What is ElevenLabs v4?

ElevenLabs v4 is the fourth-generation text to speech model from ElevenLabs. It reads up to 5,000 characters per request in a preset voice and follows inline audio tags such as [whispering], [excited] or [sighs].

How do audio tags work in ElevenLabs v4?

Write the tag in square brackets directly before the words it should color, for example "[whispering] Listen closely...". The tag itself is not spoken. Sound tags such as [laughs] or [sighs] become a vocal sound. Keep one mood per sentence.

What is the difference between ElevenLabs v4 and ElevenLabs v4 Turbo?

Both take the same text, tags, voices and settings. ElevenLabs v4 Turbo costs half as many credits and returned faster in our tests; ElevenLabs v4 is the quality tier for final reads.

How much does ElevenLabs v4 cost on Voor?

Credits scale with script length: about 28 credits per 1,000 characters on ElevenLabs v4 and about 14 on ElevenLabs v4 Turbo. A 100-character tagline costs 3 credits on ElevenLabs v4; a full 5,000-character request costs about 140. New accounts receive free welcome credits for testing.

Which languages does ElevenLabs v4 speak?

ElevenLabs v4 is multilingual. On this page we generated English, Spanish and Japanese with the same model. Leave the language code empty to detect it from the text, or set a code such as es or ja for short lines and mixed-language scripts.

How do I fix a mispronounced name in ElevenLabs v4?

Replace the word with its IPA pronunciation between forward slashes, for example "Doctor /ˈiːfə/ Brennan". Do not write the name and the IPA together: in our first test that made the voice say the name twice. For dates, prices and times, turn text normalization on.

How long can an ElevenLabs v4 script be?

Each request accepts up to 5,000 characters, which is roughly five minutes of English speech at the pace of our samples. Split longer scripts into scenes with fixed voice and settings.

Can ElevenLabs v4 clone my voice?

Not on this page. ElevenLabs v4 on Voor uses 21 preset voices such as Rachel, George, Sarah and Brian. Never use synthetic speech to impersonate someone without consent.

Should I switch from ElevenLabs v3 to ElevenLabs v4?

Start new projects on ElevenLabs v4 when you need tag-directed delivery, IPA pronunciation or the cheaper Turbo tier. Keep ElevenLabs v3 for a series that is already voiced with it.

ElevenLabs v4 Text to Speech With Audio Tags

Real render · voice George · 17.8 seconds · 9 credits

ElevenLabs v4 changes delivery where the tag says so

One ElevenLabs v4 generation, not four stitched clips. The score shows where each bracketed tag took over in the actual file.

[calm]0:00–0:04
At five in the morning the harbor is almost silent.
Even level, unhurried phrasing.
[whispering]0:04–0:08
Listen closely... the first engine is already turning over.
The quietest stretch, consonants still clear.
[excited]0:08–0:13
And there it goes: the whole fleet, one boat after another, heading for open water!
The loudest five seconds.
[sighs]0:13–0:18
Somebody still has to stay behind and mend the nets.
A breath, not the word.

Exact script · 286 characters · stability 0.5 · similarity 0.75

[calm] At five in the morning the harbor is almost silent. [whispering] Listen closely... the first engine is already turning over. [excited] And there it goes: the whole fleet, one boat after another, heading for open water! [sighs] Somebody still has to stay behind and mend the nets.

Same script, same voice, two tiers

ElevenLabs v4 or ElevenLabs v4 Turbo: half the credits, same timing

Identical script and settings on both tiers: the whisper starts at almost the same moment and the lengths differ by a tenth of a second, so Turbo is a faithful preview of pacing. Whether the quality tier is worth twice the credits is a listening call.

ElevenLabs v4

Length: 17.8 s
Whisper starts: 3.95 s
Returned in our run: about 11 s
This script: 9 credits
Per 1,000 characters: ≈ 28 credits

Final reads, trailers, narration that has to carry a scene.

ElevenLabs v4 Turbo

Length: 17.9 s
Whisper starts: 4.03 s
Returned in our run: about 7 s
This script: 5 credits
Per 1,000 characters: ≈ 14 credits

Drafts, retakes, long lists of short lines.

