DCGAN is initialized with random weights, so a random code plugged into your network would produce a totally random image. However, as you might imagine, the network has numerous parameters that we can easily tweak, along with the target is to locate a placing of those parameters that makes samples generated from random codes look like the coaching