StyleGAN's "truncation trick" pushes weights towards the average, resulting in better quality images. You can of course push weights away from the average too. I find this works best with the celebrity faces dataset.
You can also ask for the average of the dataset. Here are the average Flickr face and average celebrity. The "gal" on the left shows up in many StyleGAN articles. Not too surprising to see brown hair and a brown background. Dark skin must be underrepresented in the training data.