We propose methods to super-resolve compressed video sequences that may consist of frames with missing blocks of pixels in addition to compression artifacts. Different from traditional resolution enhancement algorithms, our methods include two key components that are crucial to handle compressed video sequences. The first component is dynamic masking that dynamically computes image masks used to reject outliers. The second component is flow-based image repairing that reconstructs missing blocks or a whole frame by exploring both temporal and spatial information. We demonstrate the proposed methods with real MPEG video sequences.