YuE
A song model that writes an editable melody-and-chord plan from lyrics and style, then renders vocals and accompaniment.
YuE2 plans melody and chords in ABC, then renders a full song; covers and agent-led edits supported.
YuE is an open music-generation project whose current README describes YuE2, which turns lyrics and a style prompt into a complete song. It suits researchers and technical musicians with a high-end NVIDIA GPU, and its model weights are not free for company commercial use.
What it is
YuE2 first writes a melody-and-chord plan as an ABC score, then generates semantic tokens and acoustic latents and decodes stereo 48 kHz audio. You can edit the score before rendering, make zero-shot covers by transcribing a recording with SheetSage2, or drive it through an included agent skill.
Who it's for
It suits musicians and researchers who want to inspect and edit a composition plan. It does not suit non-technical users, anyone without a 24 GB GPU, or companies that want to use the weights commercially without contacting the authors.
Key features
Editable score plan
Generates a melody-and-chord ABC plan you can read and change before rendering.
Zero-shot covers
Transcribe a source recording with SheetSage2 and re-render it in a new style.
Staged Python API
plan(), generate_semantic(), synthesize() and decode() let you reuse an exact plan.
Agent skill
The yue2-music skill guides an agent through songs, instrumentals, covers and score edits.
48 kHz stereo output
Produces audio without quantization when run on a suitable GPU.
ComfyUI nodes
Native ComfyUI nodes and an official text-to-music workflow are linked.
Up and running in 3 steps
Our honest verdict
What's great
- The plan is an inspectable ABC score, so you can change harmony, tempo or form before rendering.
- Personal creators and musicians may monetize generated outputs with no fees or royalties, per the README.
- A free online demo hosted by NOIZ is linked, so you can hear it before installing.
- The README publishes benchmark tables comparing YuE2 against other systems on WildSongBench.
Where it falls short
- The quick start asks for Linux, Python 3.12 and an NVIDIA GPU with BF16 support and 24 GB VRAM; other platforms are described as experimental.
- Model weights are CC BY-NC 4.0 with an additional creator permission, so companies must contact the authors for a commercial licence.
- Editing generates a new full recording; it does not preserve the original waveform outside the edit.
- The README says benchmark gaps between top systems do not establish statistical significance.
Pricing & license
The code is free under Apache 2.0, but the model weights are CC BY-NC 4.0 with an additional creator permission, and companies must contact the authors for a commercial licence. The README links a free online demo (https://yue.noizai.net/); no paid plan is shown.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I use YuE2 songs commercially?
Per the README, personal users, creators and musicians can monetize outputs with no fees to the authors, while companies need to contact them about a commercial licence for the weights.
Can I try it without a GPU?
The README links a free hosted demo by NOIZ at yue.noizai.net. Local runs need a 24 GB NVIDIA GPU for the supported path.
What happened to the original YuE?
The README says the original code, docs and licence are preserved on the YuE-v1 branch.
