Research on Automatic Game Concept Art Character Generation Based on Generative Adversarial Networks

Creating character game concept art is a longstanding practice that costs a lot of time and skill. An efficient automatic character generation technique is proposed in this study to greatly enhance the productivity of game art production. The proposed method uses an altered version of StyleGAN2 architecture, training it on a dataset of over 15,000 game character concept arts by numerous artists originally. Experimental results confirm that the new method improves upon other techniques by creating highly diverse character designs at a top level of quality. Characters generated from artificial intelligence are highly believable to professional game artists, and 78 percent of the characters could be used in real games after minor modifications. Using the system allows for tremendous reductions in the traditional character creation time processes.