3D character generation using PCGML

Abstract

In video games, both the characters and the world play an important role in creating a sense of immersion to the player. Although each character can be modelled by hand to make them feel more life-like, the process becomes inherently complex when populating a game map with more than 100000+ characters. In cases like this, developers must look towards incorporating new tools to automate and accelerate their creation pipeline. We present a new method to procedurally generate the Non- Playable Characters (NPCs) in video games using a modified Style-based generative adversarial network (StyleGAN) which is a type of neural network. PCGML acronym for Procedural Content Generation using Machine learning is the most cost- effective method for game content generation, it is employed to reduce production effort and to save storage space. Our approach adapts the use of PCGML with styleGAN to generate NPCs that are unique in both appearance and behaviour. The properties or traits influence the generation of characters making the game environment diverse and interesting for the players.