YuE released an open source music generation model, comparable to Suno, generating professional singing and accompaniment

YuE, an open source music generation model developed by the Hong Kong University of Science and Technology and DeepSeek, was officially released, attracting widespread attention. The model adopts a dual LLaMA language model architecture, which can generate music in multiple styles and simulate human vocals, supports up to 5 minutes of complete song creation, and is comparable to closed-source tools such as Suno and Udio.

YuE uses the "Dual Track Next Token Prediction" strategy to model vocals and accompaniment separately to improve sound quality and consistency. Its "structured gradual generation" technology ensures that the main and chorus passages of the song are alternated in a coherent manner. In addition, unique music context learning techniques avoid plagiarism and enhance creative uniqueness.

YuE has performed well in multiple reviews, providing music creators and enthusiasts with a new AI creation experience. Project address: https://github.com/multimodal-art-projection/YuE