agentleFS
vargHQvargHQ/skills/varg-ad-localize/SKILL.md

varg-ad-localize

Skill23 starsChanged 8 months agoMITView on GitHub

Turns one video ad into versions for 8+ TikTok Shop and Meta markets on varg: transcreates the script per market (hook, slang, claims, units, local price and currency from an offer card, each line fitted to its time slot), voices it with ElevenLabs v4 in the same voice, lip-syncs only shots where a face talks for more than 1.5 s (Sync v3), rebuilds captions and on-screen text in the target language instead of inpainting them, re-cuts every shot on the new language's words, and ends with a native-review checklist. Use when a seller asks to translate an ad, dub a video ad, localize for TikTok Shop markets (UK, DE, FR, IT, ES, ID, TH, VN, MY, PH, JP, MX, BR, SA), make a Spanish or German version of a winner, or sell cross-border. Not for subtitles on a long video (use a captions skill) or dubbing films and podcasts (see video-skills).

  • Reads credentials
  • Sends data out

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Write numbers as words when the voice should read them a certain way ("dos dólares con diez").

    line 59

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Localize a video ad for other markets

TikTok Shop sells in the US, UK and Ireland, ten EU markets (DE, FR, IT, ES since 2025; AT, BE, NL, PL since June 2026), seven in Southeast Asia and Japan (ID, TH, VN, MY, PH, SG, JP), and SA, AE, MX, BR, AU (third-party lists, Oct 2026; Shopifreaks). A winner in one market is the cheapest ad in the next one, if it is transcreated, not translated: literal captions are the most common reason localized ads flop.

1. Per market: an offer card and a brief

MARKET: es-MX · LANGUAGE: Spanish (Mexico, not Spain) · CURRENCY: MXN, price from the MX listing
OFFER: 1 pair, $349 MXN, free shipping 3-5 days · CTA: "Toca la bolsita naranja" (TikTok Shop tag)
APPROVED CLAIMS: cut-resistant level 5 (certificate) · FORBIDDEN: "irrompible", medical claims
NEVER SAY: literal English idioms, "you guys", prices in USD
FONT/SCRIPT: Latin · DIRECTION: LTR

Variants matter: es-MX ≠ es-ES, pt-BR, Gulf Arabic for SA (right-to-left), Taglish for PH, Thai and Vietnamese need fonts with diacritics, Japanese needs CJK line breaks. EU prices include VAT. Health and beauty claims differ by market (Japan's PMD Act, Indonesia's BPOM and halal wording): when unsure, cut the claim and flag it for native review.

2. Transcreate the script

Give the model the source transcript with word timings, the offer card and these rules: rewrite the hook (first 3 s) for the market, keep every line within ±10% of its source syllables so it fits its slot, swap units and currency, keep brand and product names, drop claims not on the card, keep the anchor phrases that start each beat (the cut points). Spanish runs 15–25% longer than English: shorten, don't speed the voice past 1.1×.

3. Voice

eleven_v4 with the same voice as the source (or a consented clone of the real speaker), one call for the whole script:

curl -s -X POST https://api.varg.ai/v2/speech -H "Authorization: Bearer $VARG_API_KEY" -H "Content-Type: application/json" \
  -d '{"model":"eleven_v4","voice":"Brian","text":"<transcreated script>"}'

Write numbers as words when the voice should read them a certain way ("dos dólares con diez"); write "varg.ai" as "varg dot A I".

4. Lip-sync only where it shows

Faceless ads (voiceover over b-roll, screens, hands) need no lip-sync. For shots where a face talks for more than 1.5 s, run sync_v3 on that shot with its slice of the new audio; skip profiles, hands over the mouth and fast motion (sync.so tips).

5. Re-cut, re-caption, rebuild text

  • Word timings of the new voice (local Whisper with language=es, or groq_whisper_large_v3_turbo).
  • Re-cut on the new words. For the guide-reel format: node guide-edl.mjs words.json clips --lang es (built-in en, es; any language with --anchors anchors.json, the same 11 keys).
  • Captions in the target language: captions.mjs (prices like "$2 .10" are glued back automatically).
  • On-screen text (prices on cards, banners, UI): rebuild it from the source project or re-render the overlay. Inpainting only holds on small static text over plain backgrounds; re-shoot screenshots in the local language where the UI is the point.

6. Native review before it runs

A native speaker checks: the hook sounds local, no literal idioms, prices and currency match the local listing, claims are on the card, the CTA matches the platform's local wording, and the AI label is set. Then ai-ad-compliance-check for the market.

Tested

2026-10-28: the 59 s cut-resistant gloves reel → Spanish. Transcreated script, eleven_v4 "Brian" (8 credits), local Whisper word timings, guide-edl.mjs --lang es re-cut all 14 shots on the Spanish words, captions.mjs kept "$2.10" and "$19.99" whole and highlighted GANANCIA, ADSPARK. No lip-sync needed (faceless). Total generation cost: 8 credits. https://s3.varg.ai/files/194b94d0-58ff-4e8b-91de-4f1959d04013/file_f4e15b912ef1.mp4 · The laptop UI stays in English; re-screenshot it in Spanish for a real launch.

Related skills

varg-dropship-reel (guide-edl.mjs, captions.mjs, assemble.mjs), ai-ad-compliance-check, varg-ad-variants (localize the winner, then make variants per market), varg-ai.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.