---
name: lip-sync-localization
description: "Take your winning talking-to-camera ad and launch it in any language. Same footage, a native or cloned voice, lips that match. You approve the script and hear the voice before anything is generated."
---

# Lip-sync video localization (Arcads MCP)

Turn one talking-to-camera video into the same video in other languages: same footage, new voice, matching lips. Works for any brand.

The flow is a fixed sequence of client approvals. Never skip a gate, and never regenerate anything on your own: a new take happens only when the user asks for it. After delivery, don't analyze, transcribe or rate the final video; the user gives the feedback.

## Step 1: Get the video

Ask the user to share the video file to translate.

- Check it: 5s to 5 min, .mp4/.mov, max 500 MB, one speaker talking. If there are several speakers or long b-roll parts, tell the user before going further.
- Ask which Arcads brand to file it under if it isn't obvious (`arcads_list_brands`). Create one project for the job with `arcads_create_project` and pass its `projectId` to every call.
- Upload: files over 10 MB go through `arcads_get_upload_url`, then an HTTP PUT of the raw bytes with the same Content-Type, then `arcads_register_video` with the returned `filePath`. Wait until the asset is `generated` with `arcads_get_asset` (waitSeconds) before using it.

## Step 2: Ask the languages

Ask which languages to translate into. Several are fine; run Steps 3 to 6 for each language.

## Step 3: Propose the translated script

- If the client gave their own localized script, use it word for word. Don't translate or reword it again.
- Otherwise, run `arcads_transcribe` (format `srt`) on the video, then write a native adaptation, not a literal translation:
  - Match the register of the original (UGC = casual, spoken words, informal "you" where natural).
  - Match the speaker's gender in agreements.
  - Keep the length close to the original duration. Most languages run 10 to 20% longer than English, so trim where needed.
- Show original and translation side by side, flag the choices you made (tone, cuts), and ask for approval or edits. Move on only once it's approved.

## Step 4: Choose the voice

Ask the client which they prefer:

1. **Clone the original voice:** same voice as the video, but it carries the speaker's accent into the new language. Best for same-language edits.
   - `arcads_clone_voice` with the master video as the sample. Costs 80 credits; the voice expires after 24h. Confirm the client owns the rights to the voice.
2. **Native voice:** `arcads_list_voices` filtered by language and gender (and age if relevant). Present a shortlist described by how each voice sounds; the widget lets them play samples. If they want to see everything, share the full list for that language and gender.
   - Tell them they can also connect their ElevenLabs account in Arcads (Settings → Integrations, Pro plan and above) to use their own voices.
   - Arcads can't design a voice from a text prompt. If asked, say so; the ElevenLabs MCP can design one, then the audio goes to Step 6 via Upload audio.

## Step 5: Text to Speech preview

In both cases, generate the approved script with `arcads_text_to_speech` (runs on ElevenLabs v4) so the client can hear it before any lip sync is spent.

- Share the audio link.
- Compare the audio duration with the video duration. Lip Sync trims the video to the audio length, so if they differ by more than ~1s, say so and offer: use as is (shorter video), slow the audio slightly (`arcads_change_speed_audio`, keep it above ~0.85x), or adjust the script and regenerate.
- The client may test several voices. Generate a new preview only for the voice they name.

## Step 6: Lip sync and deliver

Once the preview is approved:

- `arcads_lipsync_video` with `videoPath` = the master asset id and `audioPath` = the approved TTS asset id. Leave `script` and `voiceId` empty.
- Poll with `arcads_get_asset` (waitSeconds 40, repeat). It usually takes 2 to 3 minutes.
- Deliver exactly one final video per language: download it and send the file, plus the Arcads project link. Name assets `<subject>_<lang>_lipsync_v1`.

## Costs to mention if asked

- Text to Speech: about 8 credits per preview.
- Lip Sync: billed per second of speech (about 50 credits per second; a 14s video cost 720).
- Voice clone: 80 credits.
