HeartMuLa/heartlib

HeartMuLa Official Repo: The Most Powerful Open-Source Music Generation Model of 2026

What it solves

HeartMuLa provides a suite of open-source tools for generating and processing music. It addresses the need for high-quality, controllable music creation by allowing users to generate songs from text-based lyrics and descriptive tags, while also offering tools for audio reconstruction and lyric transcription.

How it works

The project uses a music language model to interpret your lyrics and style tags to create audio. It includes a specialized codec (HeartCodec) to convert these generated patterns into high-fidelity 48 kHz stereo sound. Users can run the model locally by providing text files for lyrics and comma-separated tags, with options to adjust generation parameters like temperature and guidance scale to fine-tune the output.

Who it’s for

It is designed for researchers, developers, and music creators who want to experiment with AI-generated music. It is particularly useful for those who need a local, open-source alternative for music production, remixing, or cover-song generation using reference audio or MIDI inputs.

Highlights

  • Includes a multilingual music language model that generates songs based on lyrics and descriptive tags.
  • Features a high-fidelity 12.5 Hz music codec for clear audio reconstruction.
  • Provides a dedicated transcription model (HeartTranscriptor) to help with lyrics.
  • Offers MuLaCover for creating controllable cover songs and remixes from reference audio or MIDI.
  • Fully open-sourced under the Apache 2.0 license.

Related

  • Project
  • Project
  • Project
  • Project
  • Project