754,000+ skills indexed

Chia1104COMMUNITY

Explore all skills by this author →

Verified:Website 𝕏@chia7423 GitHub

heartmula

Generates full songs from lyrics and style tags using open-source HeartMuLa models.

1|Updated Aug 11, 2026

Search Tags:#gpu #music generation #open-source #lyrics #AI audio #heartmula #song generation

One-click install

npx skills add https://github.com/Chia1104/agent-air --skill heartmula-chia1104

Or copy as Structured Prompt for Agent▼

Please help me install this Agent Skill.
Skill: heartmula
Source: https://github.com/Chia1104/agent-air/tree/main/skills/hermes/media/heartmula
Command: npx skills add https://github.com/Chia1104/agent-air --skill heartmula-chia1104

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Generating complete songs with vocals from text requires either paid services like Suno or complex local model setups. This Skill provides a tested, working installation and usage workflow for HeartMuLa, an open-source music generation model family, including fixes for known dependency conflicts and source code patches required for current library versions. ## Core Features & Use Cases - Local Song Generation: Generate full MP3 songs (48kHz stereo) from bracketed lyrics files and comma-separated style tags using the 3B or 7B HeartMuLa models. - Dependency & Patch Guidance: Applies required upgrades to datasets and transformers, plus two source patches (RoPE cache reinitialization and HeartCodec loading) needed for transformers 5.x compatibility. - VRAM-Constrained Operation: Supports lazy loading to run on GPUs with as little as 8GB VRAM, or multi-GPU splitting across devices. - Use Case: A user wants an open-source Suno alternative to create a wedding song locally. Provide lyrics with [Verse]/[Chorus] structure and tags like "piano,happy,wedding", and generate a 4-minute MP3 in roughly real time on a single GPU. ## Quick Start Ask the agent to install HeartMuLa and generate a song from your lyrics file and style tags, saving the result as an MP3.

Frequently Asked Questions about heartmula

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

FAQPage Schema

How do I generate music from lyrics with an open-source model?▼

HeartMuLa generates full songs from a lyrics file using bracketed structural tags like [Verse] and [Chorus], plus a comma-separated tags file for style. Run the run_music_generation.py example with your model path, lyrics, and tags to produce a 48kHz stereo MP3.

What is a good open-source Suno alternative for local song generation?▼

HeartMuLa is an Apache-2.0 licensed family of music foundation models comparable to Suno, generating songs conditioned on lyrics and tags with multilingual support. It includes the HeartCodec audio codec, HeartTranscriptor for lyrics transcription, and HeartCLAP for audio-text alignment.

How much VRAM does HeartMuLa need to run?▼

HeartMuLa requires a minimum of 8GB VRAM with lazy loading enabled, which peaks around 6.2GB for the 3B model. 16GB or more is recommended, and multi-GPU setups can split the language model and codec across separate devices.

Can HeartMuLa run on CPU or macOS?▼

CPU mode works via --mula_device cpu but is extremely slow, taking 30-60+ minutes per song and requiring 12GB+ free RAM. Triton is unavailable on macOS, so GPU acceleration is Linux/CUDA only; users without NVIDIA GPUs should use a cloud GPU or the online demo.

Why does HeartMuLa fail after installing with newer transformers versions?▼

The pinned dependencies conflict with current pyarrow and huggingface-hub 1.x, requiring upgrades to datasets and transformers. Two source patches are also needed: RoPE cache reinitialization after meta-device loading, and ignore_mismatched_sizes=True for HeartCodec checkpoint loading.

Why are my style tags ignored during HeartMuLa generation?▼

Tags being ignored is a known upstream issue (#90) where lyrics tend to dominate the generation. Experimenting with tag ordering may help, and avoid using bf16 for HeartCodec since it degrades audio quality.

Related Skills

Explore More Design & Creative Skills→

heartmula

Generates full songs from lyrics and style tags using HeartMuLa open-source music models.

sound-design

Design layered audio plans with loudness targets and dialogue cleanup.

listen

Transcribe speech with Whisper and analyze music features with librosa locally.

songwriting-and-ai-music

Generates song lyrics, parody adaptations, and Suno AI music prompts with structure and metatags.