How to Clone a Voice with ElevenLabs (Ethically): Setup, Consent Rules, and Quality Tips

Cloning a voice with ElevenLabs means building a reusable digital model of a specific person’s voice from recordings of them speaking, so the model can then read any new text in that voice. There are two ways to do it: Instant Voice Cloning, which copies the sound of a voice from one or two minutes of audio and is ready almost immediately, and Professional Voice Cloning, which trains a dedicated model on thirty minutes to three hours of audio and takes hours to finish — but sounds much closer to the real person. And there is one rule that comes before any of it: clone your own voice, or a voice you have clear permission to clone, and keep proof of that permission. Everything in this guide follows from that rule. It covers both clone types step by step — the consent checks, the recording craft, the creation flows, the verification step, the first test, and the tuning — using ElevenLabs’ own documentation as reported on 29 October to 1 November 2026, with the pricing figures re-checked on ElevenLabs’ own pricing page on 10 November 2026. Checked 10 November 2026. ElevenLabs changes its plans, models and features often, so treat the prices and model details here as current at that date and re-check the company’s own pages before you pay for anything. This is a documentation-based guide, not a hands-on test: no tool was hands-on tested for this article, so quality judgements here are judgement based on documented capabilities and the company’s own claims, never tested results.

The Short Answer: What Cloning a Voice Means — and the Rule That Comes First

<img src="

” alt=”Close-up of a studio microphone with its red recording light on, a hand hovering over the stop button, shallow depth of field, dark background”>

A voice clone is not a recording. It is a model built from recordings — a piece of software that has learned the sound of a voice well enough to speak new words in it. You feed it audio of a person speaking, ElevenLabs builds the clone, and from then on you can type any text and hear it read back in that voice. The clone lives inside your ElevenLabs account: you cannot download it as a file, and you use it through text-to-speech, long-form Studio projects, dubbing, voice agents, or the API with its voice ID.

The rule that comes first is simple, and it is not a formality. Cloning a living person’s voice without their explicit, informed consent violates ElevenLabs’ Prohibited Use Policy and can lead to a permanent ban from the platform — and, depending on what you do with the audio, legal exposure of your own. The safest way to think about it is this: your voice is yours to clone. Anyone else’s voice needs their clear permission, in writing, kept on file, before you record a single second. If you cannot get that, do not clone the voice. Use a Voice Library voice or a Voice Design voice instead — no real person is involved in either.

Cloning is one feature of a much larger platform. ElevenLabs also does text-to-speech, a library of shared voices, dubbing, voice agents, real-time speech-to-speech, transcription, sound effects and music — and if you want the full map of what the platform does before you commit to one feature, our ElevenLabs Explained guide covers all of it, with cloning as one section among fourteen. This guide does the opposite job: it stays inside cloning and walks through it properly. For the company behind the tool — how ElevenLabs grew and how big it has become — see What Is ElevenLabs?

The rest of this guide is built in the order you should actually work. Consent and the rules come first, before any recording. Then setup, then the recording craft, then how much audio each clone type needs, then the two creation walkthroughs, the verification step, the first test, tuning, putting the clone to work, what it costs, and finally what to do when it goes wrong — and when not to clone at all.

Instant vs Professional: Which Clone You Actually Need

ElevenLabs offers two kinds of clone, and the choice between them is the single most important decision in this guide. They are not two price tiers of the same thing. They are two different techniques with different costs, different audio requirements, and different rules about whose voice they can copy.

The trade is time and effort against fidelity. An instant clone is fast and cheap to make, and for short-form narration, drafts, social clips and prototyping it is often good enough. A professional clone demands hours of clean recording and a wait of several hours for training, and in return it sounds much more like the real person and holds up over hours of material. The decision rule, as ElevenLabs’ own guidance frames it, is about where the voice will be heard: short clips and drafts point to instant; audiobooks, courses, a channel where listeners hear you every week, or a brand voice across hours of material point to professional.

Two limits apply to both types. The Free plan includes neither kind of cloning — you need a paid plan before either option appears. And a clone’s accent or tone cannot be changed after creation, so the audio you record is the voice you get; if the accent is wrong, you change the audio, not a setting. The product-map view of how cloning sits beside the rest of ElevenLabs is in the guide linked above, and this guide picks up where its cloning section deliberately stops.

