DCGAN is initialized with random weights, so a random code plugged into your network would crank out a totally random image. Having said that, while you might imagine, the network has millions of parameters that we could tweak, as well as the target is to find a setting of such parameters which makes samples generated from random codes appear like