Ember, Toward Salience-Based Cinematic Generation

Automatic cinematic generation for virtual environments and games has shown to be capable of generating general purpose cinematics. There is initial work that approaches cinematic generation from the perspective of narrative instead of low level camera manipulation. In our work, we further extend this idea to take into consideration a model of user memory. Models of reader comprehension and memory have been developed to attempt to explain how people comprehend narratives. We propose using these models of narrative comprehension and memory to augment a system for cinematic generation so that it can produce a cinematic that can communicate character deliberation to the viewer by maintaining the salience of specific events.