Emotions are a key part of creative endeavours, and a core problem for computational models of creativity. In this paper we discuss an affective computing architecture for the dynamic modification of music with a view to predictablyaffectinginducedmusicalemotions. Extending previous work on the modification of perceived emotions in music, our system architecture aims to provide reliable control of both perceived and induced musical emotions: its emotionality. A rule-based system is used to modify a subset of musical features at two processing levels, namely score and performance. The interactive model leverages sensed listener affect by adapting the emotionality of the music modifications in real-time to assist the listener in reaching a desired emotional state.