I hear ya... Don't you think you're stretching the meaning of the words "composite" and "several" a little far though?
Asking as an ML layman, isn't it mixing something like 6 maybe 10 features of different people?
Maybe, maybe not. GANs are not necessarily interpretable so it's not easy to know the answer to your question.
With tens of thousands of training images, and hundreds of dimensions in the latent space, I don't think I would assume this is true. You may be thinking one feature is "eyes", but the eyes may instead be built out of 10 sub-features. Those sub-features may be inextricably linked to other, identifiable, macro-features.