DCGAN is initialized with random weights, so a random code plugged into your network would produce a totally random picture. However, while you may think, the network has numerous parameters that we could tweak, and also the purpose is to locate a setting of these parameters that makes samples produced from random codes seem like the education data