We present a novel method to create perpetual animations from a small set of given keyframes. Existing approaches either are limited to re-sequencing large amounts of existing image/video data, or to interpolating vector based drawings. Our approach benefits from several ideas and techniques from video textures, computer-assisted animation and motion graphs. It combines the re-sequencing of existing material with the automatic generation of new data. Furthermore, the animator can interfere with the animation process at each arbitrary moment. First, a given set of keyframes is used to automatically generate a set of in-betweens. The amount of in-betweens required, depends on a distance metric preventing possible visual discontinuities. Next, an optimised cost graph is derived from the generated frames, indicating for all keyframes how many steps are required to travel from one keyframe to another. Finally, by rearranging the generated sets of in-betweens according to the cost graph, n...