DCGAN is initialized with random weights, so a random code plugged in the network would create a completely random graphic. Having said that, when you might imagine, the network has countless parameters that we can tweak, as well as target is to find a setting of such parameters that makes samples generated from random codes appear like the instruc