How to Use ElevenLabs: TTS, Voice Cloning, and API Guide

The short answer (August 2026): ElevenLabs is an AI audio platform covering text-to-speech, voice cloning, video dubbing, music generation, sound effects, and real-time voice agents. You can start free at elevenlabs.io with 10,000 credits per month (roughly 10 minutes of speech at 1 credit per character), or pay from $6/month to unlock commercial rights and higher output. The API uses a single environment variable for authentication and ships official Python and JavaScript SDKs. Quality ranges from ultra-fast and cheap (Eleven Flash at ~75ms latency) to most expressive and widest language coverage (Eleven v3 across 70+ languages).

Last verified: August 3, 2026, against official docs and pricing pages.

What ElevenLabs Is Now

ElevenLabs started as a text-to-speech tool and now operates across three product surfaces. ElevenCreative covers TTS, voice cloning, dubbing, music, and sound effects. ElevenAgents lets you build and deploy conversational voice agents on phone, web, and mobile. ElevenAPI gives programmatic access to all of the above.

The core TTS workflow has not changed: paste text, pick a voice from the voice library, pick a model, generate audio. What has changed is that enterprises increasingly use the platform to replace IVR phone trees, and developers use the API for production voice pipelines that combine speech input and output in real time.

Plans, Credits, and Commercial Rights

Credits are the unit of account. In the UI, each character of text costs 1 credit regardless of which TTS model you pick. Via the API, Flash and Turbo models carry discounted rates. On paid plans, unused credits roll over for up to two months, capped at 2x your monthly quota, so your maximum balance at any point is 3x your monthly allotment.

Annual billing removes two months of cost: you pay for 10 months and get 12.

The free plan carries no commercial license and dubs carry a watermark. Every paid plan includes a commercial license.

The Creator plan at $22/month is the first tier to include Professional Voice Cloning, which matters if you need a clone that is intended to be near-indistinguishable from the original speaker.

Which Model to Use

Model choice determines expressiveness, language coverage, per-request latency, and the maximum text you can pass in a single call. Picking the wrong one is the most common source of unnecessary spend or disappointing output.

eleven_v3 covers the most languages and delivers the highest emotional range, but its 5,000-character ceiling means you need to chunk longer scripts before sending. eleven_flash_v2_5 is the right model for any latency-sensitive use case and for high-volume API work where the discounted per-character rate matters. eleven_multilingual_v2 remains the default and is the right choice for long-form narration where you are not hitting a latency requirement.

Beyond TTS, ElevenLabs also offers speech-to-text models (scribe_v2 for batch transcription in 90+ languages, scribe_v2_realtime for streaming at ~150ms latency), music generation (music_v2), and sound effects (eleven_text_to_sound_v2).

Getting Started: Browser and API

Browser: Create a free account at elevenlabs.io, open the Text to Speech tool in ElevenCreative, pick a voice from the library (searchable by language, accent, gender, and age), paste your text, select a model, and generate. MP3 and WAV downloads are available from the same screen.

API key: Go to your profile in the ElevenLabs dashboard and generate an API key. Store it as an environment variable.

Python setup:

pip install elevenlabs python-dotenv

JavaScript setup:

npm install @elevenlabs/elevenlabs-js dotenv

Basic Python TTS call:

import os
from dotenv import load_dotenv
from elevenlabs.client import ElevenLabs
from elevenlabs.play import play

load_dotenv()
client = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))

audio = client.text_to_speech.convert(
text="Your script text here.",
voice_id="JBFqnCBsd6RMkjVDRZzb", # replace with your chosen voice ID
model_id="eleven_flash_v2_5",
output_format="mp3_44100_128",
)

play(audio)

The REST endpoint is POST https://api.elevenlabs.io/v1/text-to-speech/{voice_id}. ElevenLabs exposes regional variants for US, EU, India, and Singapore if your use case requires geographic data routing.

Voice Cloning: Instant vs. Professional

The two cloning paths have fundamentally different quality ceilings, time requirements, and plan gates.

Audio quality matters more than file format for either path. Record between -23 dB and -18 dB RMS with a true peak of -3 dB, with no background noise or reverb. For Instant Cloning, ElevenLabs' own documentation warns against exceeding 3 minutes of source audio. More audio past that threshold does not improve quality and can degrade the clone.

MP3 at 192 kbps and WAV both work fine. ElevenLabs explicitly notes that using an uncompressed WAV yields little to no quality improvement over a high-bitrate MP3.

Professional Voice Cloning requires a voice captcha step to verify the uploaded audio matches your own voice. You cannot use this pathway to clone a third party's voice.

Dubbing, Sound Effects, and Music

Dubbing is available on all plans, including free (with watermarks on free). Upload a video or audio file or paste a YouTube or TikTok URL, select target languages, and ElevenLabs detects each speaker, translates the script, and regenerates speech in the new language while preserving each speaker's voice characteristics. Upload formats include MP4, MOV, MKV, MP3, WAV, and about 15 others. Output formats are MP4 (video), AAC, WAV, and SRT (subtitles). Maximum file size is 2 GB and 180 minutes. Pricing scales by content duration and number of target languages.

Sound effects: Describe the sound you want in plain English using the eleven_text_to_sound_v2 model and download the generated clip. Useful for prototyping audio in product demos or videos before commissioning production assets.

Music: The music_v2 model generates tracks from a text prompt in any style. Output quality is suitable for prototyping. For release-ready music, treat the generation as a draft and iterate.

Voice Agents

ElevenAgents combines ElevenLabs' speech recognition, a language model of your choice, and TTS into a real-time conversation loop. You configure agents through a visual dashboard: select an LLM, design conversation flows, attach a knowledge base or external API calls, and assign a voice. Agents deploy to phone calls, web chat, and mobile.

For standalone transcription, scribe_v2 handles batch jobs in 90+ languages with word-level timestamps, speaker diarization for up to 32 speakers, and detection of 65 entity types. scribe_v2_realtime streams partial transcripts at roughly 150ms latency and supports voice activity detection.

For more on building agentic systems around voice or other tools, the AI Weekly guide on how to build an AI agent covers the underlying architecture. For workflow automation ideas, see how to automate work with AI.

FAQ

Does the free plan allow commercial use?
No. ElevenLabs excludes a commercial license from the free tier. The Starter plan at $6/month is the minimum for commercial licensing.

What happens to unused credits at the end of a billing cycle?
On paid plans, credits roll over for up to two months, capped at 2x your monthly quota. Your maximum total balance across any billing cycle is 3x your monthly allotment.

Can I clone someone else's voice?
The Professional Voice Cloning pathway requires a voice captcha that verifies the uploaded audio matches your own voice. Legally, rules vary by jurisdiction. A platform permit from ElevenLabs does not substitute for applicable local law, so consult legal counsel before cloning a third party's voice for any commercial purpose.

What is the difference between eleven_v3 and eleven_flash_v2_5 in practice?
Use eleven_v3 when expressiveness is the priority: emotional range, dramatic delivery, character performance. Use eleven_flash_v2_5 when latency is the constraint (real-time agents, streaming), when you need to process very long text in a single call (40,000 character limit vs. 5,000), or when per-character API cost matters.

Can I use ElevenLabs voices in a podcast or YouTube video?
Yes, on any paid plan. Starter and above include a commercial license. Voices sourced from the Voice Library may carry additional terms set by the individual voice creator, so check those before publishing.

If this saved you time, get AI Weekly free, 3 issues a week, read by 40,000+ practitioners.