Skip to content
Repic is live. Create your account →
Repic
LoginSign up
AI TwinsLive

An AI version of you, your voice, your writing, your look.

Keep as many versions as you like; exactly one is active per modality. Consent-first, own-voice-only, always.

AI Twin Studio, voice tab, demo brand data
Real product, demo brand data.
The order matters

Consent first. Then, and only then, likeness.

A twin built before you agreed to it is not a feature, it is a problem. The sequence below is the product sequence, not a marketing one.

  1. 01

    You are asked

    Explicit, recorded consent for voice and for likeness, separately.

  2. 02

    The reference is yours

    Your samples, your recordings. Withdrawing consent removes them.

  3. 03

    The likeness resolves

    Only after both of the above. Abstract here by design: we do not depict a real face in marketing.

  4. 04

    The voice follows

    The ribbon further down this page is the voice moment, and it takes the handoff from here.

Voice

Clone your own voice, consent first, own-voice-only always. Use it for Voice Chat or narration, and retire it whenever you choose.

Versioning: keep as many versions as you like, exactly one is active per modality.
Voice cloning, the whole flow, in the open

Four steps. Consent is step one, not a checkbox.

1

Confirm the consent statement

On the recording screen, this is the gate everything else waits behind.

2

Record or upload your voice

A few minutes of natural speech is enough to start.

3

Processing

Your voice-cloning key never leaves our servers, and is never sent to your browser.

4

Your voice is ready

Activate it for Voice Chat, or keep it as one of your versions.

Consent-first, always: you can only clone your own voice on Repic. Every clone starts with an explicit consent step. Read our trust principles →

Voice Chat

Talk to the Repic Assistant out loud, and hear it answer in your own cloned voice if you choose.

No clone active? It falls back to a studio voice, stated, never faked.

On-brand by default

What this engine reads from your brand:

Tone: how your twin speaksVoice rules: words it never usesAudience: who it writes forVisual Direction: how your look reads
The same chips you'll see already filled in inside the studio, describe once, remembered everywhere.
Examples

Real twins only.

These slots hold real sample-brand twins as they land, a character sheet and a labeled voice demo. Never staged, never a stranger's face. Status today: Live

character sheet · sample twin
voice demo · labeled AI audio
writing sample · before / after
Deep specs

For the detail-minded.

Versions & activation

Keep as many twin versions as you like, exactly one is active per modality (voice, writing, visual). Switch the active version anytime; delete any version anytime, and its data goes with it.

Voice data & keys

Your consent is recorded against the twin as proof the clone was yours to make. The voice-cloning key stays server-side, never sent to any browser. Full detail in the privacy policy's voice-data section.

Statuses & history

Twin training runs show honest statuses, and every version is listed with what it was trained on. Confidence labels on visual angles are real, not decorative:

readyprocessingfailed
Go deeper

3 min read · 6 sections

What an AI Twin is here, and what it refuses to be

An AI Twin is a model of one specific modality of you: how you sound, how you write, or how you look. Not a persona you invented, not a stock presenter, not a composite. You. That distinction is the whole product, and it is also why this engine has more restrictions on it than anything else in Repic.

Repic is consent-first by construction: voice and likeness twins require an explicit consent step and a spoken consent recording, the own-voice-only rule is enforced, and deletion actually deletes. Consent is a step in the product, not a clause in a document nobody reads: there is an explicit consent gate before a voice twin can be created, the consent itself is recorded in your voice, and the own-voice-only rule means you cannot train a twin on someone who did not sit down and agree.

Keep reading

You can keep as many versions of a twin as you like, because voices and looks are not one-shot decisions. But exactly one is active per modality at any time. That constraint exists so output is never ambiguous about which version of you it used, which matters the first time you need to answer that question about something that has already been published.

Step by step, with nothing skipped

  1. Explicit consent, recorded

    Before anything is trained you pass a consent gate and record a spoken consent statement. No consent recording, no twin. This is enforced in the product, not promised in the terms.

  2. You provide your own material

    Your voice, your writing, your likeness. The own-voice-only rule blocks training on third parties, which is the single most abused capability in this category.

  3. The twin is versioned, not overwritten

    Every twin you build is kept and labelled. Exactly one per modality is active, so every generated artefact traces to one known version.

  4. Deletion actually deletes

    Withdraw consent and the twin and its source material go. The biometric policy states what is held, for how long, and what deletion covers.

Voice twin versus voice clone

Before

A generic clone: upload thirty seconds of audio, get a synthetic voice, no record of who agreed to what, usable on anyone whose audio you can find.

After

A Repic voice twin: a consent gate, a spoken consent recording tied to the twin, an own-voice-only rule, one active version at a time, and deletion that removes the source material as well as the model.

The capability is similar. The gate around it is the product. If you publish a biometric policy and a DPA, the gate has to be real, because it is trivially checkable by anyone who cares to look.

What this does not do

It will not clone anyone but you

No celebrity voices, no colleague who "would not mind", no public-figure likeness. The own-voice-only rule is the point of the feature, not a limitation of it.

It is not a deepfake tool

Twins exist to let you produce more of your own work in your own voice. Impersonation of anyone else is prohibited by the Acceptable Use Policy and blocked by the consent gate.

It does not hide that it is generated

Repic does not encourage passing synthetic output off as unassisted work. Disclosure norms are covered in Trust and Safety.

Straight answers

What exactly counts as biometric data here, and what do you keep?

Voice and likeness material used to build a twin is treated as biometric data. The Biometric and Likeness Policy sets out what is collected, the consent gate it sits behind, how long it is retained, and your deletion rights, in specific terms rather than general reassurance.

Copy link to this question
Can I create a twin of my co-founder or a client?

Only if they go through the consent gate themselves, on their own account, with their own recorded consent. There is no path where you consent on someone else behalf, and that is deliberate.

Copy link to this question
Why only one active twin per modality?

So that any piece of output has exactly one answer to "which version of me made this". Keeping many versions is useful; having two of them simultaneously authoritative is how you end up unable to explain your own archive.

Copy link to this question
If I delete a twin, is my published content affected?

No. Deleting a twin removes the model and its source material going forward; work you already generated and published is yours and stays as it is. What ends is the ability to produce more from that twin.

Copy link to this question

Free, no signup

Not ready to sign up? Run the brand voice generator instead.

Paste two or three things you have written and get a measured voice profile: sentence rhythm, register, who you address, plus rules and a rewrite.

Try it free →
Repic is open

Sound like you, everywhere, with consent built in.

Sign up →