YuE2-3B

YuE2-3B

Turns lyrics and a style prompt into a full song with vocals and accompaniment, and lets you edit the melody and chords through a score. An open 3.6B music generation model in English and Chinese for non-commercial use.

Made by m-a-p on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: YuE2-3B is an open music generation model from m-a-p. You give it lyrics and a style prompt, and it makes a complete song with vocals and accompaniment. It also plans the song as an editable score, so you can change the melody and chords before it renders the audio. The card says it runs locally and supports English and Chinese.

Who it suits: Songwriters, hobby musicians and researchers who want to draft songs from lyrics and edit the musical plan by hand.

Industry | Who uses it | How they use it | Example
Music making | Songwriters | Turn lyrics and a style note into a demo | A first draft of a song from a verse
Music production | Producers | Edit the melody and chords in the score | Changing a chord line and rendering again
Research | Music AI researchers | Study planning and synthesis as separate steps | Comparing outputs on a song benchmark
Personal projects | Hobbyists | Make background songs for their own videos | A short song to go under a family video

If you need music for paid commercial work, you can skip it, because the license is non-commercial.

Where to use it:

• Python on Windows, macOS or Linux, after you download the files
• A computer with a 24 GB GPU, which the card names for local runs
• Agent setups that revise score, style and lyrics, as the card describes

Use case | Who | What they get back
Song from lyrics | Songwriters | A complete song with vocals and accompaniment
Score editing | Producers | A new version after changes to melody or chords
Cover of a song | Hobbyists | A new rendition of a song you have rights to use

How to set it up:

• Python and pip
• About 7.3 GB of disk space in BF16, or about 1.8 GB at 4-bit (an estimate)
• A 24 GB GPU to render 48 kHz stereo songs without quantization

1. Install the Hugging Face hub tools.
2. Download the model files.
3. Follow the model card to load the model and render a song.

pip install -U huggingface_hub
hf download m-a-p/YuE2-3B

How to use it: Write the lyrics and a short style prompt, then generate. You can compose with melody and chords, with melody alone, or generate directly, and you can bring your own ABC score. The card also lists text guidance and separate planning and synthesis steps for developers.

Example: You provide two verses of lyrics and a style prompt for a calm acoustic song. The model plans a score, renders vocals with accompaniment and returns a stereo song you can then revise by editing the score.

For newcomers and experts: If you are new to this, start with short lyrics and one clear style prompt. Listen to the result, then ask for a revision.

YuE2 plans a song symbolically first, then synthesizes audio from the plan, which is why editing the score is possible. The card reports a SongBench average of 6.9632 for its best-of-eight setting on 192 WildSongBench prompts. These results are the maker's own and depend on picking the best of eight tries.

You write lyrics and a style prompt
The model plans the song as a score
You edit melody, chords or lyrics if needed
The model renders vocals and music

Good to know:

• Hardware: the card names a 24 GB GPU for local runs without quantization
• Benchmarks: the figures are the maker's own, and the top score is best-of-eight

Title: SongBench average on WildSongBench prompts, maker's own comparison
YuE2 best-of-8: 6.9632
Suno v5: 6.8721
Suno v6: 6.5562
Suno v6 Wild: 6.4195

• License: cc-by-nc-4.0, for non-commercial use
• Cost to run: no API key or paid service is named, only a capable GPU
• Data it reaches: downloading fetches model files, and your lyrics and any score you give stay in your own run
• Not for: commercial use under this license, or covers of songs and voices you do not have the right to use

On Hugging Face, 13 November 2026

• Task: text to audio
• Parameters: 3.6B
• Weights: about 7.3 GB in BF16; about 1.8 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 22,317
• Likes: 1,253
• License: cc-by-nc-4.0
• Languages: zh, en
• Quantized versions: 25
• Fine-tunes: 12
• Tags: safetensors, yue2, music-generation, symbolic-planning, agentic-editing, text-to-audio

Links

• Model: https://huggingface.co/m-a-p/YuE2-3B
• Files: https://huggingface.co/m-a-p/YuE2-3B/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download m-a-p/YuE2-3B

Then follow the usage in its model card. Nothing in this kit runs by itself.