ElevenLabs: the complete AI voice guide (2026)

Why ElevenLabs leads AI text-to-speech

In 2026, ElevenLabs is still the reference for natural AI voices. Where older text-to-speech sounded robotic, its Multilingual v2 model reproduces the breathing, pauses and intonation of a real voice. It speaks 29 languages with a natural accent, so the same video can be adapted for several markets without a recording studio.

This guide shows you how to get a truly professional voice-over: pick the voice, write the script, tune the settings and avoid the mistakes that give away a synthetic voice. The examples use the iafetch Voice module, which includes ElevenLabs Multilingual v2.

The model used: ElevenLabs Multilingual v2

iafetch uses Multilingual v2, ElevenLabs' most versatile model. It favours quality and naturalness over raw speed: a voice-over of a few sentences is ready in seconds. It covers English, French, Spanish, German, Italian, Portuguese, Arabic, Hindi, Japanese, Korean and Chinese, among others.

One key point: the voice's language follows the text's language. The same voice can read a script in English and then in Spanish while keeping its tone, which gives a brand a consistent sound across all its markets.

The 4 voices available on iafetch

Voice	Tone	Best for
Adam	Deep, calm male voice	Documentaries, podcasts, narration
Sarah	Clear, professional female voice	Tutorials, e-learning, presentations
Liam	Young, energetic male voice	YouTube, Shorts, social media
Charlotte	Warm, expressive female voice	Ads, storytelling, lifestyle brands

Tip: keep the same voice across a whole series. Your audience quickly links it to your brand, like a jingle.

Pick the right context

The Voice module offers four contexts that shape how the text is read:

Marketing: energetic, persuasive tone for ads and product presentations
Storytelling: calmer pace and clear pauses to tell a story
E-learning: very clear, steady delivery to explain and teach
Expressive: stronger emotions for creative content and characters

Write a script that works for an AI voice

The quality of a voice-over depends as much on the text as on the model. An AI reads exactly what you write: punctuation is its only guide to rhythm.

Punctuate to breathe: a comma creates a short pause, a full stop a clear one, an ellipsis (…) a dramatic one
Vary sentence length: a short sentence after a long one adds impact
Spell out numbers when the reading is ambiguous: "one hundred and twenty euros" rather than "€120"
Avoid abbreviations: "for example" rather than "e.g."
Read your text out loud: if you stumble on a sentence, so will the AI
Write directly in the target language: a text written in the language always sounds more natural than a machine translation

Example for a 20-second ad: "Got an idea? In a few minutes, it becomes an image, a video… and a voice. Try it free, no credit card needed."

The two settings that make all the difference

Stability

Stability controls how even the voice is. Low, the voice is more expressive and emotional but can vary from one sentence to the next. High, it is more consistent but can become monotonous. iafetch sets it to 0.5 by default, a good balance for most uses.

Ads, storytelling: 0.35 to 0.5
Narration, podcasts: 0.5 to 0.6
E-learning, technical content: 0.6 to 0.75

Similarity

Similarity sets how closely the voice sticks to its original tone. The default, 0.75, almost always works. Raise it for a very recognisable voice; lower it slightly if you hear artefacts on hissing sounds.

How much does a voice-over cost?

On iafetch, a voice-over costs 2 credits up to 300 characters (about 20 seconds of speech), then 1 credit per extra 150 characters. With the Creator plan (€19 for 100 credits), a short voice-over comes to about €0.40. By comparison, a studio recording by a voice actor usually costs tens to hundreds of euros.

Use cases and recommended settings

Use	Voice	Context	Stability
Podcast or documentary	Adam	Storytelling	0.55
Social media ad	Charlotte	Marketing	0.45
Online course	Sarah	E-learning	0.7
YouTube Short or TikTok	Liam	Marketing	0.5

Mistakes that give away an AI voice

One long block of text: split it into short paragraphs and generate them separately to control the rhythm
Unexplained acronyms: the AI may spell them out or read them as a word; write them the way you want to hear them
Stability too low on a long text: the voice becomes uneven; raise it for narration
Music too loud in the edit: keep the voice 6 to 10 dB above the music

Combine the voice with the other modules

A voice is most useful alongside an image or a video. On iafetch, you can generate a clip with Kling or Hailuo, write the script with Claude in the Copy module, then create the voice-over, all with the same credit balance. It is the fastest way to produce a complete ad or a Short.

Frequently asked questions

Are AI voices allowed on YouTube and Spotify?

Yes. The major platforms accept AI-generated audio. Some ask you to label synthetic content, YouTube in particular for realistic content.

Can I use the voice-over in an ad?

Yes, the generated audio belongs to you and can be used commercially, within the models' terms of use.

What is the difference with a direct ElevenLabs subscription?

With ElevenLabs alone, you only generate audio. With iafetch, the ElevenLabs voice comes with images, videos, copy and the prompt generator, in a single subscription.