yue2 huggingface

YuE2-3B is an open music-generation model published by m-a-p on Hugging Face. It generates complete songs with vocals and accompaniment from lyrics and style prompts, and supports score-guided creation and editable music workflows.

Key Features

• Generate vocal songs and accompaniment from lyrics and a style prompt.

• Use editable ABC scores to guide musical structure and revise generated sections.

• Support cover workflows and agentic editing for more controlled song creation.

• Provide separate planning and synthesis APIs, with classifier-free guidance options.

• Generate 48 kHz stereo songs using a 4B-parameter BF16 model.

How It Works

Users can load the model from Hugging Face, provide lyrics and a style prompt, and run the planning and synthesis stages locally. The model card documents Linux and Python 3.10+ requirements, along with an NVIDIA GPU offering at least 24 GB of memory and BF16 support.

Use Cases

YuE2-3B is suitable for researchers and developers exploring open music generation, score-guided composition, vocal-song prototyping, cover generation, and controllable music editing. The model card lists Chinese and English support and provides benchmark results for SongBench comparison.

Pricing and Availability

The model is openly available on Hugging Face under CC BY-NC 4.0. It is not deployed through the Hugging Face Inference Provider, so local GPU inference is required for the documented workflow.

Screenshots