Blog / How to Generate Realistic Emotional Voices with ElevenLabs in Claude

How to Generate Realistic Emotional Voices with ElevenLabs in Claude

By Published September 3, 2026 audio
How to Generate Realistic Emotional Voices with ElevenLabs in Claude

ElevenLabs v3 can whisper, shout, laugh, and switch accent mid-line when you write audio tags into the script. With CreativeClaw you generate that speech inside Claude, ChatGPT, Codex, or OpenClaw — no ElevenLabs account, and no other TTS API.

This guide follows official ElevenLabs v3 practices, then gives you copyable prompts that match the playable examples in the voice and audio collection. Copy a prompt into an AI with CreativeClaw connected and ask it to generate with ElevenLabs v3.

If you need the broader TTS lineup (MiniMax, Dia, Kokoro), start with the text-to-speech in Claude guide. For a product overview of ElevenLabs v3 on CreativeClaw, see How to Use ElevenLabs in Claude.

Why use CreativeClaw for ElevenLabs v3?

CreativeClaw is the fastest and simplest way to use ElevenLabs v3 in Claude. Here's why:

  • No API keys needed - No accounts, no configuration files. Connect one URL and every model is available instantly.
  • No subscriptions - Pay only for what you generate. $10 = 1,000 credits. No monthly fees, credits never expire.
  • MCP Apps - Preview generated media directly in Claude's UI. See results inline without opening files or navigating to external URLs.
  • Expert skills built in - CreativeClaw knows how to get the best results from ElevenLabs v3. You don't need to be a prompt engineering expert - Claude handles the optimization.
  • Let Claude iterate - This is the real power. Claude generates, evaluates the result, refines the prompt, and regenerates - all in one conversation. Your AI agent becomes your creative director.
  • Run from anywhere - CreativeClaw is a remote MCP server. Use it from Claude Code, Claude Desktop, Claude Web, or OpenClaw - same results, same account, wherever you work.

Official v3 practices (from ElevenLabs)

ElevenLabs published best practices for Eleven v3 when the expressive speech model launched. The habits that actually change the take:

  1. Write a longer spoken prompt. Aim for 250+ characters of script when you want layered emotion — whispers that build into a shout, crowd reactions, pauses. Short one-liners work for a single gag; long tagged narration is how v3 shows range. The Lunar Colony Helios example is 827 characters of tagged speech.
  2. Pick a voice suited to the delivery. A croaky witch, a husky warrior, and a confidential French whisper are different instruments. Tags cannot rescue a mismatch.
  3. Use audio tags inline. Official examples include [sarcastic], [whispers], [excited], [strong French accent], [laughs], [sighs], and [shouts].
  4. Tags are not universal. Match the voice to the emotion first, then tag the line. A [menacing laugh] on a calm narrator will not sound like a witch cackle.

The original audio-tag list groups tags into three jobs:

  • Emotions[sad] [angry] [happily]
  • Delivery[whispers] [shouts]
  • Non-verbal reactions[laughs] [clears throat] [sighs]

A later conversational-tag note adds the same idea for dialogue: direct the voice to laugh, whisper, act sarcastic, or sound curious. For multi-speaker sarcasm and interruptions, play Planet Zorgon diplomacy.

How to generate emotional voices in Claude

Step 1 — Connect CreativeClaw. Follow the setup guide. One MCP URL (https://app.creativeclaw.co/mcp), under a minute. New accounts start with 100 free credits. No card.

Step 2 — Copy a prompt into Claude (or ChatGPT, Codex, OpenClaw) with CreativeClaw connected. Name ElevenLabs v3. For Voice Design clips, ask the assistant to design a voice from the Voice Prompt, then speak only the Script Prompt.

Step 3 — Generate through CreativeClaw, not another speech API. Iterate on tags and voice if the first take is flat.

Setup by client

Claude Code — Install the CreativeClaw plugin for the full experience with skills and optimized prompts. See the Claude Code setup.

Claude apps — Add the CreativeClaw connector once, then use it on the web, desktop, or mobile. See the Claude apps setup.

ChatGPT and Codex — Enable the Creative Claw app in ChatGPT or add the hosted MCP in Codex. See the ChatGPT and Codex setup.

Grok Bot — Install Creative Claw via the Cursor Directory through Settings → Plugins, ask Grok Bot to add it, or manually add the MCP URL. See the Grok Bot installation guide.

Hermes Agent — Use the one-click MCP link or two CLI commands. See the Hermes installation guide.

OpenClaw — Add the hosted Streamable HTTP MCP and complete OAuth. See the OpenClaw installation guide.

Copyable prompts (different voices)

Each block below matches a hosted example. Copy it into your CreativeClaw-connected AI. The playable clip on the example page is the original creator media — CreativeClaw did not generate those files.

