ElevenLabs Voice Cloning: How to Clone Your Voice (Instant vs Professional) 2026

Recording every voiceover yourself gets old by the third retake.

Short answer: start with ElevenLabs Instant Voice Cloning on the $6 Starter plan. Two clean minutes of audio gets you a usable clone in seconds. Pay for Professional Voice Cloning on Creator ($22) only when you’ll publish hours of narration in your own voice.

The catch? Feeding the instant clone more audio can make it sound worse. Here’s how to clone your voice properly, and what it costs per video.

Instant or Professional: which ElevenLabs voice clone do you need?

Instant fits most creators. Professional fits people whose voice is the product: narrators, course creators and weekly channels.

They aren’t a fast and slow version of the same thing. An Instant clone uses your sample as a live reference every time it speaks. Nothing gets trained.

A Professional clone trains a model on your recordings. That’s why it holds up when the script turns emotional.

Pick Instant if you make a few videos a month or want to test a cloned voice on your channel. It’s also the only option when you need a voice by tonight.

Pick Professional if you’ll narrate a 6-hour audiobook or a 40-lesson course. Over that much audio, small slips in an Instant clone add up, and listeners notice.

Our take: make the Instant clone first, even if you plan to go Professional. It costs nothing extra on Starter. In ten minutes it tells you whether your setup deserves three hours of recording.

What do you need before you start?

A paid ElevenLabs account, a quiet room and a decent USB or XLR microphone. The free plan doesn’t include voice cloning at all.

Best for: YouTubers, podcasters and course creators who write scripts and want them read in their own voice without booking studio time.

Not ideal for: anyone hoping to clone a celebrity or a colleague from old clips. The rules section explains why.

Plans matter here. Starter at $6 a month adds Instant cloning, a commercial licence and 10 custom voice slots.

Creator at $22 adds one Professional clone and 30 slots. Our ElevenLabs pricing breakdown covers the higher tiers and annual billing.

How we checked: we worked from ElevenLabs’ product guides, help center and plan table as they stood on October 17, 2026. Both Python scripts below ran on version 2.73.0 of the official SDK with a dummy key. Each reached ElevenLabs and stopped at the key check.

We didn’t create a new clone for this post, so treat the voice-quality advice as ElevenLabs’ guidance plus our judgment.

Heads-up: we earn a commission if you subscribe through some links on this page. It doesn’t change what we recommend.

How do you record a sample ElevenLabs can clone well?

Record in a dead-quiet, soft-furnished room, two fists away from the mic, at a steady level. How you record matters far more than how long you talk.

Why? The model copies everything it hears. Room echo, a fridge hum and mouth clicks all end up in the clone.

So does your energy: read flat, and your clone reads flat forever.

• Room: a closet full of clothes or a duvet hung behind the mic beats a bare room with an expensive microphone.
• Distance: about two fists from the mic, with a pop filter to tame hard P and B sounds.
• Level: average loudness between -23 and -18 dB, with peaks no higher than -3 dB. Most free editors, including Audacity, show this.
• File: MP3 at 192 kbps. WAV doesn’t improve the clone and can trip up the upload.
• Delivery: read in the exact style you want back. For YouTube explainers, record yourself explaining, not reading a novel aloud.

Using old recordings? Keep only clips with one speaker, no music and no long silences. If you edit in Descript, our Descript review covers its Studio Sound cleanup, which helps here.

How to make an Instant voice clone in ElevenLabs

Open Voices, pick Instant Voice Clone, upload two minutes of audio and save. With a sample ready, it takes under ten minutes.

Here’s the part most guides skip. More audio doesn’t help an Instant clone.

ElevenLabs says anything past 2-3 minutes adds little and can make the clone less stable. Ten minutes of mixed clips loses to 90 clean seconds.

If the test sounds off, don’t touch the settings first. Swap the sample. A small change in what you upload beats any slider.

And you can’t change a clone’s accent or tone after it’s made.

How to make a Professional voice clone

Upload 30+ minutes of clean speech, prove it’s your voice, then wait a few hours. It’s slower, pickier and starts on Creator.

