AI Video Translator

An AI video translator takes a video in one language and gives it another: the words translated, a new voice track in the target language, and ideally a speaker whose mouth matches the new words. Full dubbing services do all of that from one upload. Mage does the voice and the lips with separate tools you control, which suits short clips such as ads, intros and character videos better than hour-long recordings.

Here is the split. Mage does not transcribe your video or translate it on its own, and it does not add subtitles. You, a translator or your AI assistant write the script in the new language. Eleven Speech then reads it in more than 90 languages, and Lipsync animates a still of the speaker, a clear frame from your video or a saved character, so their lips follow the new audio, up to 15 seconds per clip.

The result is a new talking clip made from that frame, not your original footage with its mouth replaced, so the speaker’s gestures and the camera moves of the original are not kept. For narration with no visible speaker, you need no lip sync at all: lay the new voice over the original picture in your editor. For text inside ad images, the Ad Localizer skill replaces it in place.

Automatic on Mage: Eleven Speech turns a written script into speech in the language you write it in, with pacing and emotion you can direct with tags such as [whispers] or [laughs]. Lipsync turns one image and up to 15 seconds of audio into a video of the subject speaking it, with natural expressions and subtle head movement. The Ad Localizer skill, in your assistant, transcribes the text in a static ad, has your assistant translate it, and replaces each line in its font, size and position after you approve the copy.

Not automatic: transcribing the original speech, translating spoken dialogue, matching the new line’s timing to the original, keeping the original speaker’s own voice, separating the music bed from the voice, and subtitles. Mage has no speech-to-text, no automatic dubbing and no caption tool. Lipsync also makes a new clip from a still, so it does not keep the original footage’s motion, cuts or other people in the shot.

• Automatic: the voice from your script, in more than 90 languages.
• Automatic: lips synced to the new audio, from one image.
• Automatic: on-image ad text replaced after you approve it.
• Yours: the transcript, the translation and the final review.
• Not available: subtitles, or new lips on the original footage.

A dubbing script is not a subtitle file. It has to be spoken in roughly the time the original took, so translate for meaning and length together. German and French often run longer than English, and Japanese and Chinese often shorter; shorten a line rather than speed up the voice. Keep brand and product names as they are, and say numbers, dates and prices the way a local speaker would say them.

If you work in Claude, ChatGPT or another assistant with Mage connected, the assistant can draft the translation and show it to you line by line before anything is voiced. That translation is the assistant’s work, not a Mage feature, so have a native speaker read it for anything customer-facing. Then split the script at natural pauses into pieces of 15 seconds or less, the length of one Lipsync clip.

Use Eleven Speech for most translations. Pick one of its 16 voices that fits the speaker’s age and energy, write the script in the target language, and direct the read with audio tags in square brackets. The MP3 is exactly as long as the script, up to 5,000 characters per request, so there is no silence to trim. The AI text to speech guide covers voices and tags in depth.

Eleven Speech speaks only in its preset voices. When the new track should sound like the original speaker, Seed Audio can guide a voice from up to three audio references, using only recordings of a voice you have permission to use. It renders fixed lengths of 5, 10, 30, 60 or 120 seconds, so check the timing and the language on a short line before a full script. The AI voice cloning guide covers matching a voice in more detail.

Cloning your own voice for a dub is ordinary practice: YouTube, for one, lists it among the uses of AI that need no altered-content disclosure. Cloning someone else’s voice is a different matter, and Mage does not allow imitating a real person without their consent.

When a person speaks to camera, take the clearest front-facing frame of them from the video, mouth visible and closed or relaxed, and use it as the subject in Lipsync. Add the translated audio, up to 15 seconds, and pick a model: Plum is the default, and Blueberry 2 animates the exact image you give it, which keeps the look of the original frame. Each clip is a new performance from that still, with the camera holding steady on the speaker.

For a speaker who appears across a series, save them as a character so the face stays identical in every language. Then line the clips up in Storyboard, which joins up to six into one MP4 with their sound. The AI talking photo guide covers choosing a frame that animates well.

When nobody speaks on camera, as in a narrated explainer, a product demo or a screen recording, skip lip sync. Generate the new narration and replace the audio track in your video editor. Mage does not mix a new voice into an existing video for you.

• Pick a front-facing frame with the mouth visible.
• Keep each audio piece to 15 seconds or less.
• Save a recurring speaker as a character.
• Replace the audio in an editor when no one is on camera.

Static ads carry their message in text: the headline, the button, the offer. The Ad Localizer skill handles those in your assistant. It transcribes every line, translates for meaning and tone, adjusts line breaks, number formats and right-to-left layouts, and shows you the copy for approval. Then it replaces each line in the image with GPT Image 2.5 Flare, keeping the font style, the layout, the product and the logo, and leaves the brand name and packaging text as they are. It does not translate video voiceovers. Find it on the Mage skills page, and see the AI ad generator for the rest of the ad workflow.