AI ConsultingAboutVideosBookFAQContact
DE|EN
Book Free Consultation

Home/Videos/AI Music & Voice/ElevenLabs v4 Voice Clone

ElevenLabs v4 Voice Clone: Does Instant Finally Beat Professional?

ElevenLabs says instant clones in v4 now beat the professional clones of its older Multilingual v2 model. I put four clones of my voice up against my real voice in four blind rounds. In English, the v4 instant clone from two minutes of audio came out on top three times; in German, my professional clone won. Make your own guess before you read the reveal.

Tool testVoice cloningElevenLabsBlind test

Original video by Lukas Hüttis, published October 19, 2026, 15 min · Blind test to play along · Watch on YouTube

Chapters

Each timestamp jumps to the matching section on this page. “Load video” then starts right there.

Want to try ElevenLabs yourself right away? Go to ElevenLabs*

* Affiliate link: I'm an independent ElevenLabs affiliate and earn a commission.

Five voices, one of them is me

▶ In the video at 01:09

All four clones live in my own ElevenLabs account, which I pay for myself. Every clone reads the same sentences with ElevenLabs' default settings. All samples are loudness-matched, so the loudest one doesn't win by default, and Claude shuffled the order of every round for me.

The candidates

  1. AProfessional clone, Multilingual v2My clone from Part 1 in July, trained on about two hours of audio from my videos plus a recording made directly in ElevenLabs.
  2. BInstant clone, Eleven v4, 10 secondsFrom exactly 10 seconds of audio. ElevenLabs accepts it and suggests adding more recordings.
  3. CInstant clone, Eleven v4, 2 minutesFrom 2:13 minutes of audio, the same clone as in my HeyGen blind test. How I set it up is in the tutorial.
  4. DProfessional clone on Eleven v4The same clone as A, switched to Eleven v4 in text to speech, but not retrained.
  5. MeMy real voiceThe same sentences, recorded myself.

Play along: which voice is real?

▶ In the video at 00:00

Four rounds, just like in the video, five samples each. One is my real voice, four are AI clones of my voice made with ElevenLabs. Each row jumps to that exact sample on YouTube. Make your guess first, then hit “Reveal” here. Headphones help.

Blind test · Round 1 of 4

A normal sentence

The sentenceHey, it's Lukas. Today I'm testing whether a ten-second voice clone can really replace two hours of studio recordings. I'll keep it honest: if it works, I'll say so, and if it doesn't, you'll hear why.

  1. 01Real or AI?C · Instant v4, 2 minin the video at 2:25got me
  2. 02Real or AI?My real voicein the video at 2:39
  3. 03Real or AI?B · Instant v4, 10 secin the video at 2:51
  4. 04Real or AI?A · Professional, Multilingual v2in the video at 3:05
  5. 05Real or AI?D · Professional on v4in the video at 3:18
Reveal

My real voice was number two. I had guessed one or four. Number four was A, my professional clone, and it was super close. Number one was C, the instant clone from two minutes: ElevenLabs got me there.

Blind test · Round 2 of 4

Price and name

The sentenceElevenLabs charges $22 a month, Claude Code costs more, and my name is Lukas Hüttis.

  1. 01Real or AI?My real voicein the video at 4:37
  2. 02Real or AI?A · Professional, Multilingual v2in the video at 4:44
  3. 03Real or AI?B · Instant v4, 10 secin the video at 4:51
  4. 04Real or AI?C · Instant v4, 2 minin the video at 4:59my pick
  5. 05Real or AI?D · Professional on v4in the video at 5:07
Reveal

This time I got it right, number one was me. Three and four sounded best to me, B from 10 seconds and C from two minutes. Number two, my professional clone, was pretty good too; it only mixed up my name. A tough one; I'll probably give it to C again.

Blind test · Round 3 of 4

The sentence from the start

The sentenceI built this entire project over a single weekend and honestly I still cannot believe it actually works.

  1. 01Real or AI?My real voicein the video at 6:42
  2. 02Real or AI?A · Professional, Multilingual v2in the video at 6:48
  3. 03Real or AI?D · Professional on v4in the video at 6:55
  4. 04Real or AI?C · Instant v4, 2 minin the video at 7:02my winner
  5. 05Real or AI?B · Instant v4, 10 secin the video at 7:09
Reveal

Number one sounded like me, and it was me. Three and four sounded best, D and C. Probably my winner this round: number four, the instant clone from two minutes.

Blind test · Round 4 of 4

German, my native language

The sentenceHey, hier ist Lukas. Heute teste ich, ob ein Stimmklon aus zehn Sekunden wirklich zwei Stunden Studioaufnahmen ersetzen kann. Ich bleibe ehrlich: Wenn es funktioniert, sage ich es, und wenn nicht, hörst du, warum.

