HKGAI releases YuE2 open-weight music generation model

The Hong Kong Generative AI Research Centre (HKGAI) has released YuE2, a new generation of open-weight music generation models. Unlike many commercial services that generate audio directly from prompts, YuE2 utilizes a two-stage process: it first generates symbolic musical scores (ABC format) to establish melody, chords, and structure, and then performs acoustic rendering to produce high-fidelity 48kHz stereo audio.

This architectural distinction allows for granular control, enabling creators to view and edit the underlying musical structure before final rendering. Users can perform iterative edits, such as changing harmonies or rearranging compositions, without regenerating the entire track from scratch. The model supports six languages—Chinese, English, Japanese, Korean, Russian, and Spanish—and over 60 musical genres.

In WildSongBench evaluations, YuE2 demonstrated performance levels comparable to leading closed-source commercial models like Suno v6. Because the model weights are open, users with sufficient local hardware, such as NVIDIA GPUs, can run the model offline via tools like ComfyUI. This capability offers advantages in privacy and copyright management by allowing local processing of sensitive musical materials.