAN paper and re-training them for the specified number of hours on a single GTX1080 with different data: Specifically, each model was retrained with a dataset of 100,000 images from the LSUN Scene Classification challenge. In short: Nvidia did the hard part. I just dropped some data into a directory and followed their instructions.
For more info, see How I Learned to Stop Worrying and Love Transfer Learning.
