YuE2 is an open music generation model that turns lyrics and a style prompt into a complete song with vocals and accompaniment, in 48 kHz stereo. Its twist is that it first writes a melody-and-chord plan as a readable score, then renders that plan as audio, so you or an AI agent can change the harmony, melody, tempo or form and render the edited score again.
The same checkpoint also does covers: transcribe a recording with the companion SheetSage2 model, keep its melody, and reimagine it in a new style or with new lyrics. There are native ComfyUI nodes, an agent skill for generating and editing music by conversation, and a free hosted demo for trying it without installing anything. In the team's own benchmark, it scores alongside commercial systems such as Suno.
The code is Apache-2.0, with about 12,700 stars, at yue2-v0.1.6 when scanned; the original YuE is preserved on a separate branch. Running it locally needs Linux, Python 3.12 and an NVIDIA GPU with 24 GB of memory and BF16 support. The model weights carry their own non-commercial licence, covered in the FAQ.
- Repository: github.com/multimodal-art-projection/YuE
- Licence: Apache-2.0 (Apache License 2.0)
- Language: Python. Stars: 12.7K. Forks: 1,456. Last push: Oct 6, 2026.
- Scan: safe, Oct 26, 2026, commit 1647252
Who it is for
Musicians, producers and video creators who want song drafts they can edit, and developers or researchers who want an open music model with a controllable score.
Getting started
1. Clone the repository
git clone https://github.com/multimodal-art-projection/YuE.git && cd YuE2. Create an environment and install
python3.12 -m venv .venv && source .venv/bin/activate && python -m pip install .3. Generate the example song
python examples/generate.py --output outputs/first-songThe model files download from Hugging Face on first use. The song is written to outputs/first-song/audio.flac, alongside the score and generation settings.
Safety scan
We cloned multimodal-art-projection/YuE at commit 1647252 on Oct 26, 2026 and ran the checks described on the GitHub Tools page: credential patterns, decode-and-execute code, install-time scripts, committed binaries, risky CI workflows, every host the code talks to, known vulnerabilities in pinned dependencies, and project hygiene. A person read every hit. This is what we found.
- No secrets, no suspicious patterns, no install hooks and no committed binaries across 116 files and about 13,500 lines of Python.
- No lockfile, so no dependency advisories are reported either way; pip installs current releases of the packages the project lists.
- The only network activity is downloading the model weights from Hugging Face on first use. Generation runs on your own GPU.
- 1 workflow using only GitHub's own actions. No security policy, Dependabot or CodeQL, which is common for a research release.
What the scanner counted
| Check | Result |
|---|---|
| Secrets | None found. |
| Suspicious code | None found. |
| Install-time code | None: nothing runs at install beyond the package manager itself. |
| Committed binaries | None. |
| CI workflows | 1 workflow. None use pull_request_target. No third-party actions. |
| Network hosts | 1 distinct host referenced from source; most often test.invalid. No URLs to bare IP addresses. |
| Known vulnerabilities | No lockfile to check: dependencies are declared as ranges, so what gets installed is whatever is current on the day. |
| Project hygiene | Has licence file. Missing security policy, automated dependency updates, CodeQL, contributing guide. |
| OpenSSF Scorecard | Not scored: the project is not in Scorecard's weekly index. |
By the numbers
| Stars | 12.7K |
|---|---|
| Forks | 1,456 |
| Contributors | 13 |
| Commits | 190 |
| Open issues | 37 |
| Open pull requests | 54 |
| Releases | 2 |
| Latest release | yue2-v0.1.6 |
| Licence | Apache-2.0 |
| Main language | Python |
| Project age | 1 year |
| Last push | Oct 6, 2026 |
| Tracked files | 116 |
| Lines of code | 13.5K |
| Checkout size | 7 MB |
Lines by language: Python 9,237, Markdown 2,260, JSON 1,924, YAML 52, TOML 42.
Questions
Is YuE2 free?
The code is Apache-2.0. The model weights are CC BY-NC 4.0 with an added creator permission: personal users, content creators and musicians can use YuE2 and monetise the songs without fees or royalties, academic use is free for non-commercial purposes, and companies need a commercial licence for the weights.
What hardware does YuE2 need?
Linux with Python 3.12 and an NVIDIA GPU that supports BF16 and has 24 GB of memory. If you do not have one, the NOIZ-hosted demo linked from the README runs it in the browser.
Can I edit a song after generating it?
Yes. Export the plan, change the ABC score by hand or with an agent, and pass the edited file back with --abc-file. Editing renders a new complete recording rather than patching the old audio.
This post is part of GitHub Tools, where every repository is cloned and scanned before it is written up. The scan is a snapshot of one commit on one day; the repository has moved on since, so check it before you install.