The New Open-Source AI That Writes Songs
Readers help support Windows Report. We may get a commission if you buy through our links.
Read our disclosure page to find out how can you help Windows Report sustain the editorial team. Read more
YuE, an open-source AI model that can turn written lyrics into complete five-minute songs, is officially here. A research team from the Hong Kong University of Science and Technology and M-A-P is behind this model.
The project, described in a paper on arXiv, arrives at a moment when AI-generated music is facing mounting legal questions.
YuE is available on GitHub under the permissive Apache 2.0 license, allowing both developers and musicians to explore the technology without restrictive terms. The researchers say the goal is to encourage transparent innovation in AI music while keeping it accessible to creators.
Technically, YuE builds on the LLaMA 2 architecture and was trained on trillions of tokens. Its standout feature is the ability to keep songs coherent over several minutes. The model separates vocals and accompaniment into parallel streams, avoiding the muddiness that can happen in dense arrangements.
It also uses a method called structural progressive conditioning. This breaks a track into logical sections, like verses and choruses, helping it preserve melody and rhythm through each part.
Another strength is its in-context learning ability, which lets users feed in a short audio clip to guide genre, vocal style, or even replicate a specific voice.
The official demo showcases impressive flexibility, from smooth jazz scatting to aggressive metal growls, all without direct training for those styles. With its open release, YuE could become a notable tool for musicians looking to experiment with AI, without crossing into legally murky territory.
