Back to Blog
Languages

AI Dubbing Explained: Translate Video Into Other Languages

AI dubbing translates your video into other languages with cloned voices. See how the pipeline works, current limits and prices, and when to review.

8 min read
By OndaLabs Team

AI dubbing replaces the speech in a video with a translated voice track. ElevenLabs describes its dubbing as translating audio and video while keeping each speaker's emotion, timing, tone and characteristics, and retaining the original background audio. The result is a new voice track in another language, mixed with the original sound.

This guide explains the pipeline, compares what the main services publish as of October 2026, lists their limits and prices, and shows how to build a manual dubbing workflow with OndaLabs.

How AI dubbing works: the typical pipeline

Vendor and review sites describe the same basic structure. The steps below follow the sources, and the figures attached to each step come from the services named.

| Step | What happens | What the sources say | |---|---|---| | 1. Transcribe and separate speakers | The speech is converted to text and each speaker is identified | ElevenLabs detects multiple speakers, including overlapping speech, up to 32 unique speakers per file | | 2. Translate | The transcript is translated | Translation is a separate step you can check and edit | | 3. Generate the voice | Text to speech creates the new track, optionally with a cloned voice | ElevenLabs cloning strength runs from 0 to 10, default 7; higher values favour similarity but may carry more of the original accent | | 4. Lip sync (optional) | Mouth movements are adjusted to match the new speech | Lip sync is experimental on YouTube; HeyGen describes it on its product page | | 5. Remix | The new track is mixed with the background audio | ElevenLabs retains the original background audio |

What AI dubbing services offer today

| Service | Languages | Length limits | Lip sync | Notes | |---|---|---|---|---| | ElevenLabs Dubbing | "90+ languages" in ElevenLabs' own words; the v2 table lists 94 rows | Automatic dubbing in the app up to 1 GB and 180 minutes; API up to 3 GB per file; Dubbing Studio up to 1 GB and 45 minutes | Not mentioned on the docs page | Up to 32 speakers; no realtime or live dubbing | | HeyGen video translation | 175+ languages and dialects | 30 minutes per video on Creator and Pro, 60 on Business | Described on the product page | Up to 10 languages at once; "Edit & Review" to proofread the script | | YouTube auto dubbing | 20 languages from English; 29 languages into English | Videos over 120 minutes are not eligible | Experimental, early access for select channels | Dubbed videos are marked "auto-dubbed" in the description |

Figures are as of October 2026. Language counts and length limits differ in meaning across services, so compare them with care.

Limits to know before you dub

  • Length. Each service caps file size, duration or both, as shown above.
  • Preview quality. ElevenLabs says its in-app dubbed video is a preview, mostly 1080p. For production, download the audio and replace the track on the original video.
  • Version status. ElevenLabs' Dubbing Studio is in maintenance mode, with critical bug fixes only, and Dubbing v2 is in Alpha.
  • Editing. YouTube dubs cannot be edited directly. Correct the original-language video and the dubs regenerate.
  • Expressive Speech. YouTube's Expressive Speech mimics pitch and intonation in some languages, but it "won't always be used."

When to review, or use a human

Technical terms, accents and cultural references are where machine translation most often needs a person to check. This research found no measured accuracy rate for any service, so this is advice, not a measurement. Listen to the full track before you publish.

Criticism of AI dubs has focused on quality and on the effect on voice actors. Wikipedia notes that Amazon Prime Video's 2025 test of AI-generated dubs drew criticism on both counts (Dubbing). For high-stakes work, plan a human review step.

