Visual captions: Using large language models to augment video conferences with dynamic visuals