The short answer
ElevenLabs is an AI audio platform for text to speech, voice cloning, transcription, dubbing, sound effects and music. Its newest model, Eleven v4, launched on 28 September 2026 and supports more than 90 languages with audio tags that steer emotion and delivery. The free plan gives 10,000 credits a month but no commercial licence. Voice cloning starts on the Starter plan, and Professional Voice Clones can only ever be made of your own voice.
| Made by | ElevenLabs (Eleven Labs Inc.), with offices in New York and London |
|---|---|
| Free plan | Yes, 10,000 credits a month, no commercial licence and no voice cloning |
| Models | Eleven v4 and v4 Turbo, Eleven v3, Multilingual v2, Flash v2.5, Scribe v2 for transcription, Eleven Realtime for agents |
| Best at | Natural voiceovers and narration, cloning your own voice, dubbing, voice agents |
| Shows sources | Not applicable; it generates audio from your text |
| Team plans | Scale (3 seats), Business (10 seats) and Enterprise |
| Your data | Inputs can be used to improve models unless you switch off Data use. Enterprise is excluded by default. |
What ElevenLabs is in November 2026
ElevenLabs started as a text-to-speech tool and is now a full audio platform. It sells three product lines. ElevenCreative is the web app for creators and marketers, with text to speech, Studio for long projects such as audiobooks, voice cloning, Voice Design, dubbing, sound effects and music. ElevenAgents is for building voice agents that answer phones and talk to customers. ElevenAPI gives developers the same models in their own products. There are also iOS and Android apps.
The model that matters most is Eleven v4, released on 28 September 2026 together with Eleven v4 Turbo. ElevenLabs describes v4 as its most emotive speech model and the first of a new generation. It supports more than 90 languages, up to 10,000 characters per request and natural multi-speaker dialogue. Eleven v4 Turbo brings the same technology to real-time use, with a median inference latency of about 100 milliseconds, which is what you need for a voice agent that does not leave awkward pauses. The older models are still available: Eleven v3 (70+ languages, 5,000 characters), Multilingual v2, which ElevenLabs calls the most stable for long-form work, and Flash v2.5, the fast, cheaper option for developers.
Two other models are aimed at businesses. Scribe v2, released in January 2026, is the speech-to-text model, with transcription in 90+ languages and speaker separation for up to 32 speakers. Eleven Realtime, launched on 20 October 2026, packages a whole voice agent into one speech-to-speech model in two sizes, v1 Mini for natural conversation at low cost and v1 Max for multi-step tasks such as customer support.
What ElevenLabs is genuinely good at
Voiceovers that sound performed, not read
The main reason people pick ElevenLabs is that its voices handle emotion and pacing well. Eleven v4 reads tone and context from your text, and you can steer it with audio tags in square brackets, such as [excited], [whispers] or [sighs], placed right before the line they should affect. For training videos, product demos, explainer reels and podcast intros, that often gets you a usable take in minutes instead of booking a studio.
Cloning your own voice
Instant Voice Cloning creates a clone from a short sample, and ElevenLabs recommends about one to two minutes of clean audio with no reverb or background noise. More than three minutes rarely helps. With Eleven v4, ElevenLabs says an Instant clone can work from as little as 10 seconds of audio, but the longer clean sample still gives you a safer result. Professional Voice Cloning trains a much more faithful model. ElevenLabs calls 30 minutes of audio the bare minimum and recommends at least an hour, ideally close to three hours. ElevenLabs says both kinds of clone are noticeably more accurate in Eleven v4. That makes it practical for a founder, trainer or creator to produce narration in their own voice without recording every word.
Reaching other languages
Dubbing translates and re-voices video while keeping the speaker's voice, and Dubbing Studio lets you correct the transcript and timing by hand. With 90+ languages in Eleven v4, a short internal video or a product walkthrough can be localised far faster than with traditional dubbing.
