Techniques that enable users (e.g., animators) to accurately animate models in a scene without having to load all the models involved in the scene concurrently in computer memory. For a particular model that a user wishes to animate, only a minimal set of models involved in the scene that are needed for the particular model to evaluate correctly are determined and loaded into computer memory. Additionally, if a particular model is to be unloaded from computer memory, then all models that depend, either directly or indirectly, on the particular model and that are loaded in memory are also unloaded from memory in order to avoid incorrect animation.