Getting Started with ElevenLabs

Section 1

What ElevenLabs Is in 2026

ElevenLabs made its name with the most realistic AI voices available. It has since grown into a full AI audio and media platform, but voice is still the core of everything it does.

ElevenLabs builds its own AI models for speech, transcription, music, and voice cloning, and now packages them into three products. Each runs on the same voice research, so a voice you clone or design works everywhere.

What changed recently

Section 2

Plans and Credits

ElevenLabs runs on monthly credits. For standard speech, one credit is roughly one character of text, so 1,000 credits is about a minute of audio.

Prices as of October 2026; Enterprise is custom. Annual billing gives about two months free (for example, Creator drops to about $18.33 a month). ElevenLabs often runs first-month promotions, such as a discounted first month on Starter or Creator; see current ElevenLabs pricing for today’s offers.

Which plan should you pick?

Voice agents are metered separately in call minutes (for example, 15 minutes on Free and 275 on Creator), so heavy agent use draws on a different allowance than voiceovers.

Section 3

Choosing a Voice Model

ElevenLabs offers several speech models. Picking the right one is the single easiest way to get better audio.

Section 4

Creating Your First Voiceover

Text to Speech is the heart of ElevenLabs. You can have a finished voiceover in a few minutes.

1. Sign up at elevenlabs.io and open Text to Speech in ElevenCreative.
2. Pick a voice. Browse the Voice Library by use case (narration, advertisement, conversational, characters, social media), language, accent, and age. Play samples and add favorites to My Voices.
3. Choose the model: Eleven v4 for most work (Section 3).
4. Paste your script. Write the way people speak: short sentences, contractions, and numbers written the way you want them read.
5. Adjust settings if needed. Stability trades consistency (higher) against expressiveness (lower); Similarity controls how closely the output sticks to the original voice.
6. Generate, listen, and regenerate any line that sounds off. Each generation uses credits, so fix the script first rather than re-rolling repeatedly.
7. Download the audio as MP3 or WAV and drop it into your video or podcast editor.

Pronunciation fixes

If a name or brand is mispronounced, try spelling it phonetically in the script (“FrostyStack” might become “Frosty Stack”). Eleven v4 also supports IPA phonetic spelling for precise control, and pronunciation dictionaries let you save fixes for reuse.

Section 5

Directing the Performance with Eleven v4

Older text-to-speech models read text aloud. Eleven v4 performs it. You direct the performance two ways: with inline audio tags and with plain-language direction.

Audio tags

Put short instructions in square brackets right where you want them. The voice acts them out instead of reading them. Tags can control emotion, delivery, and even background sounds:

Tips for tags that work

• Match the tag to the voice. A calm narrator voice will not shout convincingly; pick a voice whose range fits.
• Use tags sparingly. One or two per paragraph sounds natural; one per sentence sounds like an act.
• Let the text carry emotion too. v4 reads tone and context from the words, so a well-written line needs fewer tags.
• Use punctuation for pacing. Ellipses and commas slow the read. Note that Eleven v4 has no speed slider and does not read SSML break tags, so pace with the writing itself.

Multi-speaker dialogue

Eleven v4 is built to handle natural conversation between several speakers, with consistent voices across long productions. Use it for podcast-style dialogue, interviews, character scenes, and two-voice ads.

Section 6

Voices: Library, Design, and Cloning

You have four ways to get a voice, from picking one off the shelf to cloning your own.

Getting a great clone

• Record in a quiet room with a decent microphone; no music, echo, or background noise.
• Speak the way you want the clone to sound. A flat read produces a flat clone.
• For a Professional clone, provide more audio, ideally well over 30 minutes, in a consistent style.
• Use the Voice Isolator to clean up a noisy recording before cloning it.

Voice actors can also share Professional clones in the Voice Library and earn when others use them, and the Iconic Marketplace offers licensed voices of well-known figures.

Section 7

Studio: Audiobooks, Podcasts, and Long-Form Audio

Studio is ElevenLabs’ all-in-one editor for long projects. Import a manuscript, script, or article, and it splits the text into chapters and paragraphs you can voice, edit, and regenerate one piece at a time.

• Assign voices to narrators and characters, then switch voices per paragraph or line.
• Regenerate single sentences without redoing a whole chapter, which saves credits.
• Add music and sound effects on a timeline, and export finished audio or video.
• Podcasts: turn a document or URL into a two-host conversation, then edit the script before generating.

