DCGAN is initialized with random weights, so a random code plugged into your network would deliver a totally random picture. Nonetheless, as you may think, the network has a lot of parameters that we could tweak, along with the aim is to find a location of those parameters that makes samples created from random codes seem like the coaching details.