This paper presents the research steps that have been necessaries for creating a mixed reality prototype called PUPPET. The prototype provides a 3D virtual presenter that is immersed in a real TV scenario and can be driven by an actor in real time. In this way it can interact with real presenters and/or public. The key modules of this prototype improve the state-of-the-art about these systems in four different aspects: real time management of high-realistic 3D characters, animations generated automatically from actor’s speech, less equipment needs, and flexibility in the real/virtual integration. The paper describes the architecture and main modules of the prototype