DCGAN is initialized with random weights, so a random code plugged into the network would make a totally random impression. However, as you may think, the network has numerous parameters that we could tweak, as well as the intention is to find a setting of such parameters which makes samples generated from random codes look like the instruction k