Transcripts and voice agents
Scribe v2 produces transcripts with word-level timestamps and speaker labels, which is useful for interviews and meetings. ElevenAgents and the new Realtime models let support teams build phone agents with knowledge bases, guardrails and controls for how long conversations are kept.
Where ElevenLabs falls short
- Credits disappear quickly. Every product draws from one monthly credit pool. Text to speech costs about one credit per character, but automatic dubbing uses 2,000 to 3,000 credits per minute and music about 900 credits per minute, so a few videos can empty a smaller plan.
- The free plan is for testing only. It includes 10,000 credits a month (text to speech costs about one credit per character) and three Studio projects, but no commercial licence, no voice cloning and no credit rollover.
- v4 can change how a voice sounds. ElevenLabs itself says Eleven v4 may sound substantially different from v3 for the same voice, because it copies the source more faithfully, including its accent and loudness. Test before you switch a series halfway through.
- Professional clones take effort. An hour or more of clean, consistent audio is a real recording job, and the verification step can fail if your microphone or tone is different from the samples.
- It still mispronounces. Names, acronyms and brand terms often need respelling or a second take, so always listen to the full file before publishing.
The voice-cloning consent rules, in plain English
Voice cloning is where most of the risk sits, and ElevenLabs' rules are stricter than many people expect. These come from its Prohibited Use Policy, last updated on 9 October 2026, and its help centre.
- No clone without consent. You may not clone, synthesise or impersonate the voice of any living person without their explicit and informed consent or other valid authorisation. For people who have died, cloning is banned where it is likely to cause harm or infringes post-mortem rights.
- You confirm consent for every Instant clone. When you save an Instant Voice Clone, the app asks you to confirm that you have the right and consent to clone that voice. Ticking the box makes you responsible for that statement.
- Professional clones are your own voice only. ElevenLabs' help centre says you cannot create a Professional Voice Clone of someone else, even with their consent. Every Professional Voice Clone goes through a verification step to confirm the voice is yours. If a colleague wants you to use their voice, they create and verify the clone on their own account and share it with you privately.
- Some voices are blocked. ElevenLabs says it blocks the cloning of celebrity and other high-risk voices.
- No minors and no politicians. You may not submit recordings of anyone under 18 to create a voice, impersonate political candidates or elected officials without authorisation, or run political campaigns without ElevenLabs' prior written consent.
- Audio is traceable. ElevenLabs says all audio its models generate can be traced back to the account that made it, and it offers a public AI Audio Classifier for checking whether a clip came from ElevenLabs. Violations can lead to removed content and suspended or terminated accounts, and ElevenLabs says it may report serious cases to law enforcement or regulators.
The practical rule for work: only clone your own voice, or a voice whose owner has given clear written permission for that specific use, and keep that permission on file. If a client asks for a famous voice, the answer is no.
How to get better results from ElevenLabs
- Write for the ear. Short sentences, contractions and punctuation that marks pauses produce a more natural read than text written for the page.
- Use audio tags sparingly. One tag at the start of a line, such as [warm] or [serious], steers the delivery. Too many tags in one paragraph make it sound acted. In the ElevenLabs app, the Enhance button can suggest tags for your text automatically, and an ellipsis (...) gives a short, hesitant pause.
- Spell out tricky words. Write numbers, dates and acronyms the way they should be spoken, and respell brand names phonetically if the first take gets them wrong.
- Record clean samples for cloning. Use the same microphone, the same room and one consistent speaking style. ElevenLabs notes that the clone copies everything it hears, including breathing, mouth clicks and background noise.
- Pick the right model. Eleven v4 for expressive narration, Multilingual v2 when you need the most stable long reads, and Flash or v4 Turbo when speed matters more than nuance.
- Generate in short sections. Long scripts are easier to fix a paragraph at a time, especially in Studio, than as one block.
Our ElevenLabs lessons cover writing a script for voice and recording a clean clone sample, and our stacks show how it pairs with a writing assistant and a video editor.
