DCGAN is initialized with random weights, so a random code plugged in the network would crank out a completely random image. However, while you might imagine, the network has millions of parameters that we are able to tweak, and the purpose is to find a environment of these parameters that makes samples generated from random codes look like the ins