Yining Wang

Text-to-music models may appear controllable, but a closer look reveals that much of their output is simply a reflection of their training data rather than genuine instruction-following.