How to Use ElevenLabs: The Beginner's Guide (First Voiceover, v3 Audio Tags, and the Free-Credit Line)

CodexFuse Team · 2026-10-28T14:54:49.718Z · 17 min read

ElevenLabs turns text into lifelike speech. This beginner's guide walks you through your first voiceover, picking a voice, v3 audio tags, voice cloning rules, and how far the free credits go.

Learning how to use ElevenLabs takes about thirty minutes of real practice, and it can replace a voiceover budget you do not have. ElevenLabs is an AI voice platform that turns written text into natural-sounding speech, and it is the tool most small creators reach for when they need narration for a YouTube video, a podcast intro, or an online course without hiring a voice actor. This guide takes you from a blank account to your first finished voiceover: set up the free plan, write a script that fits your credits, pick a voice by its ID, direct the performance with v3 audio tags, and export audio you can drop straight into your editor. If you want a quick spoken version of a page first, our text-to-speech tool does the same job in your browser with no account at all.

The short version: you sign up free, paste a short script into Text to Speech, choose a voice and a model, generate a 15-second test, fix the words that sound wrong, and only then generate the full recording line by line. Everything else in ElevenLabs is a refinement of that loop. Cloning, dubbing, and long-form projects all build on it.

How Do You Use ElevenLabs?

To use ElevenLabs, open Text to Speech, paste a short script, choose a voice and a model, generate a sample, correct the pronunciation and delivery, and then generate and export the full recording. Your first voiceover takes seven steps:

1. Create a free account at elevenlabs.io with an email address or Google sign-in.
2. Open Text to Speech and paste a short test script: the hardest 15 to 30 seconds of your real script, not an easy sentence.
3. Choose a voice from the library and note its voice ID, so you can find the same voice again later.
4. Choose a model: Eleven v3 for expressive delivery, or Multilingual v2 for stable long narration.
5. Generate the sample and listen to every word, especially names, numbers, and acronyms.
6. Fix the script: spell out awkward words and add audio tags, then generate the full script in short sections.
7. Download the audio as MP3 for sharing or WAV-quality audio for editing, and save the voice, model, and settings with the project.

What Is ElevenLabs?

ElevenLabs is an AI audio company founded in 2022 that builds text-to-speech, voice cloning, dubbing, and conversational voice tools. Its main product is Text to Speech: you type or paste text, pick a voice, and download spoken audio that sounds close to a human recording, with intonation, rhythm, and emotion. The platform also includes Speech to Text, Sound Effects, Voice Design, Music, Dubbing Studio, and Projects (also called Studio) for long-form work such as audiobooks and podcasts.

ElevenLabs lists more than 100 ready-made AI voices across styles such as narration, news, conversational, and character voices. Its general listing covers 32 languages, while the Eleven v3 model supports more than 70 languages. That split matters in practice: if your script is in a less common language, check that the model you pick actually supports it before you spend credits on a test.

One warning before you start: realistic AI voices in monetized content need disclosure on some platforms, and a clone can make a speaker appear to say words they never recorded. Treat the voice the way you would treat a photo of a real person. Get permission, keep the approval in writing, and say the audio is AI-generated wherever the platform asks for it.

How Much Does ElevenLabs Cost in 2026?

ElevenLabs runs on credits, and one credit buys roughly one character of generated speech. The free plan gives you 10,000 credits a month, which works out to about 10 minutes of audio. The table below shows the plans exactly as listed on ElevenLabs' pricing page when this guide was checked on October 28, 2026. Prices exclude taxes, and first-month promos change often, so re-check the page before you pay.

The line that catches most beginners is the commercial license. The free plan does not include commercial use: the pricing page marks it with a cross for Free and a check for every paid plan. If your audio will appear in a monetized YouTube video, a client project, or an ad, you need Starter or higher. The free plan is for practice and personal projects only.

Two more numbers worth knowing before you generate anything. Extra minutes beyond your allowance cost about $0.36 a minute on the free plan, about $0.20 on Starter, about $0.18 on Creator, and about $0.17 on Pro and above. And failed generations still consume credits, which is why this guide keeps telling you to test short first: a bad 10-minute generation costs the same as a good one.