Before You Record Anything: Consent, Rights and the Rules

Do this section before you open a microphone. Consent is not a checkbox at the end of the process; it is the thing that decides whether the process is allowed to start at all.

The two clone types handle consent differently, and the difference matters. For an instant clone, the consent step is a confirmation: before you save the voice, you confirm that you have the right and consent to clone it. There is no technical verification of identity in that flow — ElevenLabs takes your word for it, and the responsibility is yours. For a professional clone, the check is technical as well as contractual: you must prove, by reading lines aloud into a microphone in a live verification step, that the voice belongs to you. That is what makes professional cloning own-voice-only in practice, not just in the terms. Even with another person’s consent, you cannot professionally clone their voice. If someone else wants a professional clone, they must create and verify it on their own account — and they can then share it with you.

ElevenLabs’ Prohibited Use Policy bars cloning a living person’s voice without their “explicit and informed consent or other valid authorization”. It also bars submitting voice recordings of anyone under 18 to create a voice model. Those are hard lines, not guidelines. The consequence for crossing them is severe: cloning another person without permission violates ElevenLabs’ terms and can lead to a permanent ban. ElevenLabs also operates an AI Speech Classifier to help detect audio generated by its own system, so synthetic audio can be identified as synthetic after the fact.

Disclosure is the second half of the rule. The EU AI Act’s transparency rules, in force since August 2026, require AI-generated content — including synthetic voices — to be identifiable as machine-made, and other jurisdictions are writing disclosure rules of their own. The practical rule is straightforward: if a voice is selling, informing or persuading, listeners deserve to know when it is synthetic. Label it. Do not use a cloned voice to imply an endorsement the person never gave. Do not run a voice agent that pretends to be human when a caller asks directly. And keep records of consent for every real person’s voice you use — the question, sooner or later, will be whether you can show permission, not whether you remember having it. For the wider technology behind all of this — how cloning and text-to-speech actually work — see AI Voices Explained: Cloning, Text-To-Speech And Where It’s Used.

Use this checklist before you record anything. Every item should be a yes; any no is a reason to stop.

• Whose voice is this? If it is not yours, do you have their explicit, informed consent — in writing, for this specific use?
• Is the voice owner over 18? If not, stop: recordings of anyone under 18 cannot be used to create a voice model.
• Is this a professional clone of someone else’s voice? If so, stop: professional cloning is own-voice only, even with consent. The owner must create and verify it themselves.
• Do you have a record of the consent saved somewhere you can find it later?
• Will listeners be told the voice is synthetic where it matters — in a video description, a show note, a spoken disclosure, or on-screen text?
• Will the clone be used to imply an endorsement, a statement, or a conversation the person never made? If so, do not use it.
• Is the use commercial? If so, is your plan one that carries a commercial licence? The Free plan is personal, non-commercial use only; Starter is the first plan with a commercial licence.

One more honest point: the consent confirmation in the instant-clone flow is only as good as the person clicking it. The platform cannot check whether you really have permission, so the check has to happen in your own head before you click. That is the whole point of putting this section before the recording section.

Setting Up: Your Account, Plan and the Voices Workspace

Setup is short, because the real work is the recording. Create an account on ElevenLabs’ site and open the dashboard. The work in this guide happens in the Voices section: that is where clones are created, where My Voices lists the voices you own, and where the Personal tab holds your professional clones once they are ready.

The plan you need depends on the clone type. The Free plan has no cloning of either kind, so it is not a starting point for this guide at all. Starter, at $6 per month, is the floor for instant cloning — and the first plan with a commercial licence. Creator, at $22 per month ($11 for the first month only), is the floor for professional cloning, and it includes one professional clone slot. Pro, at $99 per month, also includes one professional slot; Scale, at $299 per month, includes three; Business, at $990 per month, includes ten. Those prices and slot counts were re-checked on ElevenLabs’ pricing page on 10 November 2026 and matched the reports from 29 October to 1 November 2026. The full plan table, with credits and the traps to watch for, is in the pricing section near the end of this guide — read it before you pay for anything, because the cheapest plan that gives you the feature you need is usually the right one.

One thing to understand before you spend anything is the credit pool. Cloning itself has no separate fee — it is a plan feature. What costs credits is generating speech with the clone: text-to-speech costs about 1 credit per character on standard models, so roughly 1,000 credits buys about a minute of speech on Eleven v4 (that is arithmetic on the published rates, not a promise). Credits are a shared pool across the platform’s products at different rates per product, so a month of heavy dubbing can drain a balance that a text-to-speech user would have found generous. Unused credits on paid plans roll over for up to two months, up to twice your monthly quota, so a balance can reach at most three times the monthly quota; rollover does not apply to the Free plan.

