Can it sing? Is it an end to end multimodal model?