YuE is an open-source foundation model family for music generation, capable of turning lyrics and style prompts into complete songs with vocals and accompaniment. Developed by the Multimodal Art Projection group, the project has earned more than 12,400 stars on GitHub, is released under the Apache-2.0 license, and was actively developed in 2026, with the YuE2 generation adding symbolic planning and agentic music editing.
Most open music models generate short instrumental loops; YuE targets full songs. It models lyrics and music jointly, following a track structure so verses, choruses and bridges hang together, and it can sing in multiple languages and genres. YuE2 extends the original with symbolic planning for more coherent compositions, zero-shot cover generation that re-styles existing audio, and editing workflows driven by language instructions. The repository ships model code, inference scripts and links to weights, making it one of the most capable open alternatives to commercial song generators.
The project comes from a research group focused on open multimodal creation, and its releases are accompanied by demos and technical reports that document training data and architecture choices — a transparency that commercial music generators do not offer. For the open-source community, YuE serves both as a usable instrument and as a foundation others fine-tune for specific languages, genres and vocal styles.
YuE is free and open source. Download the code and model weights from GitHub and Hugging Face and generate music on your own hardware at no cost.
YuE is for independent musicians, AI-music hobbyists and researchers who want a powerful open song-generation model they can run, fine-tune and study without depending on a closed commercial service.