Costs $1 from your balance per run.

Reference Revoice (Singing)

Revoice a vocal to sound like any reference voice the user SUPPLIES as a separate recording. Use this ONLY when a voice-to-clone audio file is actually provided — if the user instead just NAMES a singer with no reference audio, use curated_revoice_singing instead. Requires BOTH inputs: the voice to change and the voice to clone (never pass the same file as both, and never borrow an unrelated file as the reference). Costs $1. Inputs must be isolated to the vocal first — overlapping vocals will not perform well, so use ONE lead vocal (chain Stem Separation first if needed). Nearly-equal-length inputs get passage-matched cloning; different lengths use one clear reference passage for the whole track. Running Restore Sample Rate on the result is recommended.

How it works

Drop a file or paste a direct link, pick what you want, confirm. Runs take about 7 minutes. Stems come back as individual WAV files plus a zip of everything, with download links valid for an hour. Failed runs are refunded automatically.

Pick by ear, not by benchmark

Benchmarks do not always match your material. Run the same song through several models at once, listen to the results side by side, and keep the one that sounds right. Failed models refund automatically so you only pay for stems you receive.

Use it from your assistant

Music Assistant, our MCP server, runs the same operations from inside Claude, ChatGPT, Cursor, Zapier, and Make. One URL, one free key.

Connect Music Assistant