Thirty usable minutes means far more time at the mic, because you’ll retake fluffed lines. Block an afternoon.

Trying the voice too early is the classic mistake. If you see “No model found for this voice”, training hasn’t finished. Check its status in My Voices and come back later.

What will your cloned voice cost each month?

For a creator posting one 10-minute video a week, about $22 a month on Creator. Starter’s 30,000 credits cover roughly three 10-minute voiceovers, and retakes eat into that fast.

The maths is simple. On the Multilingual v2 model, one character costs one credit.

ElevenLabs’ plan table puts 30,000 credits at about 30 minutes of speech. So a 1,500-word script runs close to 10 minutes and 10,000 credits.

Unused credits on paid plans roll over for up to two months. Annual billing drops Creator to about $18.33 a month.

Real-world tip: regenerate one paragraph at a time, not the whole script. A bad take on a 10,000-credit script costs you a third of a Starter month.

Why does my cloned voice sound wrong, and how do you fix it?

Almost every bad clone traces back to the sample, not the settings. Check these in order.

• It sounds echoey or muffled. Your room is in the sample. Re-record in a soft-furnished space; noise removal only gets you part of the way.
• It sounds bored. You read flat, so it reads flat. Record again with the energy you want back.
• The accent is off. Instant cloning can struggle with less common accents. Try a different sample first, then Professional if that fails.
• It changes mid-paragraph. Your sample mixes sessions, mics or moods. Cut it down to one consistent take.
• It breathes or clicks oddly. Those sounds were in the recording. Edit them out before uploading, not after generating.

Whatever you fix, keep your original samples. A rebuilt clone sounds slightly different even from the same audio. Those files are your only real backup.

Can you clone someone else’s voice?

With Instant cloning, only if you have their permission and the right to use it. With Professional cloning, no, not even with their written consent.

Instant clones ask you to tick a consent box. Professional clones must pass a live voice check against the samples.

Working with a partner or a voice actor? They create and verify their own clone, then share it with you privately by link.

Every clip made on the platform can be traced to the account that generated it. Stay inside the rules.

How do you use your clone outside ElevenLabs?

Through the API. You can’t download a clone as a file. But any app holding your API key can make it speak, from a script to a voice agent.

This script creates an Instant clone from one sample. It needs pip install elevenlabs and your key in the ELEVENLABS_API_KEY environment variable.

import os
from elevenlabs.client import ElevenLabs

client = ElevenLabs(api_key=os.environ["ELEVENLABS_API_KEY"])

with open("my-voice-sample.mp3", "rb") as sample:
    voice = client.voices.ivc.create(
        name="My narration voice",
        files=[sample],
        remove_background_noise=False,  # True only if the room was noisy
    )

print("Voice ID:", voice.voice_id)

Copy the voice ID it prints. This second script uses that ID to read a line you never recorded and saves it as an MP3.

import os
from elevenlabs.client import ElevenLabs

client = ElevenLabs(api_key=os.environ["ELEVENLABS_API_KEY"])

audio = client.text_to_speech.convert(
    voice_id="PASTE_YOUR_VOICE_ID",
    model_id="eleven_multilingual_v2",
    text="This line was never recorded. My clone read it.",
    output_format="mp3_44100_128",
)

with open("test-line.mp3", "wb") as f:
    for chunk in audio:
        f.write(chunk)

The same voice ID plugs into voice agent platforms. Our Vapi review and Retell review show where ElevenLabs voices fit into a phone agent. That’s a common next step once a clone sounds right.

Frequently asked questions

Which clone should you make first?

Make the Instant voice clone first. Two clean minutes on Starter tells you whether your room and mic are good enough. For most YouTube and podcast work, it’s all you’ll need.

Move to a Professional voice clone on Creator when you’re narrating hours, not minutes. Whichever you pick, protect the sample: one room, one mic, one mood. Then compare it with the alternatives in our ElevenLabs vs Murf AI comparison if the voice still isn’t right.

Ready to try ElevenLabs voice cloning? Record your two minutes today, then read the full ElevenLabs review before you upgrade.