So YuE2 (YuE2-3B) from the M-A-P team just dropped. It’s an open-weights model that generates full songs with vocals and arrangement right on your machine.
Key points:
— Feed it lyrics + a style description → get a finished 48 kHz track.
— It first writes an editable score (ABC notation), then renders the audio. You can tweak melody and chords.
— On their benchmark it beats Suno v5 and v6 in best-of-8 mode.
— On an RTX 4090 a 3–4 minute song takes about 70–90 seconds. People are also running it on 16 GB cards.
Downsides: you need a decent GPU (24 GB preferred), setup isn’t one-click, and the weights are under CC BY-NC (some commercial restrictions for companies).
Overall it’s currently the strongest open local music generator with vocals. If you’ve got a solid GPU — definitely worth trying. Suno isn’t the only option anymore.
Key points:
— Feed it lyrics + a style description → get a finished 48 kHz track.
— It first writes an editable score (ABC notation), then renders the audio. You can tweak melody and chords.
— On their benchmark it beats Suno v5 and v6 in best-of-8 mode.
— On an RTX 4090 a 3–4 minute song takes about 70–90 seconds. People are also running it on 16 GB cards.
Downsides: you need a decent GPU (24 GB preferred), setup isn’t one-click, and the weights are under CC BY-NC (some commercial restrictions for companies).
Overall it’s currently the strongest open local music generator with vocals. If you’ve got a solid GPU — definitely worth trying. Suno isn’t the only option anymore.
