heartmula

Generate full music tracks from lyrics and tags using HeartMuLa models.

Updated Apr 9, 2026
One-click install
npx skills add https://github.com/MarbleSodas/Mavis --skill heartmula-marblesodas
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: heartmula
Source: https://github.com/MarbleSodas/Mavis/tree/main/skills/media/heartmula
Command: npx skills add https://github.com/MarbleSodas/Mavis --skill heartmula-marblesodas

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

HeartMuLa turns your written lyrics plus musical tags into a complete, music-ready song—without requiring you to compose music manually.

Core Features & Use Cases

  • Lyrics + Tags Song Generation: Produces full songs conditioned on bracketed lyrics structure and comma-separated tags.
  • Local/Offline Music Pipeline: Runs music generation locally using HeartMuLa models and HeartCodec reconstruction.
  • Multilingual Creative Support: Generates music with lyrics for multiple languages and supports experimentation with different input styles.

Quick Start

Ask the agent to generate an MP3 song from your lyrics file and tags file using the HeartMuLa 3B model with lazy loading enabled, saving the result to a specified output path.

Frequently Asked Questions about heartmula

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate songs from lyrics locally using a GPU?

You can generate songs locally from lyrics by running HeartMuLa with a Python 3.10 environment and configuring GPU or CPU device settings for stable offline music synthesis.

Can I use comma-separated tags and bracketed lyrics to generate Suno-like music offline?

Yes, full music track generation is conditioned on user-provided bracketed lyrics and comma-separated tags, enabling Suno-like creative workflows for offline local-first audio synthesis.

Do I need a HeartCodec checkpoint setup to output MP3 files from text?

Yes, producing MP3 output requires a HeartMuLa/HeartCodec checkpoint setup, utilizing audio codec reconstruction to convert your lyrics and tags into a complete music-ready song.

Does local music generation support multilingual lyrics input?

Local music generation supports multilingual song synthesis, allowing you to experiment with generating full music tracks using lyrics written in multiple languages.

What are the limitations of running local-first audio generation models?

Local-first audio generation requires a Python 3.10 environment and specific HeartCodec checkpoints, with optional multi-GPU splitting needed for stable execution when handling large model inference.

What's the best way to configure multi-GPU splitting for local music generation?

For stable local music generation execution, configure multi-GPU splitting during device setup to distribute the HeartMuLa 3B model inference workload across available hardware resources.