Concept: pick a song you love, point at a style reference (a YouTube link or just a name — "the PS22 kids-choir cover of Lisztomania"), and get a full cover of your song in that style: same melody and lyrics, new arrangement and vocals.
./preview.sh
then open http://localhost:5555 — one command builds the web version of the app
and serves UI + API together on one port (mock engine, zero configuration, works
offline). First run installs dependencies and builds; later runs start in seconds.
REBUILD=1 ./preview.sh forces a fresh web build; PORT=8080 ./preview.sh changes
the port.
Current status: v0.2 app built (backend + Expo app + engine test kit) with the premium feature set — multi-take covers, refine loop, Style Studio, tiers. See FEASIBILITY.md for the research — engine landscape, API access, legal constraints, unit economics, and the staged roadmap — and docs/PRODUCT.md for the product vision.
The short version:
- Feasible. Full-arrangement style transfer is a shipped capability (Suno Covers, MiniMax Music Cover API, open-source ACE-Step). The gap in the market is legal, not technical.
- Partially API-gated. The best engine (Suno) is invite-only; a legitimate v1 stack exists today via MiniMax's cover API on resellers (~$0.15/cover) with ACE-Step as self-hosted fallback.
- One hard legal rule: never extract audio from YouTube links — use link metadata to build a text style prompt instead.
- Next step: a ~$2 listening test of the candidate engines (spec in the report) before writing any product code.