YuE

Spec sheet

What it does

YuE2 turns lyrics and a style prompt into a complete song with vocals and accompaniment.Read 15 Oct 2026 · github.com

Product

YuE2 generates songs from lyrics and a style prompt by creating an editable melody-and-chord plan, then rendering vocals and accompaniment.Read 12 Oct 2026 · github.com

Intended users

The repository describes personal users, content creators, and musicians as permitted to use YuE2 and monetize generated outputs without fees or royalties payable to the authors.Read 12 Oct 2026 · github.com

Supported platforms

webRead 21 Sep 2026 · github.com

Free plan

YesRead 21 Sep 2026 · github.com

Integrations

Official materials document Hugging Face model loading, ComfyUI native nodes, a NOIZ-hosted demo, Maestro and the SheetSage2 transcription model.Read 15 Oct 2026 · github.com

Academic origin

The YuE paper describes a family of open foundation models based on LLaMA2 for lyrics-to-song generation of up to five minutes, with style transfer and bidirectional generation.Read 15 Oct 2026 · arxiv.org

Agent editing

The repository provides an agent skill for song generation, instrumental creation, transcription, ABC-score editing and listening comparisons.Read 15 Oct 2026 · github.com

Agent workflow

The repository includes a yue2-music agent skill for song generation, instrumental music, transcription, covers, and score editing.Read 12 Oct 2026 · github.com

Audio output

The documented pipeline produces 48 kHz stereo audio and saves FLAC plus scores, tokens, latents, settings and model identities.Read 15 Oct 2026 · github.com

Audio upload

YesRead 21 Sep 2026 · github.com

Benchmark claim

The Hugging Face model card reports YuE2 best-of-8 SongBench average 6.9632 on 192 WildSongBench prompts, above the listed Suno v5 and v6 results.Read 15 Oct 2026 · huggingface.co

Code licence

The repository code is released under the Apache License 2.0.Read 15 Oct 2026 · github.com

Commercial use

YesRead 21 Sep 2026 · github.com

Covers

YuE2 can use a transcribed song melody to create a new interpretation in a target style.Read 12 Oct 2026 · github.com

Documentation depth

The repository includes quick start, creation, cover, editing, agent-skill, benchmark and generation-guide materials with runnable Python and command-line examples.Read 15 Oct 2026 · github.com

Editable composition

The system creates an explicit melody-and-chord plan that users or agents can inspect and change before audio rendering.Read 15 Oct 2026 · github.com

Editing

Users can edit an ABC score’s harmony, melody, tempo, or form and generate a new complete recording.Read 12 Oct 2026 · github.com

Generation API

The staged Python interface exposes plan(), generate_semantic(), synthesize() and decode() operations.Read 15 Oct 2026 · github.com

Hosted demo

The NOIZ demo accepts song descriptions and lyrics, and its cover studio accepts audio uploads up to 100 MB in listed formats including MP3, WAV, FLAC, M4A, AAC, and OGG.Read 12 Oct 2026 · yue.noizai.net

Independent rating availability

Capterra lists a YuE AI product page with a displayed 0.0 rating and no visible review count in the search result.Read 15 Oct 2026 · capterra.com

Inference availability

Hugging Face states that YuE2-3B is not deployed by any Inference Provider.Read 15 Oct 2026 · huggingface.co

Licensing

The repository says code, agent skill, and documentation use Apache 2.0, while model weights use CC BY-NC 4.0 with additional creator permission; commercial use by companies requires contacting the authors about a model-weights license.Read 12 Oct 2026 · github.com

Limit: audio quality

Tom's Guide said low-VRAM versions reduce audio quality and can generate extremely slowly.Read 15 Oct 2026 · tomsguide.com

Limit: commercial use

MrKeyoor notes that the code is Apache-2.0 while the model weights are CC BY-NC 4.0, restricting commercial products that cannot accept noncommercial terms.Read 15 Oct 2026 · mrkeyoor.com

Limit: editing behavior

MrKeyoor reports that editing a score renders a new complete recording and can change singing, timbre or other waveform regions.Read 15 Oct 2026 · mrkeyoor.com

Limit: low-VRAM workflows

Elma says quantized or offloaded configurations can be slower, more fragile and more prone to out-of-memory failures.Read 15 Oct 2026 · elma.sh

Limit: production readiness

Elma advises treating 4 GB community workflows as experimental and testing exact song lengths, concurrency and output standards before production use.Read 15 Oct 2026 · elma.sh

Limit: setup complexity

MrKeyoor reported a 7,073 MB environment before model weights, 77 installed packages and failed tests or collection errors during setup.Read 15 Oct 2026 · mrkeyoor.com

Local requirements

The quick start specifies Linux, Python 3.12, and an NVIDIA GPU with BF16 support and 24 GB VRAM.Read 12 Oct 2026 · github.com

Maker identity

LinkedIn describes Multimodal Art Projection as an open-source Art x AGI research community and nonprofit with 11–50 employees.Read 15 Oct 2026 · linkedin.com

Model licence

The YuE2-3B model card identifies the weights as Creative Commons Attribution-NonCommercial 4.0.Read 15 Oct 2026 · huggingface.co

Notable news: release

The repository dates the YuE2 release for September 14, 2026, with global, text-to-audio and language milestones later that month.Read 15 Oct 2026 · github.com

Older hardware limitation

Tom's Guide reported the original release required at least 24 GB VRAM and officially recommended 80 GB for full-song creation.Read 15 Oct 2026 · tomsguide.com

Performance limitation

Tom's Guide observed that an 8 GB RTX system took roughly two to two-and-a-half hours to generate 40- to 50-second tracks.Read 15 Oct 2026 · tomsguide.com

Praise: controllability

Elma highlights the editable symbolic layer as useful for iteration, covers, reproducibility and agent workflows.Read 15 Oct 2026 · elma.sh

Praise: musicality

Tom's Guide found the generated tracks musically impressive, with accurate prompt adherence, strong vocals and convincing instrumentation.Read 15 Oct 2026 · tomsguide.com

Python API

The staged Python API exposes plan(), generate_semantic(), synthesize(), and decode() operations.Read 12 Oct 2026 · github.com

Python SDK

A YuE2Pipeline class can be loaded from Hugging Face and called with style, lyrics, score and seed parameters.Read 15 Oct 2026 · huggingface.co

Repository activity

GitHub shows a public repository with about 12,000 stars, 1,400 forks, 29 issues and 52 pull requests.Read 15 Oct 2026 · github.com

Self-hosting

The project is distributed for local installation with pip and command-line examples rather than as a required cloud service.Read 15 Oct 2026 · github.com

System requirements

The current quick start specifies Linux, Python 3.12, an NVIDIA GPU with BF16 support and 24 GB VRAM.Read 15 Oct 2026 · github.com

Upload handling

The NOIZ cover page states that an uploaded song is used only to create that cover.Read 12 Oct 2026 · yue.noizai.net

Vocal generation

YesRead 21 Sep 2026 · github.com

Zero-shot covers

YuE2 can use a transcribed source melody and new lyrics or style instructions to create a changed arrangement.Read 15 Oct 2026 · huggingface.co

Plans and prices

Every YuE plan and price →

Where Specifiction places it

• AI music generators#20 of 1238.6
• AI Song Generators#9 of 318.1