How to Use ElevenLabs: Your First Voiceover in 7 Steps

Work through these steps in order. Each one is short, and skipping one is how beginners burn a month of credits in an afternoon.

Step 1: Create your account

Go to elevenlabs.io and click "Try for free" or "Get started free." Sign up with an email address and password, or with Google sign-in. The free plan starts immediately, with 10,000 credits for the month and no card required.

Step 2: Write your script before you open the tool

Write the script in plain words the listener should hear. Spell out numbers, dates, and acronyms the way you want them spoken: write "twenty twenty-six" if you want the year read that way, and "Doctor Smith" instead of "Dr. Smith" if the abbreviation comes out wrong. Unusual names, symbols, digits, and emoji can destabilize pronunciation, so clean them out before you paste anything in.

Count the words first. At a normal speaking pace of about 150 words a minute, a 300-word script is about two minutes of audio and about 1,800 credits. The script-length table later in this guide turns any word count into a credit number, so you never start a generation blind.

Step 3: Open Text to Speech and paste a test

Open Text to Speech from the dashboard and paste the hardest 15 to 30 seconds of your real script. Include the brand name, a number, a question, and the strongest emotional line. A neutral, easy sentence can make the wrong voice or model look acceptable, and you will only discover the problem after you have generated the full script.

Step 4: Choose a voice

Browse the voice library by style: narration for explainers and audiobooks, conversational for casual videos, news for formal reads, characters for fiction. Listen to the samples, pick one, and write down its voice ID. Voice names can change or repeat, but the ID is stable, and pinning the ID is how you get the same voice back for episode two.

Step 5: Choose a model

Pick the model for the job, not the newest name on the list. Eleven v3 is the expressive choice for dramatic or emotional delivery and multi-speaker dialogue. Multilingual v2 is the steady choice for long narration. Flash v2.5 is the fast, cheap choice when latency or credits matter more than expressiveness. The model table below spells out the trade-offs.

Step 6: Generate the sample, then fix the script

Generate the test and listen to all of it. When something sounds wrong, fix the input first: rewrite the awkward phrase, spell out the acronym, or add an audio tag. Change one setting at a time after that, so you can hear what actually fixed the problem. Only when the test sounds right should you generate the full script, in short sections, listening across every join.

Step 7: Download and file the audio

Download MP3 for common playback and sharing, or a WAV-quality option when the next step is editing. Save the approved audio, the script, the voice ID, the model, the settings, and the date together with the project. Six months from now, that note is the difference between matching the voice and starting over.

How to Pick a Voice (and Why You Should Pin the Voice ID)

Voice selection has the largest effect on your output, followed by the model and then the settings. That ordering comes from ElevenLabs' own product guide, and it is the reason this article spends more time on the voice than on any slider. A great voice on default settings beats a poor voice with perfect settings every time.

Start with the library voices, not a clone. The library covers the common jobs: a warm narrator for explainers, a crisp newsreader for announcements, a casual conversational voice for vlogs. Test two or three voices on the same 15-second script and compare them on the same sentence. The voice that handles your hardest line — the brand name, the number, the question — is the one to keep.

One deadline to know about: ElevenLabs' Default voices expire on December 31, 2026. If a production depends on one of those voices, migrate to a persistent replacement now and approve the new sound before the deadline. Do not build a series on a voice that will disappear at the end of the year.

Which ElevenLabs Model Should You Use?

ElevenLabs has four current model families, and each one fits a different job. The table below follows the model documentation as it stood in September 2026, so verify the model picker at generation time: model availability changes, and an old recommendation can quietly point at a retired model.

For a first voiceover, the choice is simple: use Eleven v3 if the script has emotion, dialogue, or audio tags, and Multilingual v2 if it is a long, steady narration. Flash v2.5 is for drafts and quick tests, not for the final recording you publish.

ElevenLabs Audio Tags: Directing the Performance

Audio tags are the feature that separates an ElevenLabs beginner from someone who sounds like they hired a director. In the v3 model, you write tags in square brackets inside your script, and the voice performs them. The most useful tags are [excited], [whispers], [laughs], and [sighs], and v3 also supports multi-speaker dialogue, so two voices can trade lines in one generation.

