YuE2-3B: Open Lyrics-to-Song Model With Editable Scores
ComfyUI Wiki
YuE2-3B is a 3B open music generation model that turns lyrics and a style prompt into complete songs, with an editable melody and chord score layer.
Y
YuE2-3B
Music GenerationSymbolic PlanningSong CoversAgentic Editing3B open lyrics-to-song model by the m-a-p team. Composes complete songs with vocals and accompaniment from lyrics and a style prompt, writes an editable melody and chord plan (ABC notation) before rendering, and supports agent-driven score revisions. Outscores Suno v5 on WildSongBench.
| Developer | Multimodal Art Projection (m-a-p) |
| Release Date | 2026-09-09 |
| Architecture | AR–NAR Mixture-of-Transformers (planning + flow matching) |
| License | CC-BY-NC-4.0 |
| Parameters | 3B |
| Output | 48 kHz stereo audio |
| VRAM | 24 GB GPU (BF16, no quantization) |
Capabilities
- Text-to-song: lyrics plus a style prompt become a complete song with vocals and accompaniment
- Editable score: melody and chords are planned in ABC notation and can be revised before rendering, or supplied directly
- Covers: existing songs reimagined in a new style
- Agentic editing: musical feedback is converted into score, style, and lyric revisions across render iterations
Related
Guides and workflows related to this model series.
No articles found.
Comments
Sign in with GitHub to join the discussion.