Before you record, check two things in the dashboard. First, that your plan actually unlocks the clone type you want — an instant clone will not appear as an option on Free, and a professional clone will not appear below Creator. Second, for a professional clone, that you have a free slot: the slot count is per plan, and a full set of slots means you will need to delete an existing professional clone or upgrade before you can make another. Getting the plan right now saves you from recording three hours of audio for a clone type your account cannot create.

Recording a Good Sample: Room, Mic and Delivery

The quality of a clone is decided here, in the recording, more than anywhere else. ElevenLabs’ own guidance is blunt about it: “good consistent input = good consistent output”. The model copies inflection, breathing and noise as well as the voice itself, so a bad sample does not produce a slightly worse clone — it produces a clone that faithfully reproduces the bad parts.

Start with the rules that apply to both clone types. One speaker only, in a clean recording, with no background noise, reverb or music. Consistency matters as much as cleanliness: the same tone, pacing, microphone and recording quality across every file you upload. For a professional clone, record in a single speaking style and do not sing — a clone copies your pacing and accent, so read the way you want to sound. Speak naturally, at a normal pace, with normal inflection. The clone is learning you, not a character, so do not perform a voice you do not normally use. Read from a script in the style you will actually use the clone for: if it will narrate audiobooks, record audiobook-style prose; if it will read short social scripts, record that kind of delivery.

Levels are one of the few things ElevenLabs gives a number for. Aim for about −23 dB to −18 dB RMS, with a true peak of −3 dB. In plain terms: loud enough that the recording is clear and full, quiet enough that nothing clips or distorts. Record a test passage, check the meter, and adjust the gain before you commit to a long session. Then edit the audio before you upload it: cut out long pauses and filler sounds, because the clone will reproduce them. A thoughtful edit of two minutes of speech is worth more than ten unedited minutes.

For a professional clone, the setup is more deliberate, because you are recording hours of audio that has to stay consistent. ElevenLabs’ guidance describes an XLR microphone into an audio interface, a pop filter, about “two fists” between you and the microphone, and either an acoustically treated room or a space dampened with a thick duvet. That last detail is the one most people skip and most regret: a bare, echoing room puts reverb into every file, and reverb becomes part of the clone. A small room with soft furnishings, heavy curtains, or a duvet hung behind you is a better studio than a large empty one. Split multi-hour recordings into samples of about 30 minutes each — long sessions in one file are harder to manage, and a break between takes is a chance to check levels and rest your voice. Your voice changes as you tire, so record in sessions where you sound like yourself, not in one marathon.

Choose your script with the end use in mind. The clone learns from the words you read, so a sample made of a single repeated sentence teaches it very little. Use varied, natural text — real paragraphs, with questions, lists, names and numbers — so the model hears how you handle the range of speech it will have to produce. And keep the original recordings after you upload them. You cannot download a clone as a file, and recreating one later sounds slightly different even from the same audio, so the samples are your only way back.

How Much Audio Each Clone Needs — and Why More Is Not Always Better

The amount of audio each clone type needs is the clearest difference between them, and the instant clone’s answer is counter-intuitive: more is not better.

For an instant clone, ElevenLabs recommends 1–2 minutes of good audio. The hard minimums are small: 10 seconds total, with each file at least 5 seconds. But the ceiling matters more than the minimum. Do not record more than about 3 minutes for an instant clone — more than 2–3 minutes gives little improvement and can reduce the clone’s stability. That is the opposite of what most people expect, and it is the most common beginner mistake: feeding an instant clone ten minutes of audio in the hope of a better result, and getting a less stable voice for the trouble. One to two clean minutes is the target. Stop there.

For a professional clone, the numbers run the other way. Thirty minutes is the bare minimum, one hour or more is recommended, and ideally two to three hours. Split the recording into samples of about 30 minutes each. The reason is the technique: a professional clone trains a dedicated model on your recordings, and a dedicated model needs volume — hours of your voice, in one consistent style, so it can learn the full range of how you sound. Thirty minutes will produce a clone; two to three hours will produce a noticeably better one, especially over long narration where a thin model starts to drift.

