The computer graphics field is filled with applications that require the setting of tricky parameters, which are often complex and unintuitive for non-experts.  We present an active learning method and application for setting parameters for a procedural fluid animation system.  In this system, we learn a model by showing the user a gallery of instances of different parametrized animations and asking for preference feedback.  Our method employs Bayesian optimization to incorporate prior knowledge based on previous sessions and/or expert knowledge to assist users in finding good parameter settings in as few steps as possible.