Use tags the way a director uses stage directions: sparingly, and only where the line needs them. A tag on every sentence sounds like a parody. One [whispers] before a secret, one [laughs] after a joke, and one [excited] at the reveal will do more than a script full of them. If a tag does not work, check that you are on the v3 model: tags are a v3 feature, and other models ignore them.

Speed is the other control worth learning. ElevenLabs documents speed values from 0.7 to 1.2, and extreme values can hurt quality. Stay near 1.0 for narration, slow to about 0.9 for instructional audio where the listener needs to follow steps, and never push past 1.2 to save credits: a fast voiceover that nobody understands is wasted money.

How Long Is 10,000 Credits, Really?

Credits feel abstract until you turn them into words and minutes. The math is simple: one credit buys about one character, an average English word is about six characters including its space, and a normal speaking pace is about 150 words a minute. The table below is computed from those three numbers, so you can plan any script before you paste it in.

Read the table as a budget. A 10-minute YouTube narration is roughly 1,500 words, which fits inside the free plan with a little room for one re-take. A 20-minute video needs about 3,000 words, which is already past the free allowance and into Starter territory. And a 5,000-word audiobook chapter is almost exactly one Starter month. Count the words, check the table, and pick the plan that covers the script plus one full re-generation: you will use it.

How to Clone a Voice in ElevenLabs

Voice cloning copies a real voice so ElevenLabs can speak new words in it. ElevenLabs offers two kinds. Instant Voice Cloning needs a short sample and is available from the Starter plan. Professional Voice Cloning needs a longer, cleaner sample, takes a few hours to build, and is available from the Creator plan. A cloning guide by Quantum AI Labs puts the practical minimum at a 10-second sample, and says 30 seconds to 3 minutes of clean audio gives a much better result.

The rule that matters most is not technical: you need the speaker's written consent before you clone their voice, and the consent must cover the message, the channels, and the duration. ElevenLabs runs a verification step as a safeguard, but that check does not replace the speaker's permission or your legal obligations. Never clone a voice you do not have the rights to, and never skip the consent step for a commercial project.

Record clean samples if you clone your own voice. Use one speaker, one microphone in the same position, low background noise, and the emotional range the output needs. Poor, short, or noisy samples reduce quality, and no setting fixes a bad sample. After the clone is built, verify the final script with the speaker: a clone can make them appear to say words they never recorded, so approval must cover the generated message, not only the source sample.

ElevenLabs Studio: Long-Form Projects

Text to Speech is for short recordings. Studio, the long-form workspace also shown under the ElevenCreative product line on the pricing page, is for audiobooks, podcasts, and course narration. You build a project from a script or an imported document, generate it chapter by chapter, and keep the voice, model, and settings consistent across the whole project.

The free plan includes 3 Studio projects, and Starter raises that to 20. If you are making an audiobook, work in Studio from the start: generating twenty chapters as separate Text to Speech jobs and stitching them together by hand is slower, and the joins will sound uneven. Generate in manageable sections inside the project, keep enough neighboring context for natural transitions, and listen across every join before you export.

Once the narration is exported, the job moves to editing. Our Audacity beginner guide covers trimming the audio, removing breaths and mistakes, and exporting a clean file, and our Google Vids beginner guide shows how to lay narration over video. If the narration belongs in a short-form video, our CapCut beginner guide covers the edit, and if the project also needs music, our Suno beginner guide covers generating it.

ElevenLabs vs Murf: What to Compare Before You Pay

Murf is the best-known alternative for team and business voiceover work, and the honest answer is that neither tool wins for everyone. Compare them on the five things that decide the real cost, using each company's current pricing page rather than a roundup written months ago.

