DCGAN is initialized with random weights, so a random code plugged into the network would generate a completely random picture. However, as you might imagine, the network has numerous parameters that we will tweak, along with the purpose is to locate a setting of these parameters that makes samples generated from random codes look like the training