Names, numbers and IPA

Make ElevenLabs v4 say the name right the first time

A name the voice has never seen and a date in digits: this reminder handles both.

13.2 s · 241 characters · 7 credits

[friendly] Hi, this is a reminder from Riverside Dental. Your appointment with Doctor /ˈiːfə/¹ Brennan is confirmed for Tuesday, March 3rd at 2:30 PM². [reassuring] If you need to reschedule, just reply to this message. It only takes a minute.

IPA replaces the word. Aoife is an Irish name said “EE-fa”. A transcript of the output heard “Aoife”, not a guess built from the spelling.
Normalization on. The date and time are read as a receptionist would say them.
What failed first. Writing the name and then its IPA made ElevenLabs v4 say the name twice.

Voices and languages

Three ElevenLabs v4 voices, three jobs

Preset voices, none cloned from a real person. Each card lists the exact settings, so you can rerun it and change one thing.

SpanishSaturday market greeting

Model: ElevenLabs v4
Voice: Laura
Language code: es
Stability: 0.5
Cost: 6 credits

[cheerful] ¡Buenos días a todos! Bienvenidos al mercado del sábado. Hoy tenemos pan recién horneado, tomates del huerto y, como siempre, café caliente para los que madrugan. [laughs] Y sí, también hay churros.

Good morning, everyone! Welcome to the Saturday market. Fresh bread, garden tomatoes and hot coffee for early risers. And yes, there are churros too.

JapaneseBedtime wind-down

Model: ElevenLabs v4
Voice: Lily
Language code: ja
Stability: 0.5
Cost: 5 credits

[softly] 今夜は雨の音を聞きながら、ゆっくり目を閉じてみましょう。[whispering] 明日のことは、明日の自分にまかせて。おやすみなさい。

Tonight, listen to the rain and slowly close your eyes. Leave tomorrow to tomorrow's you. Good night.

EnglishGame quest-giver

Model: ElevenLabs v4 Turbo
Voice: Brian
Language code: auto
Stability: 0.4
Cost: 3 credits

[gruff] You want the mountain pass? [laughs] Nobody has crossed it since the snow came early. [serious] Take the lantern, keep left at the broken bridge, and do not, I mean it, do not stop to rest.

Credits count characters, so a Japanese or Chinese script costs the same per character as an English one.

Credits scale with characters

What an ElevenLabs v4 script costs

Every character counts, tags included. Spoken length assumes the pace of our English samples, about sixteen characters per second. New accounts start with 50 free welcome credits.

ScriptSpokenElevenLabs v4Turbo

100 charactersA tagline or button prompt≈ 6 s3 credits2 credits

1,000 charactersA one-minute explainer≈ 1 min28 credits14 credits

5,000 charactersThe per-request maximum≈ 5 min140 credits70 credits

Generator settings

ElevenLabs v4 settings on Voor

Stabilitydefault 0.5

Lower lets tags swing harder; higher keeps a long read even. The game line used 0.4.

Similarity boostdefault 0.75

How tightly the read holds the preset timbre. Very high can sound less natural.

Voice21 presets · default Rachel

George, Sarah, Laura, Lily and Brian are the voices heard on this page.

Language codeempty = detect

Set es, ja and so on for short or mixed-language lines.

Text normalizationauto · on · off

On spells out dates, prices and times; off keeps codes and serials literal.

Seedoptional

Best-effort repeatability, not a guarantee.

The shaded band is where our takes sounded natural; the marker is the default.

Writing for the model

Six rules for ElevenLabs v4 audio tags

Put the tag directly before the words it should color.
[whispering] Listen closely...
Give each sentence one mood. Two stacked emotions force a compromise.
[excited] And there it goes!
Let punctuation carry timing: ellipses and dashes pause, exclamation marks lift.
do not, I mean it, do not stop
Sound tags become sounds, not words. Check one take, then reuse the pattern.
[sighs] Somebody still has to stay behind
Replace a hard name with its IPA in slashes. Never write both.
Doctor /ˈiːfə/ Brennan
Draft in ElevenLabs v4 Turbo, then regenerate the approved script in ElevenLabs v4.
same text · same voice · same seed

Choosing a voice model