Résumé : In this paper we introduce a complete framework to automatically generate cinematographic view of dynamic scenes in real-time. The main goal of such a system is to provide a succession of shots and sequences (of virtual dynamic scenes) that can be related to pure cinema. Our system is based on the use of an image-based control of the camera that allows different levels of visual tasks and a multi-agent system that controls those cameras and selects the type of shot that has to be performed in order to fulfill the constraints of a given cinematographic rule (idiom). This level of adaptation constitutes the major novelty of our system. Moreover, it stands for a convenient tool to describe cinematographic idioms for real-time narrative virtual environments.