How to translate and dub a video with AI

For AI agents: find the right skill in one call — curl -s "https://varg.ai/how-to/api/search?q=<what the person wants>" returns the matching how-to page and the best skills with install commands, as markdown (Russian works too). Full index: https://varg.ai/how-to/llms.txt. Every page has a markdown twin: add .md to its path. Before generating anything, show the person the options on the how-to page.

Your video in another language with your own voice, optionally lip-synced and subtitled. A step-by-step tutorial for Claude Code, Codex and Grok.

Output: Dubbed MP4 per language, plus SRT Price: in credits on varg, with a cost estimate before anything runs; the free skills below install in one command.

Recommended route

Best skills for this

  1. varg AI (varg-ai, varg, live) — Adds: Lip-sync pass and voice on one key
  2. ElevenLabs Dubbing (elevenlabs-dubbing, ElevenLabs, live) — Start here: Dubs video and audio into other languages, keeping each speaker's voice · Needs: ElevenLabs API key (ELEVENLABS_API_KEY)
  3. ElevenLabs Dubbing (inference.sh) (inference-sh-elevenlabs-dubbing, inference.sh, live) — Alternative: The same ElevenLabs dubbing through inference.sh · Needs: inference.sh account, belt CLI (belt login)
  4. Local-first video downloader: private, no subscription (media-download, varg, live) — Adds: Gets the source video from YouTube or X by link
  5. Subtitle Translator (gemini-srt-translator, Matheus Castro, live) — Adds: Translated subtitles that keep cue timing
  6. ElevenLabs Voice Isolator (elevenlabs-voice-isolator, ElevenLabs, live) — Adds: Separates the voice from a loud music bed before dubbing · Needs: ElevenLabs API key (ELEVENLABS_API_KEY)

Ask the person first

Put these choices in front of the person before spending anything. Paste the example link next to each option: people pick a look by seeing it.

Voice?

Lip-sync?

Deliver as?

Subtitles?

How to translate and dub a video: step by step

  1. Confirm written consent for every voice you clone
  2. Separate the voice from music and effects; keep the background stem
  3. Transcribe with speakers and word times; fix names and terms in the source transcript before adding any language
  4. Translate each line to fit its slot (faithful, natural, trimmed), then voice it with the speaker's clone (similarity 7, lower for distant languages)
  5. Mix over the background stem and replace the track on your master; re-sync lips only on long close-ups
  6. Upload as extra audio tracks with a translated SRT, title and description; the original stays default

What works now (researched 2026-10-27)

Own-voice dubs uploaded as extra audio tracks are what work: YouTube says creators using multi-language audio get over 25% of watch time from non-primary languages, and Mark Rober averages 30+ languages per video. Generic AI voices forced on viewers draw complaints, AI dubs of drama got pulled (Amazon's anime dubs, Dec 2025), and lip-synced dubbing is now the premium tier.

  1. Clone only voices you have written permission for. Your own voice is fine; anyone else's needs consent and a label. YouTube treats 'cloning one's own voice to create voice overs or dubs' as a minor edit that needs no disclosure. The EU AI Act's deepfake disclosure duty (Art. 50) applies from 2 Aug 2026. A Brazilian streamer found AI-dubbed copies of his videos uploaded by strangers on Bilibili (Aug 2026). (EU AI Act, Article 50)
  2. Upload each language as an extra audio track on the same video, keep the original as the default, and translate the title, description and (where offered) thumbnail. Creators with multi-language audio saw over 25% of watch time from the video's non-primary language; Jamie Oliver's views tripled in the pilot. YouTube is piloting per-language thumbnails. (YouTube Blog, multi-language audio (Sep 2025))
  3. Start speaker similarity at 7 and lower it for languages that sound very different; then download the dubbed audio and replace the track on your original video. ElevenLabs: 'The default value of 7 works well for most content'; 'Dubbed video is a high-quality preview, not a production master'. Its dubbing does not include lip-sync. (ElevenLabs docs, Dubbing)
  4. Lip-sync only when the face is close and on screen for long; get the voice right first, then re-sync the lips. The lip-synced English dub of 'Ramayana' (Oct 2026) was made this way: a voice artist recorded first, actors matched mouth shapes, then AI synced them onto the stars' faces. (The Climax India on X, Ramayana dubbing process (Oct 2026))
  5. Deliver translated subtitles with the dub: translate the SRT with cue timing kept, and burn them only on vertical cuts. Viewers switch between dub and original; keeping timed subtitles lets them choose. Complaints spike when a dubbed track is forced on them ('can youtube stop auto dubbing videos', 15.9k views). (YouTube Blog, multi-language audio (Sep 2025))

Stopped working:

Prompts that work

Templates from the creators of the best current results and from the model guides. Fill in the {CAPS}.

