While it is not the main motivation behind our system, we can also generate full sentences (e.g. to give a face to virtual assistants). We also show subsequences from which we extracted expression parameters. Notice how different the subsequences are, yet we still produce a coherent video.