The rules we set before we needed them.
Repic is built on someone's voice, likeness and brand. That only works if the boundaries are clear – so here they are, in full, with nothing hidden behind an accordion.
Last updated 13 September 2026
Six controls, and the weight behind each one.
Claims are cheap. These are the six things the studio actually does, spelled out underneath in full.
- 01
Consent
A voice clone starts with a recorded yes, in that voice, and only your own.
- 02
Verification
Misuse is designed against, not policed after the fact.
- 03
Ownership
Your voice and your twins are yours to remove, whenever you decide.
- 04
Permission
The assistant asks before anything that matters happens.
- 05
Data handling
We store what runs the studio, and nothing you cannot take back.
- 06
Disclosure
What we mark as AI today, and what we do not yet, stated in full.
Every voice clone starts with a recorded yes.
Before Repic learns a voice, you confirm a specific consent statement on the recording screen. No consent, no clone. And the rule underneath it is absolute: you can only clone your own voice. Even with their permission, you cannot clone someone else's on Repic.
“I consent to Repic creating an AI voice from my own voice.”
The cloning key is generated and used server-side only. It is never sent to your device, so it can't be copied from there.
Hard to misuse, by design.
Consent gating is enforced, not suggested. No celebrity or third-party cloning. And the cloning key never leaves our servers – it isn't sent to your browser, so it can't be lifted from there. Suspected misuse jumps the queue at contact.
Your voice and twins are yours to remove.
You own your voice model and every twin version. Delete a twin version in the product and its voice model goes with it – no support ticket, no waiting. Close your account and everything follows, surviving only briefly in backups.
Deleting a version removes its voice model too. Nothing lingers on the active clone.
Nothing that matters happens without your yes.
The assistant explains its intent first. Anything that changes or deletes your brand waits behind an Allow / Disallow card. “Just do it” phrasing can't skip the gate, and Disallow means nothing changes – verified.
It states its intent in plain words before doing anything.
Changes and deletions are gated; reads and navigation run instantly.
Your exclusions – “keep the logo”, “not salesy” – are respected and forwarded.
We store what runs the studio – nothing you can't take back.
Your persona fields, your runs, and your uploaded assets – held so the studio can do its job. We don't sell your data or train models for other customers on it. Export or delete any of it whenever you like. Full detail in the privacy policy.
What we mark as AI today, and what we do not yet.
Images Repic generates carry a machine-readable mark: studio slides, including the slides in a downloaded ZIP, exported PDFs, generated and edited images, and AI Twin character sheets. Where the image model has already marked its output, as Google's image models do with a signed Content Credentials (C2PA) manifest, Repic stores that mark exactly as it arrived. Where nothing has marked it, Repic adds the IPTC digital source type to its metadata, which is not a signed manifest. Neither is a visible label. Video, generated audio and brand kit visuals are not marked yet. The EU AI Act requires machine-readable marking from 2 December 2026, and this section changes as each gap closes. The biometric policy, section 12, predates this marking and is being revised after counsel review.
A visible label like this is not yet applied to anything Repic exports or publishes.
These are ours to keep – and answerable.
We wrote these before we needed them, and the same two people who built the permission gate answer when you write in. If we ever fall short of what's on this page, tell us – it reaches us directly.
Build on a studio that asks first.
Further reading: Consent-first voice cloning, explained