Record. Type. Listen.
Open the Audio pane, record 6–8 seconds of any voice in-app (or pick a file — reference clips are normalized automatically), type your text, and generate. The model extracts a speaker embedding from the clip — a compact fingerprint of timbre and delivery — and conditions the synthesis on it. No fine-tune, no waiting, no dataset.
The embedding pipeline was validated bit-for-bit against the reference implementation — the clone you get locally is the clone the model's authors shipped.
Your voice is a password. Treat it like one.
Voice is biometric data — the same signal that unlocks bank phone lines and impersonates you to your family. Cloud cloning services keep your reference uploads on their terms. Here, the recording, the embedding, and every synthesized sample stay on your Mac: no account, no upload, no retention policy to read.
Clone voices you have the right to use — yours, a consenting collaborator's, a character you own. The tool is local; the responsibility is too.
- No upload — the reference clip never leaves the machine.
- No account — nothing to subscribe to, nothing to leak.
- Works offline — once the model is downloaded, airplane mode is fine.