DCGAN is initialized with random weights, so a random code plugged into the network would deliver a completely random image. Even so, as you might imagine, the network has millions of parameters that we can easily tweak, plus the target is to locate a placing of such parameters that makes samples created from random codes look like the teaching dat