1. Long tagged narration — Lunar Colony Helios

Play the clip: Lunar Colony Helios. Original tagged script by franco (@francobit). Whisper, shout, laugh, and crowd reactions in one 250+ character pass.

Use CreativeClaw to generate this. If it is not connected, add the remote MCP https://app.creativeclaw.co/mcp. Generate with ElevenLabs v3 through CreativeClaw, not another image/video/TTS/music API.

[ominous] In a universe where silence swallows starlight… one signal remains. [audience ooos] [radio static] “Mayday, mayday—this is the Lunar Colony Helios…” [excited boisterous] Tonight, humanity answers. [crowd gasp] Engines ignite, [crowd gasps louder]steel hearts thunder [crowd gasps again]. [whispers] But heroes aren’t forged in daylight—they’re whispered into being inside the quiet between heartbeats. [heartbeat sound] [slow crescendo] This voice brings that whisper… [shouts] to a ROAR! [rocket launch] [fast] Precision pauses, [very fast] gravity-defying crescendos, [the fastest] laughter that echoes across vacuum[laughs]—every beat sculpted by text alone. [calm and slow] Because the greatest rescue mission… starts with a single, beautifully-spoken line. [applause] Coming soon—*in voices you’ll never forget*.

About 17 credits (827 characters at 20 per 1,000).

2. Voice design + laugh tag — witch

Play the clip: Witch menacing laugh. Original Voice Prompt and Script Prompt by ElevenLabs.

Use CreativeClaw to generate this. If it is not connected, add the remote MCP https://app.creativeclaw.co/mcp. Generate with ElevenLabs v3 through CreativeClaw, not another image/video/TTS/music API.

Design a voice matching the Voice Prompt, then speak the Script Prompt.

Voice Prompt:
A scary old and haggard witch who is sneaky and menacing. She has a croaky, harsh, shrill, high-pitch voice that cackles.

Script Prompt:
[menacing laugh] This could be the final ingredient!

About 6 credits (277 characters at 20 per 1,000).

3. Accent + whisper — French confidential

Play the clip: French confidential whisper. Original by ElevenLabs. Pair a thick-accent Voice Prompt with an inline [slight French accent] tag.

Use CreativeClaw to generate this. If it is not connected, add the remote MCP https://app.creativeclaw.co/mcp. Generate with ElevenLabs v3 through CreativeClaw, not another image/video/TTS/music API.

Design a voice matching the Voice Prompt, then speak the Script Prompt.

Voice Prompt:
A low, whispery and assertive female voice with a thick French accent. Cool, composed and seductive, with a hint of mystery.

Script Prompt:
[slight French accent] This… is strictly confidential.

About 6 credits (282 characters at 20 per 1,000).

4. Excited then mischievous — Tuxy

Play the clip: Tuxy backyard kitten. Original tagged script by Gadgetify (@Gdgtify). Two tags, one voice, a tone flip mid-sentence.

Use CreativeClaw to generate this. If it is not connected, add the remote MCP https://app.creativeclaw.co/mcp. Generate with ElevenLabs v3 through CreativeClaw, not another image/video/TTS/music API.

[excited]isn't she cute? This is Tuxy. [mischievously]My favorite backyard kitten. She has grown into a gorgeous cat. She now helps me test my gadgets. She specially likes my magnetic toys

About 4 credits (188 characters at 20 per 1,000).

Two more playable voices in the same collection: Japanese husky warrior ([dramatic] on a whispery Japanese-accent Voice Prompt) and Planet Zorgon diplomacy (multi-speaker [sarcastic] / [whispers] / [alien voice]).

Pricing

Credits come from CreativeClaw’s published table — ElevenLabs v3 is 20 credits per 1,000 characters (a 500-character message is 10 credits).

ElevenLabs v320 credits / 1,000 chars
$101,000 credits
Free100 credits on signup
SubscriptionNone. Credits do not expire.

A Helios-length script is about 17 credits. The short witch, warrior, French, and Tuxy lines are about 4–6 credits each if you send the full copyable prompt.

See pricing for the live table.

FAQ

Should I use ElevenLabs Music v1 for these voices? No. Music v1 is for generated music beds. Named speech with audio tags is ElevenLabs v3.

Can I clone a celebrity? Do not. Only clone voices you have the rights to use. The examples here are original creator clips and fictional characters.

Where do I paste the prompt? Into Claude, ChatGPT, Codex, or OpenClaw with CreativeClaw connected. The example pages prefix every copy button the same way.

Browse every tagged-speech and music example in voice and audio, or go back to ElevenLabs in Claude and the TTS guide.

Ready to try it?

Connect CreativeClaw to Claude in under a minute.

Get Started

Continue the workflow