DCGAN is initialized with random weights, so a random code plugged into your network would produce a totally random impression. However, while you may think, the network has many parameters that we can tweak, as well as the target is to locate a setting of these parameters that makes samples produced from random codes appear to be the training info