Start with a conversation, not a file

Before you upload a loved one’s voice, have the conversation. What do they want the voice used for? Are there contexts where they would find it distressing to hear themselves? Who are they happy having access? What would they want to happen if the voice were misused?

If they are still able to answer, the answers guide the consent form. If they are not, the conversation is with next of kin, and the consent is conservative by default.

The six-question checklist

Whose voice is it? You should be a next-of-kin or explicitly authorised by the person. Neighbours and friends do not count.
Have they agreed? Preferably in writing with a date. If they cannot agree, document the surrogate-consent chain carefully.
What are the allowed use-cases?Personal family use only? Extended family? Educational archive? Public memorial? Be specific.
What are the off-limit contexts?Public entertainment? Commercial use? Political? Religious? Sexually explicit? Write these down.
Who can invoke it? Only the uploader? Any family member? Anyone with the invitation link?
How is it revoked? What happens if the person changes their mind, or if you change your mind, or if a sibling disagrees?

The upload form walks you through all six. Skipping is not allowed; the listing cannot be submitted until each field is filled.

What the consent manifest records

A signed JSON document stores:

The consenting party’s identity and relationship.
Date of consent.
Allowed use-case tags (personal, family, educational, etc.).
Off-limit tags (public, commercial, political, explicit).
Invocation allow-list (which account IDs can invoke).
Revocation contacts (who can revoke, how).
An expiry policy (does consent renew periodically?).

The manifest is the legal and technical record. Every use of the voice is gated on the manifest; uses outside the declared scope are refused at runtime.

Three uses we refuse regardless of consent

Impersonation of the person to third parties without disclosure. The runtime always flags “this is a voice skin of X, recorded in Y” when the voice is used to speak to anyone who is not already in the invocation allow-list.
Financial-instruction voices. A persona cannot authorise a bank transfer or subscription. Period.
Explicit content. Sexual content with an uploaded personal voice is refused regardless of any consent document.

What happens after a death

If the voice’s owner dies, the persona moves to archive mode unless the original consent explicitly permitted continued use. Archive mode maintains access for the declared family members but refuses new uses outside the original scope. A beneficiary chain can be declared at upload (similar to the estate chain in GeraClone).

If there is family disagreement

Voice personas of deceased or incapacitated people are a common source of family disagreement. Our rule: any declared next-of-kin can pause the persona. Resuming requires the dispute to be resolved outside our platform. We are not the family judge.

When not to upload

If the person explicitly declined consent when alive, do not upload, period.
If the upload is driven by urgent grief rather than considered choice, wait. Personas do not resurrect people. Professional grief support is Cruse in the UK, or the local equivalent.
If the person is a public figure, rights around posthumous likeness vary by jurisdiction; extra care required.
If you are not confident in the provenance of the recordings, do not upload. Our verification layer catches some cases; not all.

Upload Your Grandfather's Voice — The Ethical Checklist