Machine Learning-Based Automated Cinematography for Games
Phil Carlisle · 2024
We often want our games to be more “cinematic”, and the tools we have access to in games are becoming more functionally capable of producing cinematic scenes. However, the creation of cinematic cut-scene content is still largely a hand-authored labour-intensive aspect of development. That, coupled with the relative lack of definition for what “cinematic” means, places a significant burden on smaller developers. Our approach outlined in this chapter relies on an ensemble of machine learning (ML) models to develop procedural content generation tools for Unreal Engine for use in the sequencer interface. We encourage the broad adoption of ML-enabled toolsets and identify areas of production where such an approach might be beneficial.