Lyrics-first, not prompt-first
Many music generators start from a one-line idea. YuE2 starts from the song words. That makes it well suited to songwriters who already have lyrics and want to hear them performed in a particular style, and less suited to "make me a song about summer" with no words.
The two routes
- Text-to-music (YuE2-3B): write the style brief and structured lyrics; the model arranges and sings them.
- Music-to-music (YuE2 3B): add one reference recording you own; its phrasing, pacing or feel guides a fresh rendition. It is guidance, not a request to copy the source.
What you need to provide
- A style description: genre, mood, tempo, instruments, vocal character, language and production texture.
- Lyrics with section tags: such as [Verse], [Chorus] and [Bridge], with intentional line breaks.
- Optionally, one reference audio file for music-to-music, using a recording you have permission to upload.
Where YuE2 runs here
This independent studio connects to the YuE2-3B text-to-music and music-to-music endpoints through WaveSpeed inference. Credit quotes are derived from WaveSpeed list prices of $0.10 and $0.15 per generation respectively, and finished tracks stay in your private library.