YuE2-3B Open-Weights Music Generation Model Released
The Multimodal Art Projection team has released YuE2-3B, a 3-billion parameter open-weights model designed for full-song music generation from lyrics. The release is accompanied by an interactive web demo showcasing its capabilities. High-quality, open-weights audio generation models remain scarce compared to text and image models. YuE2-3B gives developers and researchers a powerful local option for full-length AI music generation without relying on proprietary cloud services. YuE2-3B builds upon the YuE architecture, which leverages Large Language Model frameworks to handle long-form lyrics-to-song synthesis. The model's 3-billion parameter size aims to balance generation quality and track coherence with reasonable computational requirements for local execution.
## BACKGROUND
Developed by Multimodal Art Projection (m-a-p), YuE is an open foundation model family created specifically for long-form music generation. Unlike brief audio snippet generators, YuE adapts LLM architectures like LLaMA to generate structured, full-length tracks complete with vocals and accompaniment from text lyrics.