Generate new data similar to a training set across text, image, audio, and video.
No models in this category are strong on that dimension yet.