You publish a video in English. And on the same day, the exact same video goes live in Spanish, German, and Hindi. With a translated audio track that aims to preserve the original voice, timing, and expression.
That's AI dubbing.
Just two years ago, this was reserved for big YouTubers with a budget. Mr. Beast dubbed his videos into dozens of languages early on, with dedicated voice actors per country. Expensive, complex, and out of reach for most people. Today you do it with one tool and a click.
I tested Dubbing v2 in my ElevenLabs account three times. No attempt reached a finished dub.
In this guide I'll show you what AI dubbing actually is, when it's worth it, and how to translate your first video step by step. I'll also be honest about where the limits are.
Let's go.
- AI dubbing translates a video's audio track into another language and can preserve elements of the original voice and expression.
- ElevenLabs documents Dubbing v2 Alpha with more than 90 languages and automatic speaker handling. Dubbing v2 is fully automatic and has no editor. Lip-sync is not part of Dubbing.
- Three attempts with a file, my own audio URL, and an official NASA YouTube video failed at loading the source. I therefore do not have a quality assessment based on a finished output.
1. What is AI dubbing?
AI dubbing means an AI takes the spoken audio track of a video and carries it over into another language. And not with some generic computer voice, but with the cloned original voice from the video.
So the difference from traditional dubbing is this:
With normal dubbing, a different voice actor speaks your lines in the target language. With AI dubbing, your own voice stays intact, it just suddenly speaks fluent Spanish or Japanese.
A good dubbing tool handles several steps automatically, one after another.
First, the speech from the video is transcribed, meaning it's turned into text. Then that text is translated into the target language. Next, the system generates a new audio track that aims to preserve the source voice, timing, and expression.
The intended result is a localized audio track. ElevenLabs' current Dubbing documentation does not include adjusted mouth movements in Dubbing v2.
2. When is AI dubbing worth it?
Before we get into the workflow, let's look at where all this effort actually pays off. From my point of view, there are three areas where AI dubbing makes a real difference.
2.1 YouTube and multilingual reach
The most obvious use is YouTube. If you have an English video, it reaches the English-speaking market. Depending on your topic, that might be a few hundred thousand potential viewers.
The moment you also offer the same video in Spanish, you open the door to a multiple of that. YouTube even supports multiple audio tracks per video with its Multi-Language Audio feature. So you upload once and the viewer picks their language. This is exactly where AI dubbing shines, because you produce those audio tracks without a new studio session.
2.2 Film and video localization
The second area is localizing marketing and explainer videos. Imagine a company that wants to run a product video in twelve markets. In the past that meant booking twelve voice actors, coordinating twelve recordings, and paying twelve invoices.
With AI dubbing, you can translate that one original video into several languages and aim for a consistent speaker identity. Whether the result, rights, and review effort work for you needs a complete test run. For marketing, training, and internal communication, that review is part of the workflow.
2.3 E-learning and courses
The third area, and the most interesting one for many online entrepreneurs, is e-learning. If you've produced an online course in English, there's a lot of work baked into it. Re-recording that entire course in another language would be madness.
With AI dubbing, you can translate the lessons and aim to preserve a familiar speaker identity. The extra effort is not automatically minimal. Translation, terminology, rights, and each language's output all need review.
3. ElevenLabs Dubbing v2: my hands-on test
For this test I chose ElevenLabs because Dubbing v2 sits on the same speech platform that made ElevenLabs strong in text-to-speech and voice cloning.
I looked at the dubbing area directly inside my own ElevenLabs account on the Creator plan. The screenshot above shows the real upload screen from my workspace, with a file upload or URL field, language selection, and the Alpha label ElevenLabs still puts on the tool right now.
The current model is called Dubbing v2 and is still tagged Alpha in the dashboard. ElevenLabs documents automatic dubs in more than 90 languages, multiple speakers, and preserved background audio.
My test stopped earlier. The file upload hung without a detected duration, my own audio URL would not load, and an official NASA video on YouTube was rejected. That leaves the cause open. It could have been the input, the service, or my test environment.
One important distinction for your own workflow. According to the official Dubbing documentation, Dubbing v2 is fully automatic. You cannot correct a v2 dub in an editor. For transcript and translation changes, select the older Dubbing v1 model under Advanced when creating a project, then create a Dubbing Studio project. Dubbing Studio is in maintenance mode.
- More than 90 target languages according to ElevenLabs
- Automatic speaker handling and voice transfer according to ElevenLabs
- Background audio is preserved according to ElevenLabs
- Dubs on the Free plan carry a watermark, while paid plans do not
4. AI dubbing in 4 steps: how to translate your first video
The intended workflow has four steps. The interface guides you clearly through it, but in my test the process ended at step 1. So read the following steps as a guide for your own test.
4.1 Step 1: Upload your source video
In the first step, you upload your source video. According to its documentation, Dubbing v2 accepts common audio and video formats as well as public URLs, including YouTube and TikTok. ElevenLabs currently lists a maximum of 2 GB and 180 minutes for an upload.
Then you pick your target languages from the list of more than 90 languages. If you are unsure about the source language, let it be detected automatically. The interface shows the price before you start, based on the model, duration, and target languages.
