AI video dubbing without a studio: localize video at scale

Dubbing a video traditionally means a studio, voice actors, and an editor painstakingly matching new audio to footage that was never shot for it — the result is the familiar out-of-sync look where the mouth and the words drift apart. AI video dubbing fixes the visual half of that problem programmatically: it re-syncs the mouth in your footage to a new audio track so the dub looks natural instead of pasted on.

What the API actually does

It's worth being precise about scope, because dubbing is often sold as magic. Our API provides the lip-sync layer — and only that layer. Here's the honest division of labor:

• You bring (or generate) the translated audio. The API does not translate your script and does not clone or synthesize a voice. You supply the audio track in the target language, however you produced it.
• The API re-syncs the mouth. Given your video and that new audio, it aligns the speaker's mouth movements to the new track via Video Lip Sync, so the dubbed result reads as natural rather than obviously overdubbed.

For multilingual lip-sync talking characters specifically, the HappyHorse model handles lip-sync across languages, which makes it a good fit when the on-screen subject is a character that needs to speak several languages convincingly. Either way, the API's job is the sync — the translation and the voice are inputs you provide.

The API call

The request uses the same job lifecycle as everything else on the platform:

• Create the job. POST /api/ai/jobs with model_id: "video-lip-sync", a video_urls array pointing at your source footage, and an audio_urls array pointing at your translated track.
• Poll for the result. Call GET /api/ai/jobs/{id} until it reports done, then read the output URL — that's your re-synced, dubbed clip.

Because the contract matches the rest of the catalog, you reuse the same poll loop you already have for generation jobs.

Use cases

• Localize a course. Record an online course once, then produce a per-language version by re-syncing each lesson to its translated audio — no reshoot, no re-recording the visuals.
• Localize a presenter. Take a single talking-head video and ship it in every market you sell in, with the mouth matching each language.
• Localize a product video. Roll out one product explainer across regions, swapping only the audio track per language while the footage stays the same.

The economics are the point: the expensive, hard-to-repeat work — shooting the footage — happens once, and localization becomes a per-language audio swap.

Pairing dubbing with other layers

Dubbing rarely ships alone. Many teams pair it with on-screen text for accessibility and silent autoplay — add burned-in subtitles / ai subtitle generator so the dubbed clip works with the sound off and meets accessibility expectations. And if you're producing the source footage from scratch rather than re-syncing real video, look at talking avatars to generate a presenter, then dub that presenter into each language.

Start dubbing

Bring your footage and your translated audio, send a lip-sync job, poll for the result, store the URL. Start from the AI video dubbing landing to see the full workflow, and read the request contract in the AI API docs to wire it into your localization pipeline.