03 Expression tags

A directing vocabulary of 62 tags. And a door to natural language.

ElevenLabs v3 reads short directives inside square brackets and uses them to shape every line. flexVox knows the full vocabulary — and lets you stack, layer, and invent custom tags the same way a director would whisper notes from the booth.

  • 62 tags
  • 4 categories
  • Stackable
  • Natural-language descriptors
Expression in context
ALEX: [angry][laughing] I cannot believe you actually said that.
tag stacking · two tags compose into a complex delivery
JORDAN: [trying to sound brave] It's just a small thunderstorm.
custom descriptor · v3 interprets natural language inside the brackets
ALEX: [whispers]Did you hear that? [gasps]It's getting closer.
inline placement · tags work mid-sentence, not just at the start
Emotion Delivery Non-verbal Custom

The whole vocabulary, in one room.

Below is the full v3 vocabulary flexVox ships with — sixty-two tags across four categories. You don't need to memorize any of them. The composer surfaces them in the keyboard toolbar's expression menu, and the AI script writer places them automatically when enabled.

Emotion & Tone 17

  • [happy]
  • [sad]
  • [angry]
  • [nervous]
  • [calm]
  • [excited]
  • [gentle]
  • [frustrated]
  • [warmly]
  • [curious]
  • [mournful]
  • [tender]
  • [wistful]
  • [menacing]
  • [sarcastic]
  • [flirty]
  • [apologetic]

Voice Delivery 14

  • [whispers]
  • [shouts]
  • [mumbles]
  • [softly]
  • [slowly]
  • [quickly]
  • [breathy]
  • [monotone]
  • [crisply]
  • [sing-song]
  • [sotto voce]
  • [projects]
  • [over-enunciates]
  • [falsetto]

Non-verbal Sounds 15

  • [laughs]
  • [sighs]
  • [gasps]
  • [coughs]
  • [sniffs]
  • [chuckles]
  • [groans]
  • [cries]
  • [pauses]
  • [clears throat]
  • [inhales]
  • [exhales]
  • [hums]
  • [tuts]
  • [scoffs]

Inline SFX 16

  • [gunshot]
  • [knock]
  • [thunder]
  • [glass shatter]
  • [door slam]
  • [engine starts]
  • [gentle rain]
  • [tape hiss]
  • [static burst]
  • [film projector]
  • [phone ring]
  • [soft footsteps]
  • [kettle whistle]
  • [wind through trees]
  • [ice in glass]
  • [old typewriter]

Stack them. Layer them. Mean it.

A single tag is a single direction. Real performances are layered. Stack tags by writing them back-to-back — [angry][laughing] — and v3 blends them into a single delivery. The order matters: the dominant emotion goes first.

Tags can land anywhere in a line, not just at the start. Mid-sentence tags shape what comes after them: "It's quieter than I expected. [whispers] Maybe too quiet."

Tags are not stage directions for a human reader. They are real instructions to v3. Place them where you'd actually want a performer to change.

The hidden door: custom natural-language tags.

v3 doesn't limit you to the catalog. Anything in square brackets that reads as a direction — [trying to sound brave], [the way you'd tell a child], [holding back tears] — gets interpreted as a performance note. The model handles ambiguous direction the way an actor would: with reasonable interpretation, not literal mimicry.

flexVox doesn't constrain custom tags. The AI writer is allowed to invent them. The composer's keyboard toolbar surfaces categorized tags for speed, but a free-text tag is always one keystroke away.

Enhance Expression — add tags to a script that didn't have them.

Most scripts arrive un-tagged. The Enhance Expression action runs every speech turn through Apple Intelligence on-device when available, or your configured AI provider (OpenAI or Claude) as a fallback. Surrounding dialogue is sent as context so the suggested tags are emotionally consistent with the scene's flow.

Enhancement is review-first. flexVox presents each suggestion on a review sheet — accept the ones that fit, reject the ones that don't — before anything is applied to the script. You can also strip tags from any turn the same way.

  • On deviceApple Intelligence preferred — no API call when available.
  • FallbackOpenAI or Claude, configured in Settings → AI Writing.
  • Context windowSurrounding turns travel with the request so tags stay coherent.
  • ReviewBulk enhancement always opens a review sheet. Single-turn enhancement applies inline.

→ Keep going