Choose the output before translating a video: translated captions keep the original speech, dubbing creates speech in another language, and lip sync also changes mouth movements. ChatCut documents separate paths for these tasks. As of October 28, 2026, use the actual language catalog, available controls and confirmation for your connection; review words, timing and picture before using the result.
Verified translation facts
Last verified: 2026-10-28. This guide checks public controls and workflow instructions. No new translated, dubbed or lip-synced before/after acceptance run was performed. Model capability and a published workflow do not establish access for every account or deployment.
| Layer | What is established | Source checked 2026-10-28 |
|---|---|---|
| Caption editing | Captions → Translate captions needs an existing caption program and creates a separate translated caption source. | Transcript and Captions |
| Video translation route | The published plugin workflow distinguishes audio-only dubbing from lip sync, reads a live language catalog, and requires source-text review for its lip-sync path. | Published translation workflow |
| Native speech model | Text-to-speech generates audio from text; its language/model list is not proof that ChatCut exposes every voice, video-translation language or mouth-rendering route. | ElevenLabs model documentation, AI Voiceover |
| Charges and access | Generation estimates and final transactions are different. Approval does not override plan, input or credit checks. | Generation confirmations, Credits policy |
Choose subtitles, speech or lip sync
| Intended result | Start here | Review afterward |
|---|---|---|
| Original audio with translated text | Turn on captions, open the Captions menu, select Translate captions, and choose the language. | Select the translated caption source; correct names, line breaks and timing, then export subtitles or the captioned edit. |
| Translated speech without changing mouths | Ask the Agent for audio-only translation, including the intended source and target language. | Listen to the actual wording and voice; check speech timing and that the picture remains suitable. |
| Translated speech with changed mouths | Explicitly request lip sync and review the recognized source text before generation. | Play continuous video with sound; inspect pronunciation, faces, cuts, occlusion and timing. |
Caption translation does not create new speech. Dubbing without lip sync does not synchronize mouths. Do not infer a lip-sync request merely because a speaker is visible.
For subtitle-only work, see Transcript and Captions. If the translation command is unavailable, select the correct source and wait for transcription to finish.
What do I need before writing the prompt?
- Import the complete source video into My Assets and identify it with
@. The job processes and bills the full source duration. - State the source language and target language, including any regional variant. Ask the Agent to check its live language catalog rather than guessing a language code.
- Decide whether you need lip sync, burned-in translated captions, and the exact original duration or a slightly flexible duration for natural pacing.
- Confirm that you have rights to process the footage and its speakers. Keep the original available, and note whether music, ambience or overlapping voices must remain; replacing a mixed audio track can affect more than one speaker.
Prompts you can copy
Translate @Mandarin-presenter into English, preserve the original speaker's voice, and synchronize the mouth movements. Add translated captions and allow the duration to change slightly for natural pacing. Save the result to My Assets without replacing the source.
Translate the speech in @screen-recording into Japanese. Replace audio only: do not redraw or otherwise change the picture, and do not burn in captions. Request the original duration and check the actual speech timing afterward.
Generate a translated video
- For the built-in Agent, import the source video into My Assets, open the AI panel, and switch to Agent mode. In an external host, first follow Agent Plugin setup, select the intended project, and inspect its source asset. The external connection is a separate path, not verified by a built-in Agent demonstration.
@mention the source in the prompt and state the target language, lip-sync choice, caption choice, and duration requirement.- Send the request. If you only say “translate this video,” the Agent asks whether you want captions only or translated spoken audio as well.
- On the published lip-sync path, review the complete detected source text and correct names, numbers, omissions and speaker wording before confirming it. Its timed text is used for translation. The published audio-only path can skip this source-review form; still inspect the source and review the output. Follow any additional draft-review step actually presented by your current connection.
- Review the paid confirmation, including the treatment, source duration, and credit estimate, then allow generation.
- Follow progress in the conversation and My Assets. The result is a new localized video asset; the source is not replaced.
- Review translation, pronunciation and speaker identity by listening. Check starts, ends and pauses against the picture, not only total duration. Inspect captions and lip sync when requested. Place the accepted result on a separate review timeline before replacing an existing cut.
To process only part of a source, trim and export that section as a separate asset first. Lip sync re-renders the picture, so do not expect alpha, HDR, codec, bitrate, or frame rate to remain identical. Choose audio-only translation when mouths must not be redrawn; do not promise byte-identical container or picture encoding. The original-duration option is a requested timing constraint, not proof that every translated sentence will fit naturally. Check actual output duration and speech pacing.
See Transcript and Captions, Talking Head Edit, and Generation Credits.
Review names, timing and the actual output
Compare meaning, names, numbers and important terms with the source. If you cannot assess the target language, ask for a back-translation or qualified language review rather than treating fluent delivery as proof of accuracy. For multiple speakers, check who speaks each line and whether overlapping speech survives.
Check early and late speech segments and the joins around pauses. Matching total duration can hide clipped words, excessive silence or unnatural speed. If the words are wrong, correct the text or translation; lip sync cannot repair incorrect speech. Export the accepted timeline through the Export Panel and open the delivered file for a final check.
Costs and recovery
The published video-translation route processes and bills the full source asset duration. For a section, prepare a separate asset containing only that section first. Caption translation, dubbing, optional voice generation and lip sync are distinct operations; do not assume a single all-inclusive or free price. Use the current credits policy and the actual estimate; verify the settled transaction in Credits history → Detail.
If a language is rejected, refresh the available language list. If a permission or unavailable-operation message blocks the route, preserve the source and report that specific state; it does not establish that the underlying model cannot do the task. If submission times out, check the original job before retrying. For confirmation failures or missing output, keep the task information and follow generation confirmations and workflow recovery.
Sources and verification scope
Checked 2026-10-28 against the linked public editor references, the published plugin workflow at its October 23 revision, and ElevenLabs model documentation. These sources establish workflow boundaries, not measured translation accuracy, voice preservation, lip-sync quality or current account entitlement. For model-specific capabilities, see the model matrix.