multimodal-art-projection/YuE

Generates full-length songs from lyrics using a two-stage autoregressive architecture that produces semantic tokens followed by audio codec tokens, supporting style transfer via in-context learning and voice cloning. Multi-lingual models (English, Chinese, Japanese, Korean) are available on Hugging Face with configurable session counts for verse/chorus generation. Integrates with community tools like Gradio UIs and quantized inference frameworks for resource-constrained deployment (8GB+ VRAM).

6,083 stars. No commits in the last 6 months.