Dub and translate a video into other languages

Dubbing replaces the speech in a video with the same words in another language. This guide picks the right level (subtitles, a voice dub or a full video translation), runs a full translation in one go with HeyGen Video Translate, and builds the same result step by step when the script needs control. Every language ends with a check by someone who reads it.

Before you start

• Choose the lowest level the audience accepts. Each level up costs more and has more to check.
• Subtitles first for social clips watched without sound.
• Full translation for a presenter speaking to camera, where a mismatch between lips and voice shows.
• You need a reader for each language. A model can't tell you if a name, a number or the tone is wrong.
• Price grows with duration, and it shows on the button. Trim before you run, and check the real length before a whole campaign.
• Clone only voices you have the rights to.

Way 1: Full translation in one run

HeyGen Video Translate returns translated speech in a voice close to the speaker's, with the lips re-synced. Precision is made for the most demanding work, where lip sync accuracy matters. Speed is faster and costs less, for batches.

Step 1: Trim and load the video

Trim the video to the spoken part with Video Cut (see Video tools): translation is billed by length. Then open Create › Videos, click the Model card, then View all models, search for "HeyGen", choose HeyGen Video Translate Precision (or Speed) and click Use this model. Load the video.

Step 2: Set the language and the speakers

• Output language. Well over 100 options, including regional variants such as French (Canada), Spanish (Mexico) and English (UK), and several variants of Arabic and Urdu. The default is Spanish: set it every time.
• Translate audio only. Translates the voice and leaves the faces untouched. Use it when no face is on screen, or for a voice-over.
• Enable dynamic duration. On by default. Lets the translated speech run longer or shorter than the original, since languages differ in length, and adjusts the video to fit.
• Speaker num. Set the number of speakers for interviews and dialogues, so each voice is separated and translated correctly.

Step 3: Generate and repeat per language

Check the price on the button and generate. Run it once per language, and send each result to a reader of that language before you publish.

Way 2: Step by step, when you need control

Run the steps yourself when the script needs editing (brand names, legal lines, local wording), when you want a specific voice, or when one model fails a language.

Step 1: Transcribe

Run Speech to Text on the video. It returns a plain transcript and a timed SRT file. In the app it can show as Audio to Text (Subtitles).

• Language: set the spoken language (for example en or fr) instead of leaving it on auto-detect. It's more accurate and faster.
• Model size: Tiny is fastest, Large-v3 most accurate. Small or Medium suits most work; use Large-v3 for long or noisy recordings.
• Task: Transcribe keeps the source language. Translate returns English only, so don't use it for other target languages.
• Initial prompt: add product names and unusual words so they're spelled right.
• VAD filter (voice activity detection) skips silence. Leave it on.

Step 2: Translate the script and have it reviewed

Translate the transcript, then have a reader of the target language review it. Keep each line close to the original length, so the new speech fits the same shots.

Step 3: Make the speech

Use a text-to-speech model, or a cloned voice where you have the rights. See Narration and speech. For a dub that keeps the speaker's own voice in one run, ElevenLabs Dubbing translates the speech in a video or audio file into 30 languages with voice cloning, and can detect up to 10 speakers. It gives a voice dub: the lips still move in the original language.

Step 4: Re-sync the lips

Run Sync-3 Lipsync with the original video and the new audio. If the new speech is longer or shorter than the video, choose the sync mode (remap keeps both whole). See Lip sync and talking avatars.

Step 5: Subtitle if needed

Run Scenario Caption Studio with the translated SRT under Subtitles (SRT) to burn in captions with your own timing. It uses the file as it is and doesn't transcribe the audio again.

Keep text out of the picture

Dubbing changes the audio and the mouth, not a sign, a label or a title in the frame. Plan for it when you generate: keep on-screen text out of generated shots, and add it in the edit, one layer per language.

Make it repeatable

• One master, no burned-in text, and a workflow that takes the master and a language as inputs. Each market is one run.
• Store the reviewed transcripts with the outputs, so the next update reuses the approved wording. See Presenter video with avatar.

Common problems

What's next

For subtitles only, see Transcribe and subtitle. To build a presenter who speaks every language from one portrait, see Presenter video with avatar.