YuE
multimodal-art-projection/YuEReviewed by AI
Checked by RepoPicks Radar on
What is YuE?
YuE2 writes songs. You give it lyrics and a style, and it first makes a melody-and-chord score you can read and change, then sings and plays it as a 48 kHz stereo track. It can also take an existing recording, transcribe the melody and perform it in another style, and an included agent skill lets an AI assistant drive the whole process. Running it yourself means Linux, Python 3.12 and an NVIDIA card with 24 GB of video memory; the models download from Hugging Face the first time. A free hosted demo lets you try it in the browser.
YuE is free and open source (Apache-2.0) and can replace Suno, Udio, and Epidemic Sound.
How easy is it to install YuE?
A · download and open · Web
- Windows E
- macOS E
- Linux D
- Web A
The one-click way is the free hosted demo the README links ("Try YuE2 online for free ... No installation required"), which runs in the browser. Doing it on your own machine is much harder: the quick start is "Linux · Python 3.12 · NVIDIA GPU with BF16 support and 24 GB VRAM" with a clone, a virtual environment and command-line runs, and Windows, Apple Silicon, AMD and Intel are only described as "experimental" paths in docs/hardware.md.
YuE on GitHub
- Stars
- 13,370on Nov 10, 2026
- Downloads
- 9,228530 this week
- Growth
- +285stars in 6 days
Stars over time
Facts
- Runs on
- Linux, Windows, macOS, Web
- You need to install
- Nothing extra
- To build from source
- Python, NVIDIA graphics card
- Disk space
- 950 MB–4.3 GB · estimate
- Accounts or keys
- None
- Cost
- Free · Code and the agent skill are Apache 2.0. The model weights are CC BY-NC 4.0 with an extra permission: people, creators and musicians may use them and sell what they make, academic use is non-commercial, and companies have to ask the authors for a commercial license.
- License
- Apache-2.0 (Permissive)
- Can replace
- Suno, Udio, Epidemic Sound
- Last change on GitHub
- Nov 1, 2026
YuE risk signals (4)
- Attentionmodel weights non commercial
The README license table says code is Apache 2.0 but "**Model weights:** [CC BY-NC 4.0 with additional creator permission]", and "Commercial use by companies: Contact us to discuss a commercial license for the model weights".
AI review
- Attentioncovers of existing recordings
The "Cover a song" section starts from someone else's recording ("Transcribe a source recording with SheetSage2 ... provide new lyrics or a target style"), and the topics list "voice-cloning" and "style-transfers"; the rights to the original song and voice are yours to clear.
AI review
- Infoheavy hardware and downloads
Quick start: "NVIDIA GPU with BF16 support and 24 GB VRAM" and "Model files download from Hugging Face on first use", so the real disk and memory use come from downloads, not from this repository.
AI review
- Infoself reported benchmarks
The benchmark table comparing YuE2 with Suno, Mureka and MiniMax is the authors' own evaluation on their own WildSongBench set, and the README itself notes "the small gap between the highest means does not establish statistical significance".
AI review
Evidence is shown in its original language.
What Radar could not check about YuE
- Whether songs really come out at the quality shown in the demos, and how long each one takes
- How much disk space the YuE2-3B, VAE and SheetSage2 downloads take together
- Whether the experimental Windows, Apple Silicon, AMD and Intel paths work at all
- Whether the free hosted demo stays free, and what it keeps of the lyrics and audio you send
- What music the model was trained on, which matters for who owns the result