The generic central camera model can describe any imaging system with a single effective viewpoint. We are interested in the self-calibration of this model from only two rotations about non co-linear axes. Concisely, we show that the theoretical solution given in [1] performs correctly with real flow data, and that a simple iterative process can be used to improve the camera motion estimation. A method for the computation of smooth dense generic flows from rotational images is also presented and used in our experiments to compute “real” flows from both simulated and real images.