  1. 01Real or AI?D · Professional on v4in the video at 7:55
  2. 02Real or AI?A · Professional, Multilingual v2in the video at 8:10wins
  3. 03Real or AI?My real voicein the video at 8:23
  4. 04Real or AI?C · Instant v4, 2 minin the video at 8:35
  5. 05Real or AI?B · Instant v4, 10 secin the video at 8:50
Reveal

Four and five were definitely not me. At first I thought number one was me, but not after listening again. My real voice was number three. Number four, the instant clone from two minutes, was pretty good, but in German my pick is number two: my professional clone on Multilingual v2.

My result, round by round

▶ In the video at 14:08
C3 rounds
A1 round
B0
D0
RoundMy guessReal wasMy pick
1 · a normal sentence1 or 4No. 2C, instant v4 from 2 min
2 · price and nameNo. 1, rightNo. 1C, instant v4 from 2 min
3 · the sentence from the startNo. 1, rightNo. 1C, instant v4 from 2 min
4 · Germanfirst 1, then 3, rightNo. 3A, professional on Multilingual v2
Spanish–no real voiceopen: what do you hear?
Audio tags––worked in B, C and D

“Does instant finally beat professional? I feel like it really does.”

That's my verdict in the video, and it's about the v4 instant clone from two minutes of audio. It was my pick in all three English rounds, and it's fast: the clone was ready in about 11 seconds, and the speech came out practically instantly too.

In German, my native language, my professional clone on Multilingual v2 still won. The clone from 10 seconds wasn't my pick in any round. Neither was D, my professional clone on v4 without retraining.

Curious how your own voice sounds as a v4 instant clone? You need the Starter plan or higher; prices are further down.

Go to ElevenLabs*

* Affiliate link: I'm an independent ElevenLabs affiliate and earn a commission.

What ElevenLabs promises, and what I heard

▶ In the video at 00:42

In July, in Part 1, my instant clone didn't sound like me and my professional clone did. ElevenLabs says that changed with v4. The quotes come from ElevenLabs' own pages, checked on October 17, 2026.

  • Instant Voice Clones in Eleven v4 now outperform the Professional Voice Clones of Multilingual v2.
    v4 page

    In my testIn English, yes: C was my pick in all three rounds, and A was super close in round 1. In German, A won.

  • Instant Voice Clones can now capture voices with high fidelity using just 10 seconds of audio.
    v4 blog post

    In my testElevenLabs accepted the 10 seconds right away. B wasn't my pick in any round, but it made my shortlist in round 2.

  • Approximately 1-2 minutes of clear audio without any reverb, artifacts, or background noise of any kind is recommended.
    Instant Voice Cloning docs

    In my testC, from 2:13 minutes, was my pick in three of four rounds.

  • Professional Voice Clone support for Eleven v4 is currently rolling out to everyone
    Eleven v4 docs

    In my testOn October 16, my professional clone didn't list v4 under “Fine-tuned models” yet. D ran on v4 without retraining and wasn't my pick in any round.

  • They’re not perfect yet
    Eleven v4 docs, on audio tags

    In my testLaughing, whispering and sighing worked in every v4 clone I tried.

What I saw on screen

Clone B in ElevenLabs: file R10.mp3 with 10 seconds, status Ready
Clone B: ElevenLabs accepts the 10 seconds (“Ready”) and suggests adding more recordings. Screen recording from October 11, 2026.
My professional clone in ElevenLabs, list of fine-tuned models without Eleven v4
My professional clone on October 16, 2026: v4 isn't listed under “Fine-tuned models” yet. Above it in the list, the two instant clones B and C.

Spanish: this is where I need you

▶ In the video at 10:25

I don't speak Spanish, except “soy Lukas” and “una cerveza, por favor”. So there's no real voice and no blind test here, just the four clones side by side.

According to ElevenLabs, v4 speaks another language fluently and naturally instead of carrying over the reference voice's accent. ElevenLabs says it is exploring ways to make that a toggle, with no timeline. With the Spanish sentence, ElevenLabs itself shows the hint “Use a Spanish voice for best results”.

Open · not a blind test

Spanish

The sentenceHola, soy Lukas. Hoy pruebo si un clon de voz hecho con diez segundos de audio puede reemplazar dos horas de grabación en estudio. Voy a ser honesto: si funciona, lo diré; y si no, vas a escuchar por qué.

  1. AProfessional, Multilingual v2in the video at 10:59
  2. BInstant v4, 10 secin the video at 11:08
  3. CInstant v4, 2 minin the video at 11:17
  4. DProfessional on v4in the video at 11:26

Which one sounds most natural to you, and most like me? Listen to the four samples in the video and tell me in the comments.

Comment on YouTube

Audio tags: laugh, whisper, sigh

▶ In the video at 11:51

Audio tags in square brackets tell v4 how something should sound. ElevenLabs itself says they're “not perfect yet”. In my HeyGen test they had already worked pretty well.

