How to use ElevenLabs, step by step, for someone starting from zero

Setting up your account

Go to elevenlabs.io and sign up for ElevenLabs, the AI voice platform for text to speech and voice cloning, with an email address, no credit card is needed for the free plan. Once you're in, you'll land on a dashboard with several tools listed, Text to Speech, Speech to Text, Voice Design, Dubbing, and more. For a first file, Text to Speech is the one you want. Our page on whether ElevenLabs is free shows exactly where the free plan stops.

Before you generate anything: write the script properly first

The single biggest factor in how good your first file sounds isn't the voice you pick or the settings you tweak, it's how well your script is written for being read aloud. Text written to be read silently on a page often sounds stiff spoken aloud, long sentences with multiple clauses, abbreviations, and unusual punctuation can all trip up a text to speech engine. Before pasting your script in, read it aloud yourself once. If you stumble over a sentence reading it yourself, the AI voice is likely to stumble there too. Break up long sentences, spell out abbreviations the engine might mispronounce (write "ASAP" as "as soon as possible" if it matters), and add commas where you'd naturally pause for breath.

Generating your first file

1. Open the Text to Speech tool from the dashboard.
2. Type or paste the script you want read aloud into the text box.
3. Choose a voice from the library on the side. There are dozens available, filter by gender, accent, or style if you have something specific in mind.
4. Press the generate button. Within a few seconds, you'll get a playable audio clip.
5. Listen to it. If it sounds right, you're done with this step. If something feels off, pronunciation, pacing, emotion, adjust the voice settings before regenerating.

Settings explained simply

Underneath the voice selector, you'll see a few sliders. Here's what they actually do in plain terms, without the technical jargon:

• Stability: how consistent the voice sounds across the whole clip. Lower means more expressive and varied, which can sound more natural but occasionally unpredictable. Higher means flatter and more consistent, which can start to sound a bit robotic if pushed too far.
• Similarity: how closely the output sticks to the original voice sample it was trained on (relevant mainly for cloned voices). Higher keeps it closer to the source voice, lower allows more creative drift.
• Style exaggeration: how much emotional or stylistic character comes through. Useful for voices meant to sound animated or dramatic, usually best left low for straightforward narration or business content.

If your first generation sounds slightly robotic, try lowering stability a little first, that's the setting most likely to fix a flat-sounding result. My detailed guide to these settings goes deeper on recommended starting points by use case.

Saving and exporting

Once you're happy with a generated clip, there's a download button that saves it as an audio file (usually MP3) to your device. If you're working inside the Studio tool for a longer project, it saves your project automatically as you go, and you can export the finished audio from there once the whole piece is ready. Keep in mind that anything generated on the free plan isn't licensed for commercial use, you'll need at least the Starter paid plan ($5 a month on annual billing) before using a file in a business video, course, or anything you're publishing commercially.

Using Studio for a longer, multi-section project

If you're producing something longer than a single clip, a full training module, a multi-chapter audiobook, or a long-form video script, the Studio tool is built for that rather than the plain Text to Speech page. Studio lets you break a long piece of content into sections, each with its own voice and settings if needed, and manage the whole project as one unit rather than generating and stitching together individual files yourself. The free plan includes 3 Studio projects, enough to learn the workflow, paid plans raise that cap considerably (20 projects on Starter, more on higher tiers). If your first real project is going to be longer than a couple of minutes of finished audio, start in Studio from the beginning rather than building it in the plain Text to Speech tool and trying to move it over later. Long files can drift in tone, so read why AI voice changes tone.

Understanding your credit usage dashboard

Your account includes a usage dashboard showing how many credits you've used against your plan's monthly allowance. Check this periodically, especially in your first month, so you get a feel for how much a typical piece of your content actually costs in credits before you're surprised by hitting a limit mid-project. If you're close to your limit partway through the month, it's usually cheaper to finish out the month carefully (prioritizing final generations over experimental ones) than to upgrade a tier for just a few remaining days before your plan resets anyway.

A few practical tips for your first week

• Test with your real script, not the demo text: preview text in the tool is often short and doesn't reveal how a voice handles your actual, longer content.
• Save a voice and settings combination that works: once you land on something good, note the voice name and slider positions so you can reproduce the same sound later without guessing again.
• Watch your credit usage early on: the free plan's 10,000 credits go quickly once you start generating real content, keep an eye on your usage dashboard so you're not surprised.
• Don't over-adjust settings on your first try: small changes to one slider at a time make it easier to tell what's actually improving the result versus what's just different.

What to do if your first result genuinely isn't good enough

If you've tried a couple of voices and adjusted settings and it still doesn't sound right, step back and check a few things before assuming the tool can't do what you need. First, confirm the script itself reads naturally aloud, a stiff or awkwardly punctuated script will sound stiff no matter how good the voice or settings are. Second, try a genuinely different voice rather than tweaking settings further on one that isn't landing, sometimes the voice itself, not the settings, is the mismatch. Third, if it's a pronunciation issue with a specific word (a name, a technical term, an acronym), try spelling it out phonetically in the script rather than relying on the engine to guess correctly. Most "this doesn't sound right" problems trace back to one of these three, not a fundamental limitation of the tool. A cloned voice that sounds flat is a separate problem, so see how to fix a tired or slow voice.

How does Blur Studio approach this?

If you get comfortable with ElevenLabs and later want a voice tied directly into your own product, a phone agent, an in-app narrator, something more than generating files one at a time in a browser, that's a different kind of build. Tell me what you're trying to do and I'll give you an honest read on whether that's actually where you're headed.