• Commercial license: ElevenLabs includes commercial use from the $6 Starter plan. Check whether the alternative includes it at the same price or gates it behind a higher tier.
• Voice quality on your script: generate the same 30-second test on both tools and listen blind. Library size means nothing if the voice you need sounds wrong.
• Cloning rules: compare the sample length, the verification step, and which plan unlocks cloning. ElevenLabs splits Instant and Professional cloning across Starter and Creator.
• Team seats: ElevenLabs adds seats at Scale (3 seats, $299) and Business (10 seats, $990). If a team shares one account, seat pricing decides the bill.
• Long-form workflow: compare Studio-style projects and dubbing against the alternative's project tools. A cheap per-minute price is no bargain if you rebuild every chapter by hand.

Do not pay for a year of either tool before the test. Both tools change prices and plan names during the year, and the first-month promos on ElevenLabs (Starter $1, Creator $11) make a one-month trial cheap. Test on your real script, on your real voice, and keep the receipt.

7 Mistakes That Waste ElevenLabs Credits

Every mistake below costs credits, and every one is avoidable. They are the reason this guide repeats the same two rules: test short first, and count the words before you generate.

1. Generating the full script first. A 10-minute generation with the wrong voice costs the same as a good one. Test 15 to 30 seconds, then generate the rest.
2. Using the free plan for commercial work. The free plan has no commercial license. A monetized video narrated on free credits is a terms violation waiting to be found.
3. Picking a voice by name. Names repeat and change. Pin the voice ID, or episode two will sound like a different narrator.
4. Building a series on a Default voice. Default voices expire on December 31, 2026. Migrate now or re-record later.
5. Pasting a raw script. Abbreviations, symbols, and emoji break pronunciation. Clean the text and spell out the numbers first.
6. Tagging every line. Audio tags are stage directions, not seasoning. One tag per beat; a script full of [excited] sounds like a parody.
7. Cloning without written consent. Verification is a safeguard, not permission. Get the speaker's written approval for the message, the channels, and the duration.

Sources

• ElevenLabs pricing page (elevenlabs.io/pricing), read October 28, 2026: plan prices, credit allowances, commercial-use terms, overage rates, and the ElevenCreative product line.
• Magic Hour, "How to Use ElevenLabs for Voiceovers (2027)" (updated October 25, 2026): the eight-step voiceover workflow, model table, speed range of 0.7 to 1.2, Default-voice expiry on December 31, 2026, and cloning responsibility rules.
• AIPure, "How to Use ElevenLabs: Step-by-Step Guide": sign-up steps, more than 100 voices, 32 languages, ElevenLabs founded in 2022, and Eleven v3 support for more than 70 languages with audio tags.
• Quantum AI Labs, "Create Your Own AI Voice Clone using ElevenLabs": cloning steps, 10-second minimum sample, 30 seconds to 3 minutes for a better result, and the legal-rights step.
• HeyGalatea, "How to Use ElevenLabs for Voice Over, Sound Effects, and Music" (dev.to, October 25, 2026): count words first, pin the voice ID, generate line by line, and the v3 audio-tag workflow.

Frequently Asked Questions

Is ElevenLabs free?

Yes. The free plan costs $0 and includes 10,000 credits a month, about 10 minutes of audio, plus Text to Speech, Voice Design, Music, and 3 Studio projects. It does not include a commercial license, so it is for practice and personal projects only.

Can I use ElevenLabs audio commercially?

Commercial use is available on paid plans, starting with Starter at $6 a month. The free plan does not include it. Re-check the current plan terms before you publish, because commercial terms and plan names change.

Can I clone someone else's voice?

Only when you have the authorization and rights required for that voice, message, and use. ElevenLabs runs a verification step, but it does not replace the speaker's permission or your legal and contractual obligations. Get written consent that covers the message, the channels, and the duration.

Which ElevenLabs model should I use?

Use Eleven v3 for expressive performance, dialogue, and audio tags; Multilingual v2 for stable, longer narration; Eleven v3 Conversational for real-time speech; and Flash v2.5 when low latency or generation cost matters. Verify the current model picker, because model availability changes.

How many words can I generate with 10,000 credits?

About 1,650 words, which is roughly 10 to 11 minutes of audio at a normal pace of 150 words a minute. One credit buys about one character, and an average English word is about six characters including its space, so 10,000 credits cover about 1,650 words.