This paper presents a game in which players can customize the behavior of their characters using their own movements while playing the game. Players’ movements are recorded with a motion capture system. The player then labels the movements and uses them as input to a machine learning algorithm that generates a responsive behavior model. This interface supports a more embodied approach to character design that we call “Customizing by Doing”. We present a user study that shows that using their own movements made the users feel more engaged with the game and the design process, due in large part to a feeling of personal ownership of the movement.