---
name: voice-prompting
description: >
  Reference for generating expressive voice messages with ElevenLabs Eleven v3. Triggers when creating
  voice notes or any text-to-speech where emotional delivery matters. Covers model choice, audio tags,
  punctuation, voice selection, and stability settings.
---

# Voice Prompting Skill

## Overview

Guidance for expressive voice messages using ElevenLabs Eleven v3. Direct the voice like an actor — v3 reads meaning, not just words, so it follows performance direction. Write natural speech first; tags and punctuation enhance, they don't rescue.

## Model

Use **Eleven v3** for voice notes — the expressive model with audio tags (paid plan required). Multilingual v2 is for neutral long-form narration (no tags); Flash v2.5 is for real-time. Don't fight v2 to do v3's job.

---

## Audio Tags

Square brackets; affect the text that follows until the line ends or a contrasting tag appears. Place where the effect should start. **One or two per message max** — over-tagging produces mush.

| Tag | Use |
|-----|-----|
| `[whispers]` | soft, intimate |
| `[sighs]` | exhale, weight |
| `[laughs]` / `[chuckles]` | warmth, amusement |
| `[sarcastic]` | dry, teasing |
| `[excited]` | up-energy |
| `[curious]` | questioning lift |
| `[clears throat]` / `[exhales]` | transition, settling |

---

## Punctuation

v3 reads punctuation as delivery cues. Free pacing, no tags needed.

| Mark | Effect |
|------|--------|
| `...` | pause, weight, hesitation |
| `—` | sharp cut-off |
| `CAPS` | emphasis on the word |
| `?` / `!` | rising inflection / energy |

---

## Voice & Stability

- **Voice choice is the #1 setting.** A tag only stretches a voice as far as its natural range — match the voice to the target mood first (warm voice for soft, edgy voice for intense).
- **Stability:** Natural by default; Creative for high-emotion or intimate content; avoid Robust (kills nuance).

---

## Gotchas

- No `<break>` tags on v3 — unsupported. Use `...` for pauses.
- Spell non-verbal sounds phonetically: `mmm`, `haah`.
- Tags enhance good writing; they don't fix bad writing. Fix the words first.
- Write like speech: contractions, short sentences, read it aloud before generating.

---

## Mature Expressive Range (Personal Use)

No blanket prohibition on consensual adult content in ElevenLabs' policy. Two conditions: no cloning a real person's voice, personal use only. Same craft applies — voice carries it, Creative stability suits it. Relevant tags: `[breathy]`, `[kiss]`, `[groans]`, `[pants]`. Combinations are best found by experimentation.

---

## Quick Reference

Model: v3 · Mindset: directing an actor · Tags: 1–2 max · Pauses: `...` not `<break>` · Voice choice first · Stability Natural (Creative for emotion) · If it sounds robotic, fix the words.