heartmula

Configure and run HeartMuLa music generation from lyrics and tags.

3|Updated Apr 4, 2026
One-click install
npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill heartmula-twjarviszyra-web
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: heartmula
Source: https://github.com/twjarviszyra-web/hermes-unbound/tree/main/skills/media/heartmula
Command: npx skills add https://github.com/twjarviszyra-web/hermes-unbound --skill heartmula-twjarviszyra-web

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

HeartMuLa turns lyrics and tags into complete songs, removing the need to stitch together separate tools for music generation, codec handling, and model loading.

Core Features & Use Cases

  • Full-song generation from lyrics plus comma-separated tags.
  • Multilingual, local, or offline music workflows on CUDA or CPU.
  • HeartCodec-based audio reconstruction with practical setup guidance for checkpoint compatibility.
  • Use case: create a multi-minute song from verse, chorus, and mood tags, then save the output as MP3.

Quick Start

Ask me to install HeartMuLa, apply the compatibility fixes, and generate a song from your lyrics and tags.

Frequently Asked Questions about heartmula

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

FAQPage Schema
How do I generate a full song from lyrics and tags locally?

Full-song generation from lyrics and tags requires configuring the music synthesis workflow with CUDA-aware PyTorch. You input verse, chorus, and mood tags to produce a complete multi-minute audio track locally.

Can I use multilingual lyrics for audio synthesis on a GPU?

Multilingual lyric conditioning allows audio synthesis across different languages. This workflow operates on local or cloud CUDA GPUs, processing international text inputs to generate corresponding full songs.

Do I need CUDA-aware PyTorch to run music generation?

CUDA-aware PyTorch is required for GPU-accelerated music generation workflows. While the system supports offline CPU synthesis, running on CUDA GPUs significantly accelerates full-song generation and audio reconstruction processes.

Why does checkpoint loading fail with mismatched sizes during model setup?

Checkpoint loading fails with mismatched sizes due to transformer compatibility issues. Applying specific compatibility fixes for transformers, datasets, and RoPE cache rebuilding resolves these errors to successfully load model checkpoints.

What is HeartCodec audio reconstruction and when is it needed?

HeartCodec audio reconstruction is the process of decoding generated song representations into playable audio. It is needed during the final stage of the music generation pipeline to output multi-minute songs as MP3 files.