KEEP is a video face super-resolution framework based on the principles of Kalman filtering, aimed at maintaining stable face priors over time through feature propagation. It effectively captures consistent facial details across video frames by integrating information from previously recovered frames to guide and adjust the recovery process of the current frame.