DCGAN is initialized with random weights, so a random code plugged into the network would create a very random image. Having said that, as you might imagine, the network has a lot of parameters that we will tweak, along with the objective is to find a placing of these parameters which makes samples generated from random codes seem like the coaching