tinker-ai-music
A local lyrics-to-song plugin for TINKER, powered by audio.cpp via audiocpp-static.

Features
- Local lyrics-to-song with ACE-Step 1.5 Turbo (Q8 / BF16 GGUF)
- Style prompt + lyrics (supports
[Verse]/[Chorus]tags) - Lyrics language selector (Chinese / English / Japanese / Korean)
- Duration control (15–180s) and optional negative prompt
- Translate style to English when
tinker-translateis installed - Auto model download via Downloader into
~/.tinker/models/when missing - Generation queue with WaveSurfer preview, play, and save as WAV
- Local inference with auto-selected backend (Metal on macOS, CPU elsewhere)
Installation
Download and install TINKER, then run:
bash
npm i -g tinker-ai-musicUsage
- Enter lyrics to sing (structure tags like
[Verse]/[Chorus]work well) - Describe the musical style (e.g. “cinematic synth pop with clear vocals”)
- Select lyrics language; optionally set a negative prompt, duration, model, and download source
- Click Generate — missing models start downloading automatically
- Preview queued results in the audio list; save WAV when ready
Style prompts work best in English. The Turbo Q8 package is about 5.8 GB. Runtime data lives in ~/.tinker/tinker-ai-music/ (logs / outputs). Models are shared under ~/.tinker/models/.