How to clone your voice with ElevenLabs, and the consent rules that apply

This page may contain affiliate links from which Blur Studio earns commissions.

ElevenLabs is a text to speech service, and voice cloning is the feature that lets it generate speech in a specific person's voice rather than one of its library voices, built from an audio sample of that person speaking. This guide covers the two cloning tools ElevenLabs offers, practical recording tips, and the consent rules you need to take seriously before cloning anyone's voice but your own.

Instant vs Professional cloning

ElevenLabs splits cloning into two tiers, available at different plan levels:

• Instant Voice Cloning builds a usable clone quickly from a shorter audio sample. It's included starting on the Starter plan ($6 a month, or around $5 a month on annual billing, as of October 2026), which also carries ElevenLabs' commercial license.
• Professional Voice Cloning builds a higher-fidelity clone from a larger amount of audio, aimed at use cases where accuracy matters more, like a recurring brand voice or audiobook narration in your own voice. It's available from the Creator plan ($22 a month, around $18 a month annual) and up, and ElevenLabs requires identity verification before granting access to this tool as part of its safety measures.

For a quick test or a one-off project, Instant cloning is usually enough. For anything you'll reuse repeatedly as a recognizable brand or personal voice, Professional cloning is worth the extra setup step. Worth noting before you choose: moving from an Instant clone to a Professional one later generally means re-recording a longer, cleaner sample rather than simply upgrading the existing short clip, so if you already suspect a voice will become a long-term brand asset, it's worth going straight to Professional cloning and investing in a proper recording session from the start, rather than starting quick and redoing the work later.

Recording tips for a good clone

• Record in a quiet room. Background noise, echo from a bare room, or a fan running nearby all get baked into the clone and come out in every generation afterward.
• Use a consistent, reasonably good microphone. You don't need studio gear, but a phone held close and steady beats a laptop's built-in mic picking up room noise from across a desk.
• Speak naturally, not in a monotone. Read a few sentences with normal variation in pace and tone rather than a flat, robotic read, this gives the model more to work with.
• Avoid long silences or heavy breathing sounds in the sample. Trim obvious dead air before uploading if you can.
• For Professional cloning specifically, more clean audio generally produces a more accurate result. ElevenLabs' own product pages describe its exact current sample-length guidance better than I can state reliably here, since this detail changes between model versions. Check the live recording requirements on ElevenLabs' voice cloning setup page before you record, rather than assuming a fixed minute count.

Consent and legal limits

This is the part to take seriously, and I'm not a lawyer, so treat this as a plain-language summary, not legal advice. Before cloning anyone's voice, including your own if you're acting on behalf of a company, read ElevenLabs' current terms of use and safety policy directly on their site, since the exact legal language can change and I'm not going to paraphrase it as if it were settled. Earning money from your own voice covers the options and limits.

What I can confirm directly from ElevenLabs' own safety page as of October 2026: Professional Voice Cloning requires "technological verification" before you're granted access, and ElevenLabs explicitly blocks "the cloning of celebrity and other high risk voices" as a safety measure against impersonation. That tells you two things in practice. First, ElevenLabs itself treats cloning a well-known person's voice as a risk it actively tries to prevent, not something its own system is designed to permit. Second, verification exists specifically because voice cloning carries real potential for misuse, fraud, impersonation, deepfake audio, so the platform has built friction into the process on purpose.

The safe, simple rule: only clone your own voice, or a voice where you have clear, documented permission from the person whose voice it is, ideally in writing if it's for a client's commercial project. If you're building a business feature around a cloned voice, for example a company spokesperson's voice used across marketing material, get that permission in writing before you start, and keep it on file. Several of ElevenLabs' own detailed legal pages were not accessible to me during this research, so for the exact current legal terms, read them directly on ElevenLabs' site before relying on any secondhand summary, including this one. Keeping a former employee's intro message covers the consent side.

Common cloning mistakes

• Recording too little variety. A sample that's all one flat tone, reading a single type of sentence, gives the model less to generalize from than a sample with some natural variation in pace and emotion, which shows up later as a clone that sounds oddly stiff on anything outside that narrow range.
• Not testing the clone against your actual use case before relying on it. A clone that sounds convincing on a short demo sentence can behave differently on a long paragraph, a question, or a sentence with an unusual word in it. Generate a few full paragraphs of real content before building a workflow around the clone. Giving your AI receptionist a real receptionist's voice is one such use case.
• Treating a cloned voice as a finished asset instead of something you may need to refine. If the first clone doesn't sound quite right, re-recording a cleaner, more varied sample is usually more effective than trying to fix it entirely through prompting or voice settings after the fact.
• Forgetting to manage who else on your team can access or regenerate from a cloned voice, particularly on a shared workspace plan. Treat a cloned voice, especially a Professional clone of a real spokesperson, the same way you'd treat any other sensitive brand asset.

Where this fits into a real project

A common legitimate use is a founder or spokesperson who wants their own voice behind training content, a product demo, or an IVR phone greeting, without re-recording every update themselves. Clone your own voice once with ElevenLabs' voice cloning tool, then generate new lines from text whenever the script changes, instead of booking studio time for every small update. It also solves a scheduling problem a lot of small teams run into: the one person whose voice the brand is built around usually isn't available to re-record every time a product detail or a line in a training module changes, and a clone removes that bottleneck entirely once it's set up properly. Putting your own voice on every video without recording shows the workflow.

How I'd handle this for a client

When a client wants a cloned voice built into a product feature, I confirm in writing, before any build work starts, exactly whose voice it is and that we have clear permission to use it, then handle the technical setup on either tier depending on how often it'll be reused. Get in touch if you want a voice feature built around a real person's voice, done properly from the consent step onward.