This time, laughing, whispering and sighing worked in every v4 clone I tried: B, C in two takes, and D. A isn't in this round: it runs on Multilingual v2; ElevenLabs' docs describe audio tags for Eleven v4 and v3. My take in the video: “the instant v4 really rules”.

Open · three tags per sample

Audio tags

The sentence[laughs] Okay, I did not expect that. [whispers] Don't tell anyone, but this might change how I record. [sighs] Or I'll be re-recording everything tomorrow.

  1. BInstant v4, 10 secin the video at 12:24✓ ✓ ✓
  2. CInstant v4, 2 min, take 1in the video at 12:39✓ ✓ ✓
  3. CInstant v4, 2 min, take 2in the video at 12:54✓ ✓ ✓
  4. DProfessional on v4in the video at 13:14✓ ✓ ✓

What's missing in v4

▶ In the video at 13:45

v4 has no speed or style slider and no SSML. Stability and similarity are what's left. But with the default settings, it already sounded really good to me.

According to ElevenLabs, clones created before v4 should be retrained for v4: in My Voices, hover over the voice and click the plus button. On October 16, my professional clone didn't list v4 yet; the docs say professional clone support for v4 is rolling out. As soon as I can do it, there'll be a separate video.

Text to speech in ElevenLabs with Eleven v4: only the Stability and Similarity sliders
Eleven v4 in text to speech on October 16, 2026: only Stability and Similarity left. For the Spanish sentence, ElevenLabs suggests a Spanish voice.

Which plan do you need?

▶ In the video at 14:32

Regular monthly prices from the pricing page, as of October 17, 2026 (in the video: as of October 16).

  • Starter$6 a monthInstant clones like B and C. The free plan has no voice cloning.
  • Creator$22 a monthAdds a professional clone like A and D.
  • VoiceboxfreeOpen source, runs locally on your computer, takes some setting up. I have a series on it.

Frequently asked questions

Eleven v4, released on September 28, 2026, is ElevenLabs' newest text to speech model. For voice cloning, ElevenLabs promises three things: instant clones that, it says, outperform the professional clones of the older Multilingual v2 model and can be made from just ten seconds of audio, professional clones that are supported again after Eleven v3, and other languages without the reference voice's accent. Audio tags like [laughs] or [whispers] steer the delivery. v4 has no style or speed sliders and doesn't support SSML. According to ElevenLabs, clones created before v4 should be retrained for v4.

That's ElevenLabs' claim, with a catch: it refers to the professional clones of the older Multilingual v2 model, not to a professional clone on v4. In my blind test, the instant clone from two minutes was my pick in all three English rounds, ahead of my professional clone on Multilingual v2. In German, the professional clone won. I couldn't test a professional clone retrained on v4 yet: on October 16, v4 wasn't listed among my clone's fine-tuned models.

To create one, yes. ElevenLabs accepted my 10 seconds and made the clone, with a note that more recordings make it better. The v4 blog post says an instant clone can now capture a voice from just 10 seconds of audio. ElevenLabs' docs, however, still recommend about 1 to 2 minutes of clean audio: at least 1 minute and no more than 3. In my blind test, the 10-second clone wasn't my pick in any round, while the two-minute clone was my pick three times.

Yes. My clone was recorded in English and still speaks German and Spanish. According to ElevenLabs, a v4 clone speaking a different language than its reference recording sounds fluent and natural instead of carrying over the original accent. In my German blind test, my professional clone on Multilingual v2 came closest to my voice, ahead of the v4 instant clone from two minutes. At first I even took D, my professional clone on v4, for my real voice. Listening again made it clear: no. Spanish I can't judge; the four Spanish samples are linked to their spot in the video above.

For me, yes. ElevenLabs itself says the tags are “not perfect yet”. With the sentence “[laughs] Okay, I did not expect that. [whispers] Don't tell anyone …”, laughing, whispering and sighing worked in every v4 clone I tried. My professional clone on Multilingual v2 wasn't part of this round; ElevenLabs' docs describe audio tags for Eleven v4 and v3. All four samples are linked to their spot in the video in the audio tags section above.

For an instant clone, Starter at $6 a month is enough; for a professional clone you need Creator at $22 a month (regular monthly prices as of October 17, 2026). The free plan includes 10,000 credits a month but no voice cloning. In my test, the instant clone from two minutes was my pick in English, so Starter covers that. If you'd rather not pay at all, my Voicebox series runs a voice clone locally on your own computer, with some setting up.

Sources

All provider details come from ElevenLabs' official pages, with retrieval dates.

ElevenLabs

Related

Want a voice clone for your courses, explainer or product videos?

I work 1:1 with freelancers, consultants, coaches and small teams on practical AI workflows, automation and custom tools. The intro call is free and takes 15 minutes.

Book a free strategy call

More videos →