Abstract

Crowds of non-player characters are increasingly common in contemporary video games. It is often the case that individual models are re-used, lowering visual variety in the crowd and potentially affecting realism and believability. This paper explores a number of approaches to increase visual diversity in large game crowds, and discusses a procedural solution for generating diverse non-player character models. This is evaluated using mixed methods, including a “clone spotting” activity and measurement of impact on computational overheads, in order to present a multi-faceted and adjustable solution to increase believability and variety in video game crowds.