Dubbing

The Dubbing node translates spoken audio, or a whole video, into another language and keeps each speaker's voice. Wire in a recording or a video, or paste a public link, and choose the target language. You get the dubbed audio, and for a video, the same pictures with the translated dialogue.

When to use it

• Localize podcast episodes or full videos for audiences in other languages.
• Dub a video from its link, without downloading it first.
• Make versions of training or educational content in several languages.
• Reuse interview audio in another language.
• Translate the voiceover of a marketing video.

Quick start

Inputs and outputs

When both inputs are wired, the video wins. A Source Link overrides both inputs.

A Video URL node wired into Video gives Dubbing the video that it downloaded, never its page link. The Video URL node downloads the video first, under its own content rights box and by its own rules: a long video from a site whose length is read first asks for a part, or the whole video. The downloaded video counts as an uploaded file, so the 500 MB limit applies. To dub a whole long video from its link without downloading it, paste the link into Source Link instead. A Dubbing node with a Source Link reads nothing that is wired into it, so a run downloads nothing for a Video URL node wired to it.

The file decides the mode, not the input. An audio-only file on the Video input is dubbed as audio. A video on the Audio input is treated as a request for audio. With a Source Link, the mode follows what the link points at.

Settings

Content rights for a Source Link

A Source Link is an import of that link, so the settings panel shows the content rights box under it. Tick the box to confirm the declaration for that link. See Content rights for what the declaration says.

• A run stops until the box is ticked for the link.
• Each run confirms the link again.
• The box stays ticked only for the link that you ticked it for, and only until the declaration changes.
• A frozen node is never run, so it does not ask.
• A Dubbing node inside a sub-workflow has its box in the sub-workflow's own editor. When you run it from the workflow that holds it, the node sends its link without a confirmation, and a deployment that requires one refuses the run.

Limits

Credits

Dubbing costs 44 credits per started minute of the dubbed span, with a minimum of 1 minute. The dubbed span is the Start and End window when you set one, and the whole source otherwise.

Nodaro measures the length of the source before the run starts, for a wired file and for a Source Link alike. A link to a post on a supported site is read from the post, and a direct media link from the file. A span longer than 30 minutes is refused before anything is charged.

Sometimes the length cannot be read before the run. When no Start and End window is set, the run then holds the price of 30 minutes, and your balance must cover it. When the dub is delivered, the run is charged for the span that was dubbed, as ElevenLabs reads it from the media or as the delivered file shows it. The rest of the hold is refunded. In that case, the 30-minute limit is checked once the media is read.

A run from the node's Run button, a workflow run and a call from the API, the SDK or MCP are all charged this way.

Dub into Hebrew

Dubbing into Hebrew uses ElevenLabs Dubbing v2 and costs 1,210 credits per started minute. It works differently from the other target languages:

• Upload the source first. Wire an uploaded audio or video. A Source Link is not supported for Hebrew.
• Trim before you dub. The Start and End window is not supported. Cut the part you need with Trim Video or Trim Audio first.
• Voices and background stay. The dub keeps the speakers' voices and the background sound, and detects the speakers automatically.
• Fewer options. Native voice, Number of Speakers, Drop background audio, Profanity filter and Target Accent are refused for Hebrew before the run starts.

A Hebrew video dub keeps the source resolution.

Clone the speaker or use a native voice

By default, the dub keeps each speaker's identity. A Hebrew speaker dubbed into English sounds like the same person speaking English, with their own accent.

Turn on Native voice when the dub should sound like a native speaker of the target language. A similar voice from the Voice Library then speaks for each speaker.

Tips

• Set the source language when you know it. Auto-detect is reliable, but an explicit language avoids mistakes with accented or mixed-language speech.
• Set the number of speakers. Automatic detection can merge two speakers or split one speaker in two.
• Use clean sources. Moderate background noise is fine, but music under the speech lowers the quality.
• Test on a short window first. Dub one minute with Start and End, check the language quality, then dub the rest.
• Dub long sources in parts. Each window is its own run and its own per-minute charge.
• Match the lips. A dubbed video keeps the original mouth movements. To match the lips to the new language, wire the dubbed audio and the video into Lip Sync.

Troubleshooting

The run is refused because the source is too long. The dubbed span is over 30 minutes. Set Start (sec) and End (sec) to dub one part at a time.

A dub with Native voice fails. Native voice needs a free voice slot, and the run fails when none is free. Turn Native voice off to dub with each speaker's own cloned voice.

A run stops at the Dubbing node. The content rights box under the Source Link is not ticked for that link. Tick the box and run again.

A Hebrew dub is refused. One of the options that Hebrew does not support is set, or a Source Link is used. Clear the option, or upload the source and wire it in.

From the API

POST /v1/dubbing takes exactly one of audioUrl, videoUrl or sourceUrl, plus targetLanguage and the optional sourceLanguage, numSpeakers, disableVoiceCloning, dropBackgroundAudio, startTime, endTime, highestResolution, useProfanityFilter and targetAccent. A sourceUrl is fetched by the dubbing service for you, so it is an import of that link. Send the confirmationId confirmed for it, from POST /v1/content-rights/confirm. Without one, a deployment that requires a confirmation refuses the dub with 403 before the job is created. A key or an agent whose account holds a standing authorization needs none. See Content rights. A video dub returns videoUrl and the dubbed audioUrl. A span over 30 minutes is refused with status 413. The MCP tool is dubbing.

To quote a dub, ask POST /v1/credits/model-costs or GET /v1/credits/model-cost for the identifier elevenlabs-dubbing:<N>u, or elevenlabs-dubbing-v2:<N>u for Hebrew. N is the started minutes of the span, from 1 to 30. The answer is the credits the dub reserves. The run applies your account's pricing to the whole span once, so N times the per-minute price can come to more. A fractional N, or more than 30 minutes, lands in missing. See Exact prices of runs priced per unit. See Voice and media.