ElevenLabs

What ElevenLabs Covers

ElevenLabs began as a text-to-speech company and now runs a wider audio stack from one account. The model catalogue documented by the vendor spans four areas: text to speech, speech to text, music generation, and voice changing. That breadth matters for anyone producing content at volume, because the alternative is stitching a narration tool, a transcription service and a music library together and reconciling three bills.

On the text-to-speech side, Eleven v3 is the expressive flagship with "70+ languages supported" and a documented 5,000 character limit per request. Eleven Multilingual v2 covers 29 languages with a 10,000 character limit. Eleven Flash v2.5 trades some nuance for speed, documented at "Ultra-low latency (~75ms)" across 32 languages with a 40,000 character limit, and the docs note that numbers are not normalized by default on that model, so a script full of figures wants either v3 or a pass of manual expansion. Eleven Flash v2 is the English-only equivalent with a 30,000 character limit.

For transcription, Scribe v2 handles "Accurate transcription in 90+ languages", and Scribe v2 Realtime does the same set at "Low latency (~150ms)". Eleven Music v2 generates music and is documented for "English, Spanish, German, Japanese and more". Eleven Multilingual STS v2 is the voice changer, matching the 29 languages of Multilingual v2.

Choosing a Model Is the Real Skill

The interesting decision inside ElevenLabs is not whether the audio is good, it is which model fits the job. A long-form YouTube narration wants Eleven v3 for delivery, and the 5,000 character ceiling simply means chunking the script, which you would do anyway for editing. A live agent or an interactive product wants Flash v2.5, where 75 milliseconds is the difference between a conversation and a wait. A publisher converting an existing back catalogue wants the larger character windows so fewer requests are needed. Experienced producers end up using two or three of these models in the same pipeline rather than picking one.

Voice Cloning

Two cloning routes exist. Instant Voice Cloning is available from the Starter tier upward and works from a short sample. Professional Voice Cloning is the fine-tuned route, and the documentation is specific about what it asks for: "we recommend uploading at least an hour of training audio", "ideally as close to three hours as possible", with the note that the "bare minimum we recommend is 30 minutes" and best results at two to three hours. Training itself is not instant either, since "Usually fine-tuning will take 3-6 hours" and the process "can take up to 24 hours" depending on the queue.

Before fine-tuning begins you must verify the voice, and if verification fails "you can wait 24 hours to try verification again, or reach out to support for help". The vendor is also direct about scope: "For now, we only allow you to clone your own voice." That is a deliberate consent design rather than an oversight, and it tells you exactly which use cases fit. A creator building a personal voice for their own catalogue is the intended user. An agency wanting to clone a client voice needs the client to hold the account and run the verification themselves.

Professional clone slots are allocated by tier: Creator and Pro carry one, Scale carries three, Business carries ten, and Enterprise is custom. Free and Starter do not include professional slots, which reads correctly once you see it as a ladder rather than a wall, since the entry tiers exist for evaluating output quality before anyone commits a three-hour recording session.

Pricing, checked August 2026

The published tiers run as follows. Free is $0 with 10,000 credits per month and access to text to speech, speech to text, sound effects, Voice Design, Music, Productions, Image and three Studio projects. Starter is $6 with 30,000 credits and adds the commercial license, Instant Voice Cloning, 20 Studio projects, commercial use of Music, Dubbing Studio, and Image and Video.

Creator is $22 per month, listed at $11 for the first month, with 121,000 credits and Professional Voice Cloning. Pro is $99 with 600,000 credits and adds "44.1kHz PCM audio output via API" and 192 kbps audio quality. Scale is $299 with 1.8 million credits, three workspace seats, team collaboration and three professional voice clones. Business is $990 with 6 million credits, ten professional voice clones, ten workspace seats and low-latency text to speech quoted as low as five cents per minute. Enterprise is custom and adds DPA and SLA terms, BAAs for HIPAA customers, custom SSO and elevated concurrency limits.

Dubbing is metered separately and priced by output: 2,000 credits per minute for automatic dubbing with a watermark, rising to 10,000 credits per minute for Dubbing Studio without one.

Who It Suits

ElevenLabs suits producers whose output is audio-first and multilingual: podcasters, course creators, localisation teams and anyone building a voice interface. If your work is video editing with narration attached, Descript keeps the audio and the timeline in one place and may serve you better. If you need a presenter on camera rather than a voice over footage, Synthesia is the closer fit. Where ElevenLabs is difficult to beat is raw voice quality across many languages, and the option to go from a free evaluation to a fine-tuned personal voice without changing vendors.

Is ElevenLabs the Right Audio Vendor for You?

Start with what you are producing, because the audio tools have sorted into clear lanes and picking the wrong lane costs more than picking the wrong tier.

If your work is video and the narration is one track inside a timeline, Descript keeps editing and audio together and will feel less like a detour. If you need a presenter visible on screen, Synthesia is built for that and ElevenLabs is not. If what you need is a voice, in many languages, at a quality that survives a paying audience, ElevenLabs is where the conversation ends up.

The breadth is the underrated part. One account covers text to speech across 70 or more languages on Eleven v3, transcription across 90 or more languages on Scribe v2, a model at roughly 75 milliseconds for interactive work, a voice changer and music generation. Teams that start with narration and later need transcription do not have to go shopping again.

Picking a Tier

Checked August 2026, the free plan gives 10,000 credits per month and is genuinely enough to judge output quality on your own scripts before spending anything. It is an evaluation tier, so the commercial license and Instant Voice Cloning begin at Starter, which is $6 per month with 30,000 credits and also opens Dubbing Studio.

Creator at $22 per month with 121,000 credits is where most working creators land, because it is the first tier carrying a Professional Voice Cloning slot. If you are building a product on the API rather than publishing content, Pro at $99 with 600,000 credits adds 44.1 kHz PCM output via the API and 192 kbps audio. Teams move to Scale at $299 for three workspace seats and three professional clones, or Business at $990 for ten of each. Enterprise covers DPA and SLA terms, HIPAA BAAs and custom SSO.

One planning note if a fine-tuned voice is the goal: the documentation recommends at least an hour of training audio and ideally close to three hours, plus a verification step and a fine-tuning run of three to six hours. Book the recording session before you need the voice, not the week you launch.

The Verdict

ElevenLabs is the strongest general-purpose choice when the output is voice and the audience is international. Use the free tier to test it on your own material, move to Starter or Creator once the quality holds up, and climb further only when seats, clone slots or API audio quality become the constraint.