DCGAN is initialized with random weights, so a random code plugged in the network would crank out a totally random image. Even so, when you might imagine, the network has an incredible number of parameters that we could tweak, along with the goal is to find a environment of these parameters which makes samples generated from random codes look like