Formats are straightforward. For an instant clone, MP3 at 128 kbps and above; higher bitrates make no significant difference. For a professional clone, MP3 at 192 kbps or higher — and since the instant-clone guidance repeats the 192 kbps figure, 192 kbps is a safe choice for both. Uncompressed formats such as WAV usually do not improve the clone and can cause upload problems, so there is no reason to reach for them. ElevenLabs states no file-size limit and no cap on the number of files: total runtime and audio quality matter, not how many files you upload.

The rule underneath all of this is that quality beats quantity. A noisy three-minute sample produces a worse clone than a clean one-minute sample, and no setting rescues a bad recording. That is worth repeating, because every other section in this guide — the first test, the tuning, the troubleshooting — is about fixing small problems in a good clone, not rescuing a bad one. If the sample is wrong, the answer is always to re-record, never to add more audio or turn a dial further.

Making an Instant Voice Clone, Step by Step

With a clean one-to-two-minute sample ready and consent settled, making an instant clone takes minutes. The steps below follow ElevenLabs’ own instant-cloning documentation as reported on 1 November 2026.

1. In the dashboard, open Voices and click the plus icon.
2. Select Instant Voice Clone.
3. Upload your audio file or files, or record directly in the browser, following the on-screen instructions. Each file must be at least 5 seconds; the total must be at least 10 seconds; aim for 1–2 minutes, and do not go past about 3 minutes.
4. Name and label the clone — an accent or style label helps you find it later — then confirm you have the right and consent to clone the voice, and click Save voice. This is the consent step from the earlier section, and it belongs here, inside the workflow, not as a footnote: if you cannot honestly confirm it, stop here.
5. Under Voices, open My Voices and click Use voice.
6. Test it with a paragraph from a real script — one you did not record — and listen on headphones and on a phone speaker. If it sounds off, try a different 1–2 minute sample before adding more audio. Adding more audio to an instant clone is the wrong fix; a different sample is the right one.

A few practical notes. Name the clone for the voice, not the project — “Narrator, warm, slow” will still make sense in six months, while “Test 3” will not. Labels are worth the ten seconds they take, because once you have several clones, the dashboard becomes hard to navigate without them. And test with text the clone has never heard: a paragraph from the sample itself will always sound good, because the model has already heard those exact words. The test that matters is new text, in the style you will actually use.

One more thing to know before you move on: an instant clone can struggle with uncommon accents. ElevenLabs’ documentation says so plainly, and it is a limit of the technique — a short sample gives the model less to learn from. If your accent is unusual and the instant clone sounds wrong, that is not necessarily a bad sample. It may be the point at which a professional clone, with hours of audio to learn from, is the honest answer.

Making a Professional Voice Clone, Step by Step

A professional clone is a project, not a task. It needs the Creator plan or higher, a free professional slot, hours of clean audio, a live verification step, and then a wait while the model trains. The steps below follow ElevenLabs’ own professional-cloning documentation as reported on 1 November 2026.

1. Go to Voices, click Create Voice, and choose Professional Voice Clone. You need the Creator plan or higher before a slot is available, so confirm the plan first — the option will not appear below Creator.
2. Add samples with Upload samples, or use Record yourself. The page shows length feedback as you add audio; ElevenLabs recommends at least an hour. The total must be at least 30 minutes, and ideally 2–3 hours, split into samples of about 30 minutes each.
3. Use Audio settings if you need to remove background noise or separate speakers. This is a cleanup step, not a rescue step — it can tidy a good recording, but it cannot fix a noisy one.
4. Verify your voice. This is the live captcha step, covered in full in the next section: you read lines aloud so ElevenLabs can confirm the voice belongs to you.
5. Wait for fine-tuning. Track progress in My Voices. Training usually takes 3–6 hours; ElevenLabs’ docs also give 6–24 hours, and a queue can make it longer, so plan for the range, not the short end. If you try to use the clone before it finishes, you will get the error “No model found for this voice” — that is the sign training is not finished, not a fault in the clone.
6. When it is ready, open the Personal tab under Voices and click Use.

The most important discipline in this whole flow is the one before step one: make sure your samples are right before you start. A professional clone trains on the audio you give it, and a flaw in the samples — a second speaker in one file, a change of microphone halfway through, an hour of tired, flat delivery — becomes part of the model. Re-recording after training means starting the wait again. So listen to every file before you upload it, check that the tone and room match across the set, and only then begin.

