Dub a video with AI, lips in sync

A video where someone speaks, a new voice in another language: Sync Lipsync 3 re-times the lips to the new text. The same shot in English and in Spanish, without reshooting. The whole workflow is visible without an account.

Interactive demo

Real workflow, real results. Pan and zoom freely.

Use this workflow

The real workflow: the avatar's video in a Media node, two translations each read by a Voice-over node, and two Video nodes set to Sync Lipsync 3 that receive the video through the video port and the voice through the voice port.

At a glance

What the workflow produces: 2 videos of 12s in 720p, 9:16 · 2 voice-overs
Canvas structure: 8 nodes wired by 6 connections
Models wired in: ElevenLabs v3, Sync Lipsync 3
Cost of one full run: about 208 credits, roughly $2.50

A video shot once, delivered in as many languages as needed: that is what dubbing with lip sync does. The person on screen no longer says the original sentence, they say the translation, and their mouth follows the new voice frame by frame. The face, the lighting, the framing stay those of the shoot.

The translated text is read in a Voice-over node, with a catalog voice or a cloned voice, yours for instance. The Video node set to Sync Lipsync 3 receives the video and the voice, and renders the dubbed shot. The price follows the produced duration, measured on your two files, and shows before you click.

The demo starts from the talking avatar video, in French, and renders it in English then in Spanish. A spot, an internal training video, a product presentation: one shoot, one file per market.

How it works

1
Upload the spoken video
A face seen from the front, well lit, mouth visible, with no hand or microphone in front. Up to two minutes per pass. A shot filmed on a phone works.
2
Write the translation and pick the voice
Each Voice-over node reads a text in one language. Pick a voice close to the person, or clone theirs to keep their timbre from one language to the next.
3
Set how the durations meet
By default, the result stops at the shorter of the two durations. If the voice runs past the shot, choose “loop”: the video repeats under the voice, which works well on a to-camera take.
4
Generate and compare
One node per language, all wired to the same video. The price of each follows its duration. Relaunch only the language you do not like.

Variations and use cases

The same canvas, tuned for different needs: each variation is two or three lines of prompt away.

The spot in five languages

An ad shot in French becomes five files: one Voice-over node per language, one Sync Lipsync 3 node per language, the same video as input. The budget reads on each button before launching.

The translated training video

A trainer filmed to camera, the text translated by a Text node, read by a Voice-over node, and the video dubbed for teams abroad. The edit stays the same.

Fix a sentence without reshooting

Dubbing also serves in the same language: a sentence said wrong, a price that changed, a name to correct. Rewrite the text, relaunch, the mouth follows.

The customer testimonial subtitled, then dubbed

An authentic testimonial wins a foreign audience without losing the person's face: dub it with a close voice, and keep the subtitles for accessibility.

The generated avatar in every language

Start from a video produced by the talking avatar or a UGC ad, and dub it: the synthetic character speaks every language without regenerating the shot.

A photo rather than a video

If you have no starting video, make a photo talk: InfiniTalk invents the face motion from a portrait and a voice.

Common mistakes

A video with several faces

The model syncs one face. With two people on screen, the result is unpredictable. Frame the one who speaks.

A voice much longer than the shot

In cut mode, the result stops with the video and the end of the text disappears. Shorten the translation or switch to loop mode.

A mouth hidden during the take

Hand, microphone, cup: anything passing in front of the mouth breaks the sync on those frames. Keep the face clear when shooting.

Forgetting the music and ambience

The result only contains the voice. Put the music back in the Montage node, otherwise the dubbed version sounds barer than the original.

Recommended models

Frequently asked questions

Do I need to reshoot the video for each language?

No: that is the whole point. A single video, one Voice-over node per language, and the lips follow each translation.

Is the original voice kept?

No, the audio track of the result is the wired voice. To keep music or ambience, put them back in a Montage node over the dubbed video.

Can I dub with my own voice in another language?

Yes: clone your voice in the Voice-over node, write the translation, and the cloned voice reads it. The lips follow.

What video length is accepted?

From one second to two minutes per pass. Beyond that, cut the video in the Montage node and dub the pieces.

How much does dubbing a video cost?

The price follows the produced duration, per second, and shows on the button as soon as the video and the voice are wired. A technical failure is refunded automatically.

Does dubbing work on an AI-generated video?

Yes, as well as on a real shoot. A Wan, Seedance or Veo shot with a frontal face dubs like any video.

Are gestures and expressions kept?

Yes: only the mouth is recomputed. Hands, eyebrows, head movements are those of the original take.

Which voice for a credible dub?

A voice in the same register as the person, or their clone. Listen to the voice alone in the Voice-over node before launching the sync.

Dub a video with AI, lips in sync

Use this workflow

Free account, no card required. The workflow opens pre-filled in your canvas.