DCGAN is initialized with random weights, so a random code plugged in the network would create a completely random image. Having said that, while you might imagine, the network has countless parameters that we can tweak, as well as goal is to locate a placing of those parameters that makes samples created from random codes appear like the education