Also worth knowing: both clone types work with Eleven v4, ElevenLabs’ current flagship model, and an existing professional clone can be fine-tuned on v4 from My Voices, using the plus button next to Eleven v4. Multilingual v2 is also still recommended for quality, and Flash v2.5 is the fast option. The model choice matters less than the sample — but if you are choosing between them, v4 is the current default to start from.

Voice Verification: Proving the Voice Is Yours

Verification is the step that makes professional cloning own-voice-only in practice. It is a short live recording: you read on-screen lines aloud so ElevenLabs can confirm the voice in the samples belongs to the person sitting at the microphone. No one else can pass it for you, and that is the point.

Three details decide whether it passes. First, use the same or similar equipment you recorded the samples with, and a similar tone and delivery. A mismatch between the verification recording and the samples is the most common failure — a different microphone, a different room, or a different speaking style makes the two recordings sound like two different people, and the check fails. Second, read each verification line only once, then press Stop. Reading a line more than once can cause verification to fail. Third, read naturally: the check is comparing voices, not testing your acting, so deliver the lines the way you delivered the samples.

If verification fails, you can wait 24 hours and retry, or contact support. Before you retry, change something about the setup rather than repeating the same attempt: go back to the exact microphone and room you recorded in, match the tone of the samples, and read each line once. A failed verification is usually a mismatch problem, not a voice problem, and the fix is consistency.

Why does this step exist at all? Because without it, the own-voice-only rule would be words on a page. Anyone could upload someone else’s recordings and claim them as their own. The live captcha is what turns a policy into an enforced boundary: the voice has to belong to the person in front of the microphone, right now, reading lines they have not seen before. That is also why the verification step is the reason to get your samples right first — it is the gate between the recording work and the training wait, and a failure costs you a day.

The First Test: Judging Your Clone Honestly

A new clone deserves a proper first test, not a quick listen and a shrug. The test is simple to run and easy to get wrong, because the temptation is to test with text the clone has already heard — and that text will always sound good.

Use a paragraph from a real script that was not in the recording. Something the clone will actually have to do: a paragraph from the audiobook chapter, the course lesson, or the video script you made it for. Generate it, then listen twice: once on headphones, and once on a phone speaker. Headphones reveal the fine detail — odd breaths, mispronunciations, a flattened emotional line. A phone speaker reveals what most listeners will actually hear, which is a rougher, smaller version of the same thing. A clone that only sounds good on headphones is not finished.

Listen for five things. Accent: does it sound like you, or like a flattened version of you? Pace: does it rush, or drag, or pause in the wrong places? Odd breaths: are there breaths in the middle of sentences, or a gasp where none belongs? Mispronunciations: names, numbers, and unusual words are where clones stumble first. And flattened emotion: does a question sound like a question, and does an excited sentence sound excited? If you want to check the output word by word, ElevenLabs’ Cue timing layer is the transcript-and-timing technology behind checking a clone’s output against the text it was given.

The fix order matters. If the clone sounds unlike you, change the sample before you touch the settings. For an instant clone, try a different 1–2 minute sample — a different passage, a different session, a cleaner take. For a professional clone, re-record the weakest part of the audio. Only once the sample is right should you reach for the tuning settings in the next section, because settings adjust a good clone; they do not repair a bad sample.

And know what cannot be fixed after creation. A clone’s accent or tone cannot be changed after creation — change the audio instead, because small changes to the samples can make a large difference. If the accent is wrong, the answer is a new sample in the right accent, not a slider. The same applies to speaking style: the output replicates the style you recorded, so a flat, tired sample produces a flat clone. The honest verdict at the end of a first test is often “re-record”, and that is a normal outcome, not a failure.

Tuning the Clone: Stability, Similarity, Clone Strength and Audio Tags

Once the sample is right and the first test sounds close, tuning is where a good clone becomes a usable one. The settings live on the text-to-speech page, and the discipline is the same as the rest of this guide: change one thing at a time, generate the same paragraph at two or three settings, and listen like an editor.

Three settings matter for clones. Stability controls how consistent the output is: lower values give a broader emotional range and more variation between generations. Start around 50. Similarity controls how closely the output follows the original voice: high values can reduce naturalness, so a clone pushed to maximum similarity can sound accurate and lifeless at the same time. Start around 75. Clone Strength is shown for professional clones with Eleven v4: start at 0% for most voices, and raise it only if the clone sounds too far from the original. Eleven v4 has no Style or Speed slider, so do not look for one — those controls belong to other models.

