ElevenLabs

You need to consider one thing: whether you need your own voice rather than any voice from the library.

Who it is for. For people making videos, podcasts, and courses who are tired of waiting for a voice actor: paste in a script, choose a voice, and a file is ready in a minute. Starting with the $6 plan, you can record your own voice, and it will then read anything you give it in 74 languages.

Who else. For developers: the service has a full API, a command-line utility, and speech transcription that ranks in the top three for accuracy in an independent arena as of 06.09.2026. One key covers voice generation, transcription, and voice agents.

Who it is not for. Anyone who needs free voiceover to sell: without a subscription the use is personal only, with "elevenlabs.io" credited in the title.

Good fit. To try a different voice, there is voiceover in MiniMax Audio.

What vibecoding isStarting from scratch: how to set tasks for agents with no coding experience

Five things on one credit balance: voice generation, your own voice, transcription, video translation, and music.

Voice generation. You paste text, pick a voice and a model, and click «Generate speech». Since Sep 28, 2026, the flagship has been Eleven v4 with emotion tags in square brackets like «whispering» and «excited» and a limit of 10,000 characters per request. The previous Eleven v3 takes up to 5,000, Multilingual v2 suits long texts, and Flash v2.5 responds in 75 milliseconds and costs half as much via the API. The vendor counts more than ten thousand voices in its library.

Your own voice. An instant clone is made from a short recording and is available from Starter at $6. A professional clone is trained on recordings; the vendor recommends about two hours of clean audio, it is available with Creator at $22, and beforehand the vendor asks you to read a verification phrase aloud within a set time to confirm the voice is yours. Free users get three slots, but can fill them only with a voice created from a description, not their own.

Transcription and video translation. The Scribe v2 model turns recordings into text in over 90 languages and distinguishes up to 32 speakers. Dubbing translates a video into another language and preserves the speaker's voice; it also supports over ninety languages, while the product window still carries an "Alpha" label. Free dubbing comes with an audio watermark that cannot be removed.

Music and everything else. Eleven Music composes tracks with and without vocals under licenses from Merlin and Kobalt. Images and video on the platform are made by third-party models, including Sora 2 and Veo 3.

Studio 3.0 assembles a video with voiceover in a single editor, and ElevenAgents builds voice agents for customer support. To compare ways of editing music, there is also Udio generator, which has downloading results disabled.

For background music that you need to edit note by note, consider AIVA music editorseparately. It lets you export a composition as MIDI for further work.

What the vendor itself shows. The header video comes from the Studio 3.0 landing page: an editor window with project text on the left, a voice replacement panel on the right, and voice and effects tracks below, while a product list from voice generation to noise removal slides in from the left. We did not run ElevenLabs for this task ourselves; the figures come from docs, help pages, the storefront, and third-party tests.

You can start in a browser without installing anything.

Step 1. Open elevenlabs.io and sign up with Google or email. For personal use, the vendor asks for a personal email address; work requires a paid plan. Without signing in, you can only generate speech in the homepage demo.

Step 2. In the Text to Speech section, paste text, choose a voice and model, then click "Generate speech". The finished file downloads as an mp3, with an uncompressed format also available on Pro and higher. Add your own voice in the Voices section after upgrading to Starter.

Apps. For iPhone there is an app in the App Store. For Android there is a Google Play page with five million installs. There is no desktop app, and the query "download for pc" hits nothing: on a computer you use the website, and since Aug 24, 2026 developers have a command-line tool.

Where voice generation can be used. On paid plans, the file is yours: you can put it in a video, podcast, game, or ad without mentioning the vendor. On the free plan, you can publish only with "elevenlabs.io" or "11.ai" in the title and only not for money.

Steps verified Sep 6, 2026 against the live elevenlabs.io storefront, the vendor's docs and help center, and the app stores

The company has a lot of money, and since Sep 28, 2026, its new model has ranked first in an independent speech synthesis arena.