Privacy settings: the training opt-out
ElevenLabs says it uses certain data you provide to improve its audio models for everyone, unless you opt out. To switch it off, click your profile icon in the top right corner of the ElevenLabs platform, go to Terms and privacy, then Data use, turn off the toggle called Improve the models for everyone and click Update your choice. From then on, new data you submit is not used for training. By default ElevenLabs does not train on Enterprise customers' data, and it says it never uses your data to clone your voice without your consent. Enterprise also offers Zero Retention Mode, custom data processing terms and BAAs for HIPAA customers, and voice agents have their own settings for retention, audio saving and redaction of sensitive details.
Which ElevenLabs plan suits whom
| Plan | Good for | What it adds |
|---|---|---|
| Free | Hearing what the voices can do | 10,000 credits a month, no commercial use |
| Starter | Occasional commercial voiceovers | Commercial licence, Instant Voice Cloning, Dubbing Studio |
| Creator | Regular creators and trainers | Professional Voice Cloning (one slot), more credits |
| Pro | Heavy users and developers | Higher-quality audio output, many more credits |
| Scale and Business | Teams | 3 or 10 seats and 3 or 10 Professional Voice Clones |
| Enterprise | Regulated or large organisations | Custom terms, SSO, HIPAA BAAs, training excluded by default |
Pick Starter if you publish anything commercial, because the free plan does not allow it. Move to Creator only when you have recorded the hour of audio a Professional Voice Clone needs. Teams handling customer recordings should look at Enterprise for the default training exclusion and Zero Retention Mode, or at minimum switch off Data use on every account. Check ElevenLabs' pricing page for current prices and credit costs.
ElevenLabs or something else?
ElevenLabs leads on voice quality, but it is not the only choice. If you edit video and podcasts by editing a transcript, compare it with Descript in ElevenLabs vs Descript. For other voice tools see ElevenLabs alternatives and the best AI for voiceovers, or compare tools side by side in our comparison tool.
Prompts to copy
Rewrite a script for the ear
Rewrite this script so it sounds natural when read aloud by an AI voice. Use short sentences and contractions, write numbers and acronyms the way they are spoken, mark natural pauses with commas or full stops, and keep it under [number] words. [paste script]
You get: A voice-ready script (use it in any writing assistant, then paste into ElevenLabs)
Warm product explainer with audio tags
[warm] Most teams lose an hour a day to status updates. [short pause] Here is how [product] gives that hour back. [upbeat] You connect your tools once, and every morning your update writes itself.
You get: A friendly, paced read for a demo or reel
Serious internal announcement
[calm, serious] I want to share an important change to how we handle [topic]. Starting on [date], [what changes]. [reassuring] Nothing changes for your day-to-day work, and your manager will walk you through the details this week.
You get: A steady, trustworthy tone for staff updates
Two-speaker podcast intro
Speaker 1: [excited] Welcome back to [show name]! Today we are talking about [topic]. Speaker 2: [laughs] And I have strong opinions. Speaker 1: You always do. Let's start with the question everyone asks: [question]?
You get: A natural back-and-forth opening using multi-speaker dialogue
Training module narration
[clear, measured] Module three: handling a customer complaint. [short pause] By the end of this lesson you will know the three steps: listen, acknowledge, resolve. Let's take them one at a time, starting with listening.
You get: Calm, clear narration for e-learning
Recording script for a clean voice clone
Write a two-minute script I can read aloud to record a voice-clone sample in my natural speaking style for [use, such as training videos]. Mix statements, a question and a short list, avoid tongue twisters, and keep the tone [conversational or formal] throughout.
You get: A consistent sample script (draft it in any assistant, then record it for ElevenLabs)
Short ad read with a clear call to action
[energetic] Stop guessing what your customers want. [short pause] [product] shows you, in plain words, every Monday morning. [confident] Try it free at [website].
You get: A punchy 10-second ad read
Pronunciation check paragraph
Our new partner, [brand name, spelled phonetically], launches on the fourteenth of November. Call us on [number written as spoken] or email hello at [domain] dot com.
You get: A quick test of names, dates and numbers before a long render