The method is to generate the same test paragraph at two or three settings and compare them directly. Change Stability by a step, listen, and note what changed. Then change Similarity, listen, and note what changed. Do not change both at once, or you will not know which one fixed the problem — or caused it. Keep the test paragraph fixed across every comparison, and keep the listening conditions fixed too: same headphones, same volume, same room. A setting that sounds better on a phone speaker and worse on headphones is telling you something about the listening conditions, not the clone.

Eleven v4 also supports audio tags in the text itself — tags such as [whispering], [shouting] and [laughing] — which let you direct the delivery inside a sentence. ElevenLabs says tag-following is “not perfect yet”, so treat tags as a direction to the model, not a guarantee: a [whispering] tag should produce a quieter, closer delivery, but it may not, and the only way to know is to generate and listen. Multilingual v2 and the Flash models do not use these tags; they use SSML break tags for pauses instead. If you are working on one of those models, a break tag is the way to control a pause, not an audio tag.

A final caution on tuning: the settings are a finishing step, not a rescue. If the clone sounds unlike you, the sample is the problem, and the previous section’s fix order applies — change the sample first, tune second. A well-tuned clone built on a bad sample is still a bad clone, and no combination of Stability and Similarity will fix it.

Putting the Clone to Work: Narration, Dubbing, Studio and the API

A finished clone earns its keep in four places, all from the same voice, under one account and one credit pool. The first is text-to-speech narration: paste a script, pick the clone, and generate the audio. This is the everyday use — voiceovers for videos, narration for short content, a consistent voice across a series. If the video is silent footage that needs a voice, the pairing runs the other way too: AI video generators make silent clips, and a cloned voice is the natural voiceover route for them.

The second is Studio, ElevenLabs’ long-form workspace for audiobooks and courses. A professional clone is the right tool here, because long narration is where a thin instant clone starts to drift — the voice flattens, the pace wanders, and listeners who hear you every week notice. Studio lets you work sentence by sentence, regenerating a single line without re-recording a chapter, which is exactly the workflow a professional clone is built for.

The third is dubbing in your own voice: the same speaker, the same voice, in another language. That is the use case where a clone is most obviously valuable — your videos, your voice, in a language you do not speak. There is no separate dubbing workflow guide on this site yet; for the platform view of dubbing alongside the other features, see ElevenLabs Explained. The fourth is voice agents: a cloned voice can serve as the voice of an agent, so callers hear a consistent, recognisable voice instead of a generic one. For the speech-to-speech models a cloned voice can serve inside an agent, see ElevenLabs Eleven Realtime.

There is also an API route. You can use a clone elsewhere through the API with your key, passing its voice ID to the text-to-speech endpoint. That is how a clone leaves the dashboard and becomes part of an app or a workflow — but the clone itself stays in your account, and you cannot download it as a file. Keep your original audio samples: recreating a clone later sounds slightly different even from the same audio, so the samples are the only reliable backup you have.

One more route deserves a short paragraph, because it is the only way a clone can earn money directly. A professional clone can be shared in the ElevenLabs Voice Library after a review by the ElevenLabs team, and payouts come through Stripe Connect, based on use by paying customers. You choose a notice period from 30 days to 2 years; a longer one earns more. ElevenLabs does not publish payout rates, so do not estimate earnings — there is no published figure to estimate from. Instant clones and Voice Design voices cannot be shared in the Library. For the business context behind Library payouts — how ElevenLabs itself makes money — see How ElevenLabs Makes Money: Dubbing, Agents And Audiobooks. And for the finishing touch on a narrated video, ElevenLabs Instant Soundtracks generate the music bed for finished narration and video.

What It Costs: Plans, Credits and the Traps

Cloning has no separate fee — it is a plan feature, and what you pay for is the plan and the credits you spend generating speech. The table below is taken from ElevenLabs’ pricing page, re-checked on 10 November 2026. Prices are in US dollars, exclude taxes, and are the monthly-billing prices. Re-check the page itself before you pay for anything: prices are the fastest-moving facts in this article.

Annual billing charges ten months for twelve, giving monthly equivalents of $5 for Starter, $18.33 for Creator, $82.50 for Pro, $249.17 for Scale, and $825 for Business. Those equivalents were also confirmed on the pricing page on 10 November 2026.

