ElevenLabs Prompts (2026): 100+ Voice Design, v3 Audio Tags & Dialogue Scripts

Introduction to ElevenLabs Prompting

ElevenLabs is a leading text-to-speech and voice cloning platform that allows creators, developers, and enterprises to generate lifelike voiceovers in multiple languages. Unlike generic text-to-speech engines, ElevenLabs uses advanced AI to produce natural, expressive voices that sound human-like with emotional nuance. The platform is increasingly used for YouTube videos, podcasts, e-learning, audiobooks, customer service, and commercial productions.

Effective prompting on ElevenLabs involves two distinct processes: voice design (creating or customizing a voice character) and voiceover scripting (writing and optimizing the text that the voice will narrate). Mastering both will help you produce professional-quality audio content at scale.

How ElevenLabs Prompts Work

ElevenLabs uses prompts in two key ways: voice design prompts describe the characteristics of a custom voice (age, gender, accent, energy level, emotional tone), while voiceover text prompts are the actual scripts you want the voice to narrate. Voice design prompts help you create a unique voice character that matches your brand or content style. Voiceover script prompts are optimized for natural delivery, pacing, emotional expression, and listener engagement.

The platform supports real-time voice cloning, where you can upload a short voice sample (your own or a licensed voice) to create a personalized voice model. This opens up possibilities for brand voice consistency, multilingual narration, and character-specific voices for storytelling and entertainment projects.

Voice Design Prompt Template

When creating or describing a custom voice in ElevenLabs, use this structure to get the most accurate result:

"A [age-range] [gender] voice with a [accent] accent, 
[overall tone: calm/energetic/authoritative/warm], 
[pacing: slow/medium/fast], 
[emotional feel: warm, playful, serious, mysterious], 
suitable for [use case], 
with [extra traits: slight rasp, subtle smile in voice, authoritative confidence, etc.]."

Example: "A mid-30s female voice with a neutral American accent, calm and confident, medium pacing, warm emotional feel, suitable for SaaS tutorials, with a professional but approachable tone."

35+ Voice Design Prompt Ideas

40+ Voiceover Script Ideas

ElevenLabs v3 Audio Tags: Complete Reference (2026)

Audio tags are inline directives in square brackets that tell ElevenLabs v3 how to perform a specific line. They are the single biggest leap in output expressiveness since v2 launched, and they dramatically reduce the number of takes you need for emotionally varied content. Tags work inline — you embed them in the script at the moment the vocal action should happen.

Example: “I can't believe you did that. [laughs] That's incredible.”

Emotional tags

Non-verbal sounds

Worked examples — audio tag scripts

Hey everyone, welcome back to the show. [pause] Today we're talking about something that completely changed how I think about productivity. [chuckles] I know, I know — another productivity episode. But stay with me, because this one's actually different.

She opened the letter slowly. [inhales] Her eyes scanned the first line, then the second. [pause] [whispering] No. [pause] [louder, shaken] This can't be right.

[confident] For ten years, we've been building the tools creators actually need. [warm] Not the ones that look good in demos — the ones that ship. [excited] And today, we're launching something we've been working on for two years. [pause] We can't wait to show you.

[sarcastic] Oh, fantastic. Another meeting that could have been an email. [sighs] Let me pull up the agenda. [deadpan] I'm sure it'll be thrilling.

Audio tag usage rules

• One or two per paragraph — overuse makes output unstable and forces the model to exaggerate.
• Tags persist until the next tag — [whispering] applies to everything after it until you explicitly reset with another tag or a sentence break.
• Combine tags sparingly — [nervous, whispering] works; stacking four tags rarely does.
• Stability matters — drop stability to 0.3-0.5 for expressive tags; tags at 0.8+ stability get smoothed out.
• v3 and Multilingual v2 only — legacy models (v1, Monolingual v1) will read the tags as literal text.

Multi-Speaker Dialogue Prompts (12)

ElevenLabs Studio supports multi-speaker dialogue natively — assign different voices per line and the system handles voice switching. For API-only workflows, generate each speaker separately and concatenate. Use distinct pacing and audio tags to reinforce character separation.

HOST: So you've spent the last three years building this. [pause] When did you know it was going to work?
GUEST: [chuckles] Honestly? The moment a beta user emailed me at 2 a.m. asking where the invoice was. [laughs] That's when I knew we had something.
HOST: [warm] That's such a good sign.

MAYA: The electric bill came in again.
JAMES: [sighs] How bad?
MAYA: [amused] Bad. But guess what the solar quote says?
JAMES: [curious] Tell me.
MAYA: [excited] We break even in four years.

