YuE2 is a music generation system that unifies symbolic and audio generation to produce complete songs from text prompts.
The tool addresses the challenge of generating high-quality music while maintaining human control over composition. It works by accepting lyrics and style descriptions, then generating an explicit melody-and-chord plan in symbolic form before rendering that plan as a full song with vocals and accompaniment. This two-stage approach makes the composition editable and interpretable at the symbolic level before audio synthesis occurs.
Developers should adopt this tool if they need music generation with white-box control over melodic and harmonic content. It suits projects requiring either original composition from text prompts or style transfer of existing songs. The system supports zero-shot cover generation, where a transcribed song can be reimagined in a new style using the same checkpoint. It also enables agentic music editing, where an agent can iteratively refine a composition through conversation about its score, arrangement, and lyrics. The tool is positioned as competitive with proprietary systems on its evaluation benchmark.
The project maintains active development with regular updates to its codebase and model releases on Hugging Face. Documentation includes detailed benchmarking methodology and architectural explanations. The team provides multiple demonstration modes showing creation, cover generation, and agentic editing workflows, with specific examples like a nine-step transformation of a song from Mandarin pop to English jazz. Community engagement occurs through an active Discord channel for users and contributors.