Prices: what the published figures say

ElevenLabs prices automatic dubbing in credits per minute: 2,000 with a watermark and 3,000 without. Dubbing Studio costs 5,000 with a watermark and 10,000 without. Free-tier dubs are watermarked, and paid-tier dubs are not. A 10-minute dub without a watermark therefore costs 30,000 credits.

| ElevenLabs plan | Monthly price | Credits per month | |---|---|---| | Free | $0 | 10,000 | | Starter | $6 (standard monthly price) | 30,000 | | Creator | $22 | 121,000 | | Pro | $99 | 600,000 | | Scale | $299 | 1,800,000 | | Business | $990 | 6,000,000 |

HeyGen's Creator plan is $29 a month with 600 credits, Pro is $49 with credits from 1,000, and Business is $149 plus $20 per seat with 1,500 credits. Credits and dollars do not convert evenly across services, so compare plans rather than per-minute prices. The ElevenLabs $6 and $22 figures are standard monthly prices, not promotional ones.

Disclosure and the EU AI Act

Article 50 of the EU AI Act sets transparency duties. Providers of systems that generate synthetic audio must mark outputs in a machine-readable, detectable format. Deployers who generate or manipulate audio or video that could be mistaken for authentic must disclose that. The duty is lighter for evidently artistic, creative, satirical or fictional works (Article 50).

The European Commission's regulatory page, last updated on 3 August 2026, says the AI Act's transparency rules come into effect in August 2026 (European Commission). Whether a particular dub needs a label depends on its facts, so check with a lawyer. YouTube's auto-dubbed label is a platform feature, not a legal finding.

A manual dubbing workflow with OndaLabs

OndaLabs offers text to speech, two-speaker dialogue, speech to text, custom voices, image generation and music generation. It does not offer translation, video upload or lip sync. A manual workflow for a short project looks like this:

  1. Transcribe the original audio with the speech to text tool. Transcription costs 50 credits per minute, so a 10-minute pass costs 500 credits.
  2. Translate the transcript with a tool of your choice. Translation happens outside OndaLabs.
  3. Generate the target-language track with text to speech, at 1 credit per character on Onda Lite or 1.5 on Onda Flash, or use the dialogue tool for two speakers.
  4. If you want the original speaker's voice, create a custom voice from a recording you have permission to use. It costs 1,500 credits. Custom voice slots are 3 on Creator, 10 on Pro and 30 on Business.
  5. Replace the audio track in your video editor. This step also happens outside OndaLabs.

The Free plan at $0 includes 2,000 credits a month. Starter at $9 a month includes 170,000 credits, about 189 minutes of speech, Onda Flash and commercial use.

For the voice-only side of a project, see Chinese text to speech and German text to speech.

Frequently asked questions

How does AI dubbing work?

A typical pipeline transcribes the audio and separates the speakers, translates the text, generates a new voice track, optionally adjusts lip movement, and mixes the result with the background audio. ElevenLabs keeps the background audio and detects up to 32 speakers per file. Lip sync is the least consistent step across services.

How much does AI dubbing cost?

ElevenLabs charges 2,000 credits per minute with a watermark and 3,000 without, so a 10-minute dub without a watermark is 30,000 credits. HeyGen's paid plans start at $29 a month for Creator. Credits and dollars do not convert evenly, so compare plans rather than per-minute prices.

Is AI dubbing accurate enough to publish without review?

This research found no measured accuracy rate, so it cannot say. Check technical terms, accents and cultural references, and listen to the whole track. Criticism of AI dubs has focused on quality, so plan a review step for anything important.

What is the difference between AI dubbing, voice-over and subtitles?

Wikipedia defines dubbing as adding to or replacing the recorded audio of a video during post-production. Subtitles keep the original audio and add text. Choose dubbing when viewers should hear a voice in their language, and subtitles when they should hear the original.

Can AI dubbing keep the original speaker's voice?

ElevenLabs offers cloning with a strength setting from 0 to 10, default 7. Higher values favour similarity but may carry more of the original accent. OndaLabs' custom voices clone from a recording you have permission to use, for 1,500 credits.

Do I have to label AI-dubbed videos?

EU AI Act Article 50 requires disclosure of AI-generated audio or video that could be mistaken for authentic, with a lighter duty for evidently artistic, creative, satirical or fictional works. YouTube marks dubbed videos as "auto-dubbed" in the description. Whether your video needs a label depends on its facts, so check with a lawyer.