DCGAN is initialized with random weights, so a random code plugged into your network would produce a totally random picture. Having said that, while you might imagine, the network has millions of parameters that we are able to tweak, along with the purpose is to find a environment of these parameters which makes samples created from random codes