This open-source song generation model takes lyrics and a style description as input to produce a complete song with vocals and accompaniment. It first generates melody and chord scores, then creates audio from the score. It supports editing the score to regenerate the song, as well as creating covers based on reference recordings. The companion Skill also supports generating purely instrumental music.