Which ElevenLabs audio tags work in Eleven v4
Updated Invalid Date
The picks
Press play to hear each voice read this guide's sample line.
These picks compare my fifteen voices. I own them, and ElevenLabs may pay me when you use them.
Felix·Character reactions
The catalog favors animation and large performance swings. It is a poor match for restrained instructions.
Drift·Quiet delivery
Close-mic atmosphere suits a soft scene. Keep the music sparse so the words survive.
Titan Epic·A restrained reveal
Fiction and trailer weight can build into a reveal. Let the words carry the stakes instead of tagging every clause.
ElevenLabs audio tags are short directions in square brackets, such as [whispers], [sighs] or [laughs], typed into the script at the exact spot where the delivery should change. Three models read them. Those are Eleven v4, Eleven v4 Turbo and Eleven v3, and the same tags work across all of them, though ElevenLabs recommends v4. Other models are not on ElevenLabs’ list of tag-capable models, so glance at the model selector before you count on a tag doing anything.
Where do audio tags go in a script?
An audio tag goes immediately before the words it should change. ElevenLabs’ v4 guidance says to place the tag before the exact clause you want, add a fresh tag where the mood shifts, and keep to one tag per sentence segment so the model isn’t juggling two instructions. Tags are free-form. A short phrase works as well as a single word, and ElevenLabs’ own v4 examples include [lower, thoughtful] and [quietly curious].
[quietly] I checked the lock twice before bed. [sighs] It was open again this morning.
[flat, tired] Third time this week.
An opening emotion also carries. ElevenLabs says that when a sentence starts with an emotion, v4 holds that tone across the line. You rarely need a tag on every sentence.
Which ElevenLabs audio tags are documented?
| Group | Examples from ElevenLabs’ docs | How I’d use them |
|---|---|---|
| Emotion | [curious], [excited], [sarcastic], [mischievously], [crying] |
Setting the mood of a line before it starts |
| Delivery | [whispers], [shouts] |
Changing volume and closeness |
| Reactions | [laughs], [sighs], [exhales], [clears throat], [snorts] |
Human sounds between the words |
| Sound effects | [applause], [gunshot], [explosion], [clapping] |
Rarely. I add effects in the edit, where I control the level. |
| Experimental | [strong X accent], [sings] |
Tests only. ElevenLabs flags these as less consistent across voices. |
Why does ElevenLabs ignore or misread an audio tag?
Five causes cover nearly every miss I’ve had.
The voice was never recorded that way. ElevenLabs says a tag lands more readily when the delivery already exists in the voice’s training data. v4 can follow a tag the voice wasn’t trained on, with weaker results. A close-mic, restrained voice will whisper beautifully and shout badly. No tag changes that.
The tag gets read as a sound effect. Because v4 can generate sound effects, a tag sometimes comes out as a noise when you meant a delivery, or the reverse. ElevenLabs’ fix is to describe the voice itself. A tag like [low, gravelly voice] can’t be mistaken for a sound, so write it that way.
Two moods at once. ElevenLabs warns that contrasting emotions in one direction make the output less accurate. If a character is angry and amused, choose the one the line is about. The words can carry the other.
The passage is too short. ElevenLabs says v4 works best on paragraphs, and one tagged sentence gives the model nothing to build from. Generate the line inside its paragraph and cut it out later.
The tag doesn’t suit the voice. ElevenLabs’ guidance is that a serious or professional voice may not suit playful tags. Put [mischievously] on a steady support voice and the two pull in opposite directions.
When a line still won’t land, ElevenLabs suggests trying a different tag before you touch the text. I add one more step. Try the line with no tag at all, because a good voice reading a well-built sentence often needs no direction whatsoever.
How do I add a pause on Eleven v4?
Use punctuation. <break> tags don’t work on v3 or v4, and ElevenLabs points you to audio tags, ellipses and text structure for pauses. On v4 the toolkit looks like this.
- An ellipsis slows a line down or lets it trail.
- A dash cuts the speaker off mid-sentence.
- Capitals on one word add emphasis.
- A paragraph break gives the longest beat.
A pause direction in brackets is free-form like any other tag, and ElevenLabs uses one in a v4 audiobook example. Punctuation is more predictable, though. I start there and only reach for a bracket when a full stop is too short.
How many audio tags should one script have?
For narration, one tag every few sentences is plenty. Dialogue and character work can go denser, since the performance is the point of those. A narration script with a tag on every line sounds like an actor who doesn’t trust the writer. If you’re tagging every sentence, the script probably needs a rewrite more than the voice needs direction.
My voiceover script guide covers the punctuation side. Eleven v4 vs v3 covers what changed between the two tag-capable models.
Which of my voices respond best to audio tags?
Felix is the voice where tags pay off most. It’s a character read with big swings, made for animation and story beats that want the performance out front, and [excited] or [laughs] pushes it where it was already heading. Restraint is the problem. Instructions and news are too much for Felix, tags or no tags. See Felix for characters and animation.
Drift is the opposite case. Close-mic and restrained, it takes [whispers] and [exhales] along its grain, and [shouts] goes against it. It also disappears under loud music and sounds odd on marketing copy, so keep the mix sparse and the script gentle. Hear it on Drift for ASMR and relaxation.
I’d start casting with Felix for reactions, Drift for quiet delivery, or Titan Epic for a reveal. The samples below use punctuation alone, so you can hear a baseline before trying directions on your own script.
Sources
Related guides
Other jobs where I use the same voices.