Black Forest Labs' first video model generates audio and video from the same pass, and it takes direction like nothing else we've tried.