MuLaCover

Controllable AI cover-song and music-remix generation. Give the model a reference song, your own lyrics and a one-line style prompt; MuLaCover transcribes the reference melody, chords and drums with YourMT3 + ChordNet, encodes the style with Qwen3-Embedding, generates HeartCodec audio tokens with a 3B llama backbone, and decodes a 48 kHz WAV.

• Model: HeartMuLa/MuLaCover
• Code: HeartMuLa/MuLaCover on GitHub
• Paper: HeartMuLa

License

The MuLaCover weights are released under CC BY-NC 4.0 with additional terms in the model repository's MODEL_LICENSE. Generated outputs are restricted to noncommercial use unless separately authorized by MuLa Labs.

Example audio credits

The bundled reference excerpts are 45-second clips of Josh Woodward — "Perfect" and Josh Woodward — "Let's Pretend", both licensed CC BY 4.0 (source: https://www.joshwoodward.com, mirrored on the Internet Archive).