Three traps deserve attention. The first is the shared credit pool. Generated speech draws on the monthly credit pool at about 1 credit per character for text-to-speech, so roughly 1,000 credits buys about a minute of speech on Eleven v4 — that is arithmetic on the published rates, not a promise — and the pool is shared across products at different rates. A month of heavy dubbing can drain a balance that a text-to-speech user would have found generous, so budget for everything the account will do, not just the cloning.

The second trap is rollover. Unused credits on paid plans roll over for up to two months, up to twice your monthly quota, so a balance can reach at most three times the monthly quota — and rollover does not apply to the Free plan. That makes a Creator plan more forgiving than it looks: a quiet month’s credits are still there next month, within the limit.

The third trap is the first-month price. The $11 Creator figure is month one only; month two is $22. The $1 Starter first month is likewise a promo, not a standing price. Budget on the list price, and treat any first-month offer as a trial discount, not the cost of the plan. The decision rule is simple: buy the smallest plan that gives you the licence and the feature you need. If you only need an instant clone for non-commercial use, Starter is the floor. If you need a professional clone, Creator is the floor — and the commercial licence comes with it from Starter upward.

When It Goes Wrong: Fixes, Limits and the Misuse Line

Most cloning problems are diagnosable, and most have the same fix: change the sample, not the settings. The table below covers the failures you will actually meet.

The hard limits are worth naming plainly. An instant clone has a stability ceiling: past about three minutes of audio it gets worse, not better, and it will never match a professional clone over long narration. A professional clone has a training wait: hours, not minutes, and a queue can make it longer. Language coverage is not settled in the documentation — one section reports 90 or more languages for Eleven v4, another lists a smaller model set — so test your target language before recording hours of audio. And an accent or tone cannot be changed after creation, in either clone type.

Then there is the misuse line, and it is the most important part of this section. Do not clone a living person’s voice without their explicit, informed consent. Do not clone a minor — recordings of anyone under 18 cannot be used to create a voice model. Do not use a cloned voice to imply an endorsement, a statement, or a conversation the person never made. Do not use a cloned voice for political robocalls or any deceptive call. The consequences are a permanent ban, legal exposure, and detection — ElevenLabs operates an AI Speech Classifier to help detect audio generated by its own system, and synthetic audio can be identified after the fact.

When not to clone at all is a question worth asking before you start. If you need a voice and no real person is involved, use the Voice Library or Voice Design instead — a shared professional voice, or a designed voice built from a description, gives you a voice with no consent question attached. If the voice belongs to someone else and they will not give permission, that is the answer too. And if the use is deceptive — a fake endorsement, a fake call, a voice pretending to be human — no consent can make it right, because the consent was never for that.

The honest verdict is that the limiting factor in voice cloning is rarely the model. It is the sample, the consent, and the discipline to listen. A clean two minutes, a clear permission, and an honest first test will take you further than any setting. Get those three right, and the clone will sound like the person it is supposed to sound like — because it was built from them, with their permission, and checked by someone who was actually listening.

Sources

All facts in this guide were checked against ElevenLabs’ own pages and the sources below on 10 November 2026. The pricing figures in the plan table were re-checked on ElevenLabs’ pricing page on 10 November 2026 and matched the reports dated 29 October to 1 November 2026. The cloning documentation figures — sample lengths, training-time range, verification steps and tuning settings — are reports of ElevenLabs’ voice-cloning documentation as read by independent checks dated 29 October to 1 November 2026; ElevenLabs’ own documentation pages could not be read directly during this research, so re-check the company’s own pages before you rely on a figure. ElevenLabs changes its plans, models and features often, so treat the prices and model details here as current at 10 November 2026 and re-check the company’s own pages before you pay for anything.

• ElevenLabs pricing page — plan prices, credits, cloning gates and professional clone slots, re-checked 10 November 2026
• ElevenLabs voice-cloning documentation and help centre (Instant Voice Cloning and Professional Voice Cloning guides) — sample lengths, formats, levels, creation steps, verification and tuning settings, as reported 29 October to 1 November 2026; exact documentation URLs could not be confirmed at draft time, so the guides are named here without a link rather than guessed
• ElevenLabs Explained: Text-to-Speech, Voice Cloning, Dubbing and Voice Agents — geroundtable.com — the full product map this guide sits inside, and the corroborating source for the consent rules, the AI Speech Classifier, EU AI Act disclosure and the licence boundary