The developer created a local music production tool based on the YuE2 AI model. The system allows users to generate tracks by defining genres, rhythms, and instruments, and includes features for structural arrangement, such as crescendos and drops.
The studio environment provides track visualization, including volume waves and spectrum analysis. It uses the demucs program to separate audio into four tracks: vocals, drums, bass, and a combined track for guitars and keyboards. Users can also perform overdubs and apply mastering effects like peak limiting and gain adjustment.
The tool includes an experimental voice-swapping feature using Seed-VC, though it currently faces limitations with singing voices and audio quality degradation. To simplify the workflow, the developer integrated an MCP server, enabling AI agents to manage generation, editing, and effects application.
The software requires a PC with an NVIDIA GPU featuring at least 16 GB of VRAM and approximately 20 GB of disk space for the worker and model weights. Optional components like Whisper for text recognition and Seed-VC for voice experiments require additional storage.
