Film restoration involves locating the position of artifacts and replacing the "missing" portion of the film (obscured by the artifact) with pixels that had been lost. Computer vision research has recently developed many techniques for constraining and predicting parts of a scene based upon the assumption of rigid motion. In this paper, we show how the constraints can help identify artifacts as well as how the prediction can be used to replace the artifact with natural looking portions of the scene. These techniques can be superior, when the rigid motion assumption is valid, to other techniques for film restoration.