Section 8

Beyond Voice: Everything Else in ElevenCreative

ElevenLabs is no longer just a voice tool. These features live alongside Text to Speech in the same account and use the same credits.

A Flows example: one product, many ads

1. In ElevenCreative, open Flows and click New Flow. Name it as a reusable template.
2. Add an Image Generation node with photos of your product as references.
3. Connect a Video Generation node and describe the motion (“slow push-in, soft light”).
4. Add Text to Speech for a voiceover and Music for a bed, then combine them.
5. Swap the product, hook, voice, or language to generate variations without rebuilding.

Video generation in Flows requires a paid plan. If video is your main focus, compare with dedicated video platforms such as Higgsfield, which FrostyStack also covers in a free Getting Started with Higgsfield guide.

Section 9

ElevenAgents: Voice Agents for Your Business

ElevenAgents lets you build AI agents that hold natural conversations by voice or text, in 90+ languages, across phone, website chat, email, and WhatsApp.

Typical uses are answering customer questions, booking appointments, qualifying leads, and handling after-hours calls.

Building a simple agent

1. Open Agents and create a new agent from a template or from scratch.
2. Write its instructions: who it is, what it should do, what it must never do, and when to hand off to a human.
3. Choose its voice (v4 Turbo or another low-latency model for natural back-and-forth) and add a knowledge base, such as your FAQ, services, and hours.
4. Connect tools if needed, such as a calendar or CRM, and set guardrails for compliance.
5. Use Testing to simulate conversations before going live, then connect a phone number or embed the chat widget.
6. Watch Analytics (resolution rate and other metrics) and refine the instructions over time.

Agent call minutes are a separate allowance on each plan (Section 2), and live calls also depend on concurrency limits.

Section 10

Using ElevenLabs from Claude (MCP)

ElevenLabs maintains an official MCP server, so Claude can generate speech, clone voices, transcribe audio, and more from a chat.

MCP (Model Context Protocol) is an open standard that lets an AI assistant use outside tools. With the ElevenLabs server connected, you can ask Claude to write a script and voice it in one conversation.

Set it up in Claude Desktop

1. In ElevenLabs, create an API key under Developers > API Keys.
2. Install uv, a small Python tool runner that the server uses (instructions at astral.sh/uv).
3. In Claude Desktop, open Settings > Developer > Edit Config and add the server below, with your key.
4. Restart Claude Desktop. ElevenLabs tools appear in the chat’s tool list.

{ "mcpServers": {
    "ElevenLabs": {
      "command": "uvx",
      "args": ["elevenlabs-mcp"],
      "env": { "ELEVENLABS_API_KEY": "your-api-key-here" }
    } } }

This server runs on your own computer, so it works in the Claude Desktop app rather than claude.ai in a browser. Generated audio files are saved to your computer. ElevenLabs entries for its Agents product have also started appearing in Claude’s connector directory; check there for the latest one-click options.

Things to ask Claude

Section 11

Rights, Safety, and Disclosure

A few rules protect your account, your clients, and the people whose voices you use.

• Commercial use starts on Starter. Free-plan audio cannot be used commercially and must credit ElevenLabs.
• Consent for every clone. Never clone a real person’s voice without permission, including celebrities, politicians, or clients’ staff.
• Disclose AI where it matters: on platforms that require AI labels, in political or news content, and to callers speaking with an AI agent.
• Music rights vary by plan. Check the license tier before using generated music in ads or client work.
• Platform rules apply. YouTube, TikTok, audiobook distributors, and ad networks each have their own policies on synthetic media.

Section 12

Saving Credits and Avoiding Common Mistakes

Most wasted credits come from the same handful of habits.

Section 13

Quick-Reference Checklist

Tick items off as you go. The PDF version has a printable page.

Getting started

• Account created via the FrostyStack ElevenLabs link; plan chosen (Starter or higher for commercial use)
• Favorite voices saved from the Voice Library, or a voice designed or cloned (with consent)
• Default model set to Eleven v4 for creative work

Every voiceover

• Script written for the ear: short sentences, numbers spelled out as spoken
• Tested on a short paragraph; stability and similarity adjusted
• Audio tags added sparingly; pronunciations fixed
• Listened on headphones before downloading

Before publishing

• Commercial license confirmed for your plan
• Consent on file for any cloned voice
• AI disclosure added where the platform or law requires it