Gary4Claude – Skill de Produção Musical com IA
betweentwomidnights/gary4claude
Rede de APIs de geração e transformação de áudio para produção musical assistida
Ver no GitHubSkill que integra múltiplas APIs de geração e transformação de áudio para agentes Claude. Oferece acesso a síntese de áudio a partir de texto, continuação de tracks, transfer de estilo, injeção de LoRAs e ferramentas utilitárias para manipulação de WAV.
Instalação
npx convez-ai betweentwomidnights/gary4claude1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Gerar tracks musicais completos a partir de descrições em texto
- Estender e continuar composições existentes com consistência estilística
- Aplicar estilos de áudio e adaptadores LoRA a gravações
- Automatizar pipeline de produção musical com encadeamento de gerações
Veredito

Skill experimental mas funcional para automação de produção musical. Requer familiaridade com workflows de geração de áudio e APIs externas; bem documentado com exemplos práticos de encadeamento de operações.
README
A tradução para o português deste README ainda está na fila.
gary4claude
AI music production skill for Claude Code agents. Gives your agent access to a network of music generation APIs — text-to-audio, MusicGen continuation, structured synthesis, ACE-Step cover/continuation, LoRA style injection, and audio transforms.
The Cast
| Name | What it does |
|---|---|
| Jerry | Text-to-audio generation with BPM-aware looping (stable-audio + finetunes) |
| SA3 | Stable Audio 3 generation, loops, continuation, transform, and blendable LoRAs |
| Gary | MusicGen continuation — chain generations to build long tracks |
| Foundation | Structured synth/sample generation with randomizable presets |
| Carey | ACE-Step cover/continuation with LoRA captions; optional stem generation |
| Terry | Audio style transfer — transform audio with a text prompt (MelodyFlow) |
Install
Copy the skill files into your project's .claude/skills/ directory:
# From your project root
mkdir -p .claude/skills/gary4claude
curl -sL https://github.com/betweentwomidnights/gary4claude/archive/main.tar.gz | tar xz --strip-components=1 -C .claude/skills/gary4claude
Or clone directly:
git clone https://github.com/betweentwomidnights/gary4claude .claude/skills/gary4claude
Usage
In a Claude Code session:
Use the gary4claude skill to make me a 2-minute DnB track with AI vocals.
Or get specific:
Use the gary4claude skill to make a 2-minute 174bpm DnB track in A minor,
then use Carey's billie LoRA adapter with cover/complete.
Agents should build songs sequentially: make one request, wait for the returned audio, save it, then decide the next step. Do not launch parallel generations unless the user explicitly asks for alternate takes.
Skill Files
gary4claude.md— Top-level overview, routing, workflows, audio flow between servicesjerry.md— stable-audio: text-to-audio, loops, finetune promptssa3.md— Stable Audio 3: loops, continuation, transform, LoRA blendinggary.md— MusicGen: continuation, retry, transform, model switchingfoundation.md— Foundation-1: randomize + generate, BPM handlingcarey.md— ACE-Step: cover, complete continuation, LoRAs, captions, optional lego stemsterry.md— MelodyFlow: audio transforms via Gary
Utility Scripts
The scripts/ directory contains stdlib-only WAV helpers for agents that should not need to write their own audio glue code:
scripts/mix_wavs.py— mix compatible WAV loops/stemsscripts/tile_wav.py— repeat a loop to a target bar count, duration, or reference WAV lengthscripts/trim_wav.py— crop by seconds or barsscripts/wav_info.py— inspect duration, sample rate, channels, and peakscripts/gary_preflight.py— verify a WAV is long enough for Garyprompt_durationand get a tiling recommendationscripts/base64_wav.py— encode/decode WAV files for API payloads
Example:
python scripts/gary_preflight.py seed_4bar.wav --prompt-duration 6
python scripts/tile_wav.py drums_4bar.wav drums_8bar.wav --bars 8 --bpm 174
python scripts/mix_wavs.py drums_8bar.wav synth_8bar.wav -o seed.wav --gain 0.8 --gain 0.7 --normalize
python scripts/tile_wav.py seed.wav seed_2min.wav --seconds 120
API Base URL
All services are hosted at https://g4l.thecollabagepatch.com
License
MIT