How do you make a voiceover with ElevenLabs?

ElevenLabs is a text-to-speech service: you give it written text and it returns a spoken recording in a synthetic voice. People use it to narrate explainer videos, audiobooks, podcasts, training modules and social media clips, and for accessibility, such as giving a voice to someone who has lost theirs. For a beginner, the useful thing to know is that a good voiceover depends on three choices: the script, the voice and a handful of settings.

This guide walks through a short narration from start to finish. It uses one of the platform's ready-made voices, which is the simplest and safest place to start.

Step 1: Write a script for the ear

A script for listening is not the same as one for reading. Keep sentences short, put one idea in each, and read it aloud yourself before you paste it in. If you stumble, a synthetic voice may too. Here is a 30-second script for a local library's video, made up for this example:

Looking for a quiet place to work this summer?
The Riverside Library is open every weekday until eight.
Bring your laptop. The Wi-Fi is free, and so are the study rooms.
Book a room online, or just ask at the front desk.
Riverside Library. Your place to think.

Step 2: Choose a voice

In ElevenLabs you can pick from a set of pre-made voices or browse the larger Voice Library. Listen to three or four with your own script in mind. For a library promotion you might want a warm, unhurried voice; for a product demo, something brisker.

Step 3: Set the sliders

The speech tool has four voice settings. ElevenLabs' documentation explains what each one does:

The 50 and 80 values come from the documentation, which says it is the combination users most often settle on. The company recommends leaving style exaggeration at 0. Speaker boost usually makes only a subtle difference, so test it once and keep it if you hear an improvement.

Step 4: Generate, compare, adjust

Generate the clip three times. Because the model is non-deterministic, meaning it does not produce exactly the same output twice, each take will differ a little even with identical settings. The sliders set how wide that variation can be.

Then adjust with a purpose:

• Too flat or robotic? Lower stability by 10 to 15 points and generate again. Lower values bring more emotion, though going too low can make the voice erratic or rushed.
• Too dramatic for a serious message? Raise stability. Higher values sound more even and usually need fewer takes to get a usable one.
• Odd noises or artifacts? If you are using a cloned voice from a poor recording, a very high similarity setting can copy the background noise too. Lower it a little.

Download the take you like best and place it in your video editor under the visuals. Listen once more at normal speed with the pictures before you publish.

What are the consent rules for voice cloning?

Voice cloning means creating a synthetic copy of a real person's voice from recordings. ElevenLabs offers two kinds, and the rules differ. For Professional Voice Cloning, which trains a custom model on your recordings, the documentation says plainly that you may only clone your own voice. Before training starts, you must pass a verification step, a voice captcha, in which you record your voice again and the system checks that it matches the samples you uploaded. If it does not, you have to contact support to be verified by hand.

Beyond that check, the company's terms forbid using the platform to impersonate or harm anyone. ElevenLabs also describes a "no-go voices" safeguard, designed to block clones of candidates in presidential and prime ministerial elections, starting with the US and the UK. It also offers a classifier that lets anyone upload audio and check whether it was generated with ElevenLabs, and it says breaking its rules can lead to warnings, removed voices, banned accounts and, in serious cases, reports to authorities.

The platform's rules are a floor, not the whole picture. If a colleague or client wants their voice cloned, get their clear permission in writing, explain where the voice will be used and for how long, and agree on what happens to the recordings afterwards. Tell your audience when a voice is synthetic, especially if it could be mistaken for a real person speaking. Laws on using someone's voice or likeness differ by country and state, so for commercial projects read the current ElevenLabs terms and check with a lawyer if you are unsure. When in doubt, use a pre-made voice: it avoids the consent question entirely.