I shudder to think how current LLMs would go with this. I guess we can currently do this easily for still images and audio. Kind of reminds me of Translation Party.