Matching a voice

What voice match does

You can hand the clone an optional voice sample, and it will aim to make its generated speech sound like that voice. The sample is a file already sitting in your project's storage, and it feeds the model as a reference audio track right alongside the reference video.

Why it asks for consent

A voice is a likeness, exactly like a face. So the instant you attach a voice-match reference, the tool makes you attest that you have the right to use that voice. This is not a polite checkbox you can wave past. The server enforces it: a run that sends a voice reference without the consent attestation is rejected before it ever dispatches, so ticking the box in the browser alone will not sneak it through.

In plain terms, only match a voice you own or have clear permission to use. Your own recorded voiceover is fine. A voice actor who agreed is fine. A licensed sample is fine. Cloning a stranger's voice, or a celebrity's, is not.

When to skip it

Leave the voice reference off and the clone simply generates its own synchronized speech from the script. That is the normal path, and it is what most runs use. Voice match is for the narrower case where the delivery genuinely needs to sound like one particular, authorized voice.