4.2 Step 2: Set the speaker identity
With Dubbing v2, ElevenLabs handles the speaker identity automatically from the source. Under Advanced, the provider says you can adjust speaker similarity on a scale from 0 to 10. The setting trades voice similarity against natural pronunciation in the target language.
For several people, ElevenLabs documents automatic speaker detection. I did not reach this step in my test.
4.3 Step 3: Review and adjust the translation
This is the important product boundary. Dubbing v2 has no editor. You cannot directly correct a generated v2 dub's transcript or translation in the Dubbing area.
For an editable workflow, choose the older Dubbing v1 model under Advanced when creating the project, then create a Dubbing Studio project. According to the Dubbing Studio documentation, you can edit the transcript, translation, and speaker assignments there. Dubbing Studio is in maintenance mode, though, and no new features are announced.
Even with an editor, human review is essential. Technical terms, product names, brand names, and wordplay need a check by someone who understands the target language.
4.4 Step 4: Export
After a successful Dubbing v2 run, ElevenLabs provides a lossless audio file. Dubbing v2 does not create a video with adjusted mouth movements. In the older Dubbing Studio, ElevenLabs documents exports as AAC, MP3, or WAV and, for projects, separate tracks, AAF, SRT, and CSV among others.
I did not reach this step with Dubbing v2.
5. The limits of AI dubbing
AI dubbing can bundle a lot of production work. Whether it saves time depends on the whole workflow, including the source, translation, review, and export.
The most important limit in my test was that three different sources failed before generation.
For a successful dub, you would need to check technical terms and proper nouns carefully. Especially in technical or industry-specific videos, you must not assume every specialist term transferred correctly.
Without a finished dub, I also could not check the advertised preservation of emotion, expression, and voice similarity.
AI dubbing can be a useful lever for YouTube, e-learning, marketing, and social media. For a specific project, though, you need a complete trial run with your own material.
6. Alternatives to ElevenLabs
ElevenLabs is the best all-round solution for me, but there are of course other providers. I want to briefly introduce two of them so you get the full picture. For a broader overview of AI voice tools, also check my article on the best AI voice generators.
The first is Synthesia. Synthesia actually comes from the AI avatar video corner, but it also offers translation and dubbing features. If you're working with AI avatars instead of real video footage anyway, Synthesia can be a sensible choice.
The second is Rask AI. Rask AI was built as a video translation tool from the ground up and specializes in localization. It also covers many languages and is tailored to the pure dubbing use case.
For a dubbing project with a deadline, fully test each provider with your own material before choosing one. If you want to dig deeper into the ElevenLabs ecosystem, read my ElevenLabs review or take a closer look at the ElevenLabs alternatives.
7. Conclusion: how to take your videos global
AI dubbing can be a useful lever in 2026 when the technical workflow works for your material. My test could not complete that workflow.
There are two things you should remember.
First, you need a complete test run to export before any real project. A nice upload screen and a long feature list are not enough.
Second, review and document the rights to the video, audio, text, voices, and music for each dub.
Since August 2, 2026, Article 50(4) of the EU AI Act has required professional deployers to disclose AI-generated or manipulated audiovisual content that closely resembles an existing person, object, place, entity, or event and would falsely appear authentic or truthful. A cloned voice may qualify. For evidently artistic, creative, satirical, or fictional works, an adapted notice is enough.
If you test Dubbing v2 yourself, start with a short, non-critical test file and check that it actually reaches a result. For text-to-speech and voice cloning, ElevenLabs remains my first choice. I do not derive a dubbing recommendation from three failed attempts, though.
Frequently Asked Questions About AI Dubbing
ElevenLabs documents Dubbing v2 with more than 90 languages.
The list includes English, Spanish, French, Hindi, Japanese, Czech, and Bulgarian. English is listed as both a source and target language. My three attempts failed before generation, so I could not verify the language coverage in practice.
ElevenLabs charges for Dubbing by source-media minute and target language. Its Dubbing v2 API starts at $2.20 per source minute. These API prices exclude taxes, levies, and duties. In the app, the total depends on the model, duration, and target languages, and is shown for confirmation before the dub starts.
Dubbing is available on every plan. Free-plan results carry an automatic watermark that cannot be removed on that plan. My three attempts failed before generation, so I could not check a project price in the app.
Since August 2, 2026, Article 50(4) has applied to professional uses within the Act's scope. Deployers must disclose AI-generated or manipulated image, audio, or video content when it closely resembles an existing person, object, place, entity, or event and would falsely appear authentic or truthful in context. The Act defines that as a deepfake. A cloned voice may qualify. The disclosure must be clear and provided no later than the audience's first exposure.
The Act excludes personal, non-professional use. Its disclosure duty also does not apply where the use is authorised by law to detect, prevent, investigate, or prosecute criminal offences. For evidently artistic, creative, satirical, fictional, or analogous works, disclosure is limited to stating that deepfake content exists in an appropriate way that does not hamper the work's display or enjoyment. For the wider legal context, read the EU AI Act guide.
