Video compression algorithms used to stream videos are lossy, and when compression rates increase they result in strong degradation of visual quality. We show how deep neural networks can eliminate compression artefacts and restore lost details.