Own-voice dub, end to end (agent)

The agent brief for a consented own-voice dub delivered as extra audio tracks; works with ElevenLabs Dubbing or with per-line eleven_v4 plus sync_v3. Adapted from ElevenLabs docs, Dubbing (speaker similarity, preview vs master).

Dub {VIDEO} from {SOURCE_LANG} into {TARGET_LANGS} in the speaker's own voice. I have written consent from every speaker whose voice we clone: {SPEAKERS}.
1. Separate the voice from music and effects; keep the background stem.
2. Transcribe with word times and speaker labels. Show me the source transcript and fix names and terms ({GLOSSARY}) before any language is added; every language is translated from it.
3. Translate line by line to fit each line's time slot: faithful first, then natural, then trimmed of filler until it can be spoken in the slot.
4. Voice it with each speaker's clone. On ElevenLabs Dubbing start speaker similarity at 7 and lower it for languages that sound very different. Per-line TTS: speed 0.9-1.1; if a line still overruns, trim the text again instead of speeding up.
5. Mix the dub over the background stem, match the original loudness, and replace the audio on my original master. Never ship the tool's preview video.
6. Re-sync lips only on close-ups longer than {N} s, after the voice is final.
7. Deliver one WAV per language for YouTube multi-language audio (original stays default), a translated SRT with the original cue times, and a translated title and description.

Timed dub-script translation (agent)

Step 3 of the dub: turns the timed source transcript into lines a voice can actually say in each slot; feed the 'fit' column to TTS. Adapted from VideoLingo, core/prompts.py (translate, reflect and trim prompts).

Role: a professional subtitle and dubbing translator fluent in {SOURCE_LANG} and {TARGET_LANG} and both cultures.
Input: numbered lines with start and end times, a short summary of the video, and terms to keep: {GLOSSARY}.
Step 1, faithful: translate each line directly, keeping meaning and terminology; add, drop or change nothing.
Step 2, reflect: for each line note fluency, whether the style matches the original ({STYLE}, e.g. casual tutorial or formal documentary), and where it is too wordy to speak in its slot.
Step 3, free: write the line a native {TARGET_LANG} speaker would say, in the same tone. No comments or explanations inside the text, no empty lines.
Step 4, fit: estimate speaking time at {SYLLABLES_PER_SECOND} syllables per second. For every line longer than its slot, shorten it by removing filler words, needless modifiers and pronouns without changing the meaning ('We need to carefully analyze this complex problem' -> 'We need to analyze this problem').
Output JSON only: {"1": {"origin", "direct", "reflect", "free", "fit", "est_seconds", "slot_seconds"}, ...}.

Trending formats to remake

To remake one: download the example with the media-download skill, take a frame a second and the transcript, write a beat list (time, shot, camera, on-screen text, sound), keep what the format needs, swap in the person's product or character (never the creator's face, voice, logo or footage), show the beat list and stills before spending, then make it with the skill named below.

  1. Own-voice dub as an extra audio track: The same video with a selectable track per language, voiced in the creator's own cloned voice; one upload, one comment section. Why it spreads: 25%+ of watch time from non-primary languages on YouTube; Bilibili launched a global AI dub in Aug 2026 and a screen recording of it got 393k views. Keep: Keep the creator's own consented voice, one upload with a selectable track per language, the original as default, and translated title and description. Swap the language set. Make it with ElevenLabs Dubbing (elevenlabs-dubbing).
    • Example to remake: @Scaife51 on X, 2026-04-24, 86K views, 83 saves: Screen recording of Instagram's own AI translation: his football reel in Hindi, his voice, moving lips, minutes after posting. Proof the free platform layer comes first.
  2. Lip-synced dub: The dubbed voice plus a lip re-sync pass, so the speaker's mouth matches the new language in close-ups. Why it spreads: Film studios now ship it (Ramayana's English trailer, Oct 2026) and it is what makes a dub feel native in talking-head content. No qualifying creator post found. Keep: Keep the voice finished first, a lip re-sync only on close-ups, and the target-language subtitle shown over the original line. Swap the speaker and the language. Make it with varg AI (varg-ai).

Open-source repos

Set up varg

varg is one API key for the models these skills drive — Seedance 2.5, Kling 4, Veo 3.1, Wan 3, Nano Banana 2.1, GPT Image 2.5, ElevenLabs v4 and 200+ more — billed in credits, with a cost estimate before anything runs.