DETECTIVE: Where were you between nine and ten last night?
SUSPECT: [nervous] At the gallery. I told you that.
WITNESS: [firmly] That's a lie. I saw him leave at eight forty-five.

Additional dialogue prompt starters:

• Two-person onboarding walkthrough — friendly host + curious new user asking clarifying questions
• Customer service training scenario — frustrated caller + empathetic agent de-escalation
• Classroom explainer — teacher + student asking follow-ups, teacher scaffolding answers
• Courtroom cross-examination — lawyer pressing, witness evasive then collapsing
• Therapy role-play — therapist reflecting back, client working through a realization
• Sci-fi bridge scene — captain issuing orders, helm officer confirming, AI warning
• Morning-show banter — two co-hosts ribbing each other about a news story
• Documentary interview — calm narrator voice-over + recorded participant anecdote
• Kids' audiobook — parrot character (quick, squawky) + grandparent (slow, warm) telling a tale

Voice Cloning Description Prompts (10)

After cloning a voice from an uploaded audio sample, ElevenLabs lets you attach a text description that refines how the model interprets the clone. Descriptions have the biggest impact when the sample is short (under two minutes) or only covers a narrow emotional range. Use these templates to expand what a cloned voice can do.

• “This voice is warm, measured, and conversational — leans into light self-deprecating humor, never shouts, naturally slows down on important points.”
• “Professional narration voice for technical tutorials — clear enunciation, confident but not authoritarian, treats the listener as intelligent, avoids salesy enthusiasm.”
• “Storytelling voice for fiction — pitches up slightly for children's dialogue, lowers and slows for suspense, uses breath and pauses for emphasis.”
• “Brand-spokesperson voice — energetic without being hyped, articulate, pauses after key claims, sounds like someone you'd trust with a recommendation.”
• “Meditation and wellness voice — slow pacing (around 110 words per minute), soft volume, long pauses, gentle downward inflection at sentence ends.”
• “News anchor voice — even pacing, authoritative without being severe, clean breaks between sentences, neutral on emotionally charged material.”
• “Audiobook character voice for a curmudgeonly old wizard — low pitch, gravelly, slow, with dry sarcastic asides.”
• “Corporate training voice — clear, approachable, slightly more formal than conversational but never stiff, natural hesitations at thinking-points.”
• “Fitness coach voice — energetic, encouraging, well-paced, rises in intensity for instructions and softens for water-break reminders.”
• “Documentary narrator voice — measured gravitas, cinematic pauses, descends in pitch for historical weight, rises slightly for revelatory moments.”

Pronunciation & SSML Control

ElevenLabs doesn't support full W3C SSML, but it does respect several pronunciation controls for brand names, technical terms, and abbreviations:

• Phonetic respelling — for a hard-to-pronounce brand, rewrite it the way it sounds: “Huawei” → “Hwah-way”, “Nguyen” → “Win”.
• Acronym spacing — for letter-by-letter reads, add spaces: “A P I” reads as “A. P. I.”, not “AP-pee”.
• Number expansion — write numbers how you want them spoken: “2025” can read as “twenty twenty-five” or “two thousand and twenty-five” depending on context.
• Capitalization for emphasis — WRITING A WORD in all caps often produces slight vocal emphasis, though the effect varies by voice.
• Em dashes for rhythm — em dash breaks (—) produce a distinctive dramatic pause different from a comma or period.

For reliable brand-name pronunciation, generate a short test clip with your phonetic respelling before committing to a long script.

Best Practices for ElevenLabs Voice Prompts

• Choose the right voice first: Voice selection has more impact than micro-tweaks. Select a voice that matches your content's tone and audience expectations before fine-tuning delivery.
• Write scripts for the ear: Use short sentences, clear punctuation, natural phrasing, and conversational language. Read your script aloud to catch awkward phrasing.
• Use longer, expressive prompts: Provide at least 250+ characters of context. ElevenLabs' AI performs better with detailed voice descriptions that include emotional cues and use-case context.
• Control emotion with text formatting: Use punctuation strategically—exclamation marks for excitement, ellipses for thoughtfulness, commas for pauses. Some platforms support emphasis tags.
• Iterate and segment: Break long scripts into sections (intro, body, CTA) and generate voiceover for each part separately. This gives you more control and flexibility in final editing.
• Test pronunciation: Check how ElevenLabs pronounces technical terms, brand names, and non-standard words. Adjust spelling or add pronunciation guides if needed.
• Match pacing to content: Fast-paced scripts work for ads and promotional content. Slower pacing suits educational and meditative content.

FAQ: ElevenLabs Prompting