How to create an AI voice-over with ElevenLabs

Disclosure: Some links may be affiliate links; we may earn a commission at no extra cost to you. How this works.

A voice-over used to mean booking a studio or a freelancer. Text-to-speech tools like ElevenLabs let you produce narration for a product demo, a tutorial or a promo video from a script. The quality depends heavily on three things you control: the script, the voice and the settings.

This guide follows ElevenLabs' own documentation for the Text to Speech tool. The product changes often, so check the current docs and pricing page when you start. We checked them on 2 November 2026.

Step by step

1. Write the script for the ear

   Text meant for reading and text meant for listening are different. Use short sentences. Read the script aloud once and mark where you run out of breath.

   ElevenLabs advises writing numbers, symbols and abbreviations out in words, because digits and symbols can be read differently across languages. Write 'one hundred euros' rather than '€100'. Proofread too: the documentation says the AI will not correct a misspelled word and reads it as written.

2. Choose the voice first

   According to ElevenLabs' documentation, voice selection has the biggest influence on the result, followed by model selection and then the model settings. Spend your time there. Listen to several voices with your own text, not just the sample sentence.

   If you are not generating English, choose a voice that was made for your language. The documentation notes that any voice can technically speak other languages but keeps its original accent. The voice library lets you filter by language.

3. Pick a model and test two

   ElevenLabs offers several models, including Eleven v4, Eleven v3, Multilingual v2 and the low-latency Flash models. They differ in language coverage, character limits and expressiveness. On the website, Eleven v4 accepts up to 10,000 characters per generation and other models up to 5,000, so longer scripts need to be split or produced in Studio. For English-only content, the documentation recommends an English-only model. For other languages you need a multilingual one. Generate the same paragraph with two models and keep the one that sounds more like what you want.

4. Start with the common settings

   The documentation names stability around 50 and similarity around 75 as the most common starting point, with the style setting kept at 0. Lower stability gives a wider emotional range and more variation; higher stability sounds steadier and can become monotone. Some models do not offer every slider, so use the ones you see.

   The tool is non-deterministic, which means the same text can come out slightly different each time. That is normal. Generate, listen, adjust.

5. Generate and use the free regeneration

   Click Generate. If a line comes out wrong, ElevenLabs says you can regenerate the same text up to two times at no extra charge on the website, provided the text, voice and model stay the same, the first generation is under two hours old and you have not refreshed the page. If you change the text, you pay credits again, so fix the script before you generate.

   CreditsOn the pricing page we checked on 2 November 2026, the Free plan was listed at $0 per month with 10,000 credits per month. Credits are shared across all features, and the pricing FAQ lists text to speech at 1 credit per character, so a long script uses a lot. Prices and allowances change.

6. Download and check the file

   Use the download button after generating, or open your history later. The documentation lists MP3 (128 kbps) and WAV there, and MP3 at higher bitrates, M4A and FLAC under Advanced. Listen to the full file with headphones for mispronounced names, odd pauses and clipped endings. Generate that sentence again if needed.

7. Check the licence before you publish

   On the pricing page, the Commercial License is listed starting with the Starter plan, described as 'everything in Free, plus' it. If the voice-over will appear in a client project or an advert, do not assume the free plan covers it: read the current terms and plan details first, and keep a record of which plan you were on when you generated the audio.

A note on cloning voices

ElevenLabs offers instant and professional voice cloning. Only clone your own voice, or a voice whose owner has clearly agreed. Cloning someone else without permission can cause legal problems and damage trust. If you use a synthetic voice in client work, consider telling the client.

Where to get it

The free plan is enough to try this guide on a short script. For commercial work, read the current pricing page first. Open ElevenLabs: ElevenLabs (affiliate link, see the note above).

Keep going

1. 01How to make a short promo video with CapCutVideoTurn a few phone clips into a short promo video with CapCut: write the script, cut the clips, add an AI voice and auto captions, then export for social.
2. 02How to write client emails faster with AIEmailA practical six-step workflow for freelancers and small businesses: write a voice note, use a five-part prompt, check facts and save templates to reuse.