DCGAN is initialized with random weights, so a random code plugged into your network would crank out a completely random picture. On the other hand, as you might imagine, the network has millions of parameters that we will tweak, plus the goal is to find a location of these parameters that makes samples generated from random codes appear like the c