Are these algorithmically generated, say by characterizing the faces in some way and then finding a face that fits the characteristic of both the input faces?

Or did some human do this, in which case I'm having trouble seeing why it is here.