Arena. Artificial Analysis compares voices in blind pairs. As of Sep 29, 2026, Eleven v4 ranks first out of 92 models, the previous flagship Eleven v3 is seventeenth, and in the single-voice-for-all ranking v4 trails Alibaba's Qwen-Audio 3.1. As of Sep 6, 2026, Cartesia, Inworld, Qwen and Simba from Speechifyranked above Eleven v3. In speech transcription, Scribe v2 is third by overall word error rate, 2.2%, though it runs slower than the leaders.

Company funding. The Feb 4, 2026 round brought in $500 million at an $11 billion valuation, with $781 million raised across five rounds. Annual revenue at the end of 2025 was $330 or $350 million according to two vendor posts, and exceeded $500 million by May 2026. These are all company figures; the private firm has no public reporting.

Stores. The US App Store rates the app 4.84 across 19,800 ratings, Google Play 4.7 across 212,000 reviews and more than five million installs (Sep 6, 2026).

One currency for all products: credits per text character, while a minute of different work costs from one thousand to ten thousand.

Free. 10,000 credits per month, about ten minutes of voiceover. Voiceover, transcription, effects, voice generation from a description, and three Studio projects are available. Cloning, commercial use, and rolling unused credits into the next month are unavailable, and published work must include "elevenlabs.io" or "11.ai" in its title.

Starter and Creator. Starter at $6 gives 30,000 credits, a commercial license, and instant voice cloning. Creator at $22 gives 121,000 credits and professional cloning; the vendor sells the first month for $11 and displays that prominently in the storefront with $22 crossed out. Annual billing makes any plan two months cheaper.

Pro and above. Pro at $99 unlocks 600,000 credits and uncompressed audio via API. Scale at $299 and Business at $990 add team seats and up to ten professional clones. The vendor revised the plans during Apr 8–15, 2026: Business dropped from $1,320 to $990, but credits fell from 11 million to 6 million, so a thousand credits became 37% more expensive.

Credits in practice. A thousand credits provide about a minute of speech, and the same balance is used by other products: transcription costs 330 per recorded minute, music 900, voice changing and noise removal 1,000 each, video translation from 2,000 to 10,000 per minute depending on mode and watermark. Top-ups beyond the limit are prepaid from $5, a minute of speech costs from $0.36 without a subscription to $0.17 on Business, and unused balance expires after a year.

If you only need to clean up recorded speech, compare cleanup in Adobe Podcast: the free tier and daily limits are covered separately from voice generation.

Four questions: who owns the voiceover, what happens to recordings of your voice, who is suing ElevenLabs, and how to tell its voice from a human one.

Who owns the voiceover. On paid plans, rights to generated content are yours forever, including after cancellation, because they are granted at the time of generation, not download. Free use is personal only, with attribution in the title, and subscribing later does not grant rights to older free recordings. Anything made with beta features cannot be sold, even by paid subscribers.

Someone else's voice. The vendor provides a professional clone only for your own voice and verifies this by having you read a verification phrase aloud. Celebrity voices are blocked, but the protection does not yet cover every language, as the vendor states itself. It can trace any generated recording to the account that made it.

Your data. All personal data and voice recordings are stored in the US regardless of the user's country, with voice data retained for up to three years after the last login. You can opt out of training on your data in account settings, but it applies only to material uploaded afterward. European storage and a zero-retention mode are sold only to enterprise clients.

Court cases. Audiobook narrators filed a lawsuit in Delaware on Aug 29, 2024 alleging that their recordings were used for training; the case was closed on Nov 6, 2025 and the terms were not disclosed. On May 11, 2026, seven voice actors and journalists filed a BIPA class action in Illinois; Apple, Google, Meta, Microsoft, and others were sued in the same wave. On Aug 17, 2026, ElevenLabs asked the court to dismiss the case; no decision has been made. Two patent suits are nearby, dated Sep 22, 2025 and 01.09.2026.

Tell it from a human. The vendor released a free detector for its voices, but its page says it does not reliably recognize recordings from the flagship v3 model. The C2PA file-signing standard is stated on the safety page, but neither the docs nor the help center explains how to verify it.

Over three and a half years, ElevenLabs grew from a beta with one model into a $11 billion company with a full audio stack that voice actors sue over their voices.