Plan short ecommerce video ads for TikTok, TikTok Shop, Reels, and Meta. Turn product facts, available footage, and budget into an angle, timed script, production route, platform checks, and distinct concepts to test.
Ecommerce video ad skills for Claude Code and AI agents: plans and makes short product videos that sell on TikTok, TikTok Shop, Instagram Reels and Meta ads, with AI or real footage. Asks the four questions that change the rules (where it runs, product facts, assets, budget), picks angle, hook, format and on-camera character, writes a second-by-second script, keeps the real product identical in AI shots, applies platform rules (AI labels, TikTok Shop bans on AI voices, slideshows and price mentions, the FTC fake-testimonial rule, music licensing), quotes cost before generating and plans different-concept variants. Use when someone wants a viral TikTok, Reels or TikTok Shop video for a product, ecommerce ad creatives or "10 ads for my product". Hands off to product-video-hooks, tiktok-shop-video, ai-ugc-ad-video, product-photo-to-video, ad-creative-testing and ecommerce-video-edit. NOT for app launch videos, music videos or brand films with nothing to sell.
Run score Not tested
See how to installReview the instructions the agent will receive before you install.
Make short product videos that stop the scroll and sell: TikTok, TikTok Shop, Instagram Reels and Meta ads. This skill is the plan and the rules. The sibling skills do each part.
Cost gate. Nothing is generated before the person has seen an estimate (estimate_cost on the varg MCP, or POST /v2/estimate) and said yes. Quote credits (1 credit = 1 cent). Draft cheap first: stills, kling_v4_flash or seedance_2_fast_preview, 480–720p. Spend on full models only for concepts the person approved.
Agents fail here in both directions: a wall of 15 questions, or invented product facts. Ask exactly these, in one message, then state what you assumed for everything else:
Defaults when unanswered: 9:16 1080×1920, 15–25 s, hook in the first 2 s, burned-in captions, licensed music or none, AI people labelled, 3 different concepts.
An angle = buyer × problem × promise. Mine it from the buyer's own words (reviews, comments, Reddit), not from adjectives. Write one line per angle: "Women 50+ whose knees crack on the stairs → a sleeve that lets them climb without thinking". Then:
Details and casting by category: references/characters.md.
Hook (0–3 s): the product or the result on screen, the promise said out loud and written as text, all three agreeing. TikTok: value in 3 s, hook done by 6 s (90% of ad recall happens in the first 6 s). Meta: brand and message within 5 s gives 1.7× the purchase intent, and a hook in both picture and sound gives 1.5×. The hook must lead into the product. Bank of 60+ openers by type: product-video-hooks.
Format: choose from references/formats.md. Start with what wins most often: unboxing (9.8% hit rate on Meta, yet only 2% of what brands make), hands-on demo (8.1%), problem and fix in one clip (the top TikTok Shop structure), yapper (one person talking to camera, unpolished), street interview, podcast clip, us vs them on ONE attribute.
Structure (15–30 s):
| Time | Beat | On screen |
|---|---|---|
| 0–3 s | Hook | Product or result in frame from the first second, line spoken + written |
| 3–6 s | Bridge | The hook turns into the product: "this is why…" |
| 6–20 s | Demo | The product working in a real setting, shown more than once, 1–3 benefits, one specific number |
| 20–26 s | Proof | A real review, a real result, a count ("2,100 reviews") |
| last 3–4 s | CTA | One action, said and shown ("tap the orange cart") |
Meta's Reels study (Meta + Toluna) behind this order: listing benefits 5.3×, brand on screen under 25% of the runtime 4.8×, product shown more than once 2.7×, native elements such as emoji 2.5×, speech plus music 2.1×, a spoken or on-screen CTA 1.9×. Text on screen: at most 5–10 words per second. Long education ads (90 s to 2 min) can carry big budgets, but the product must appear before the halfway point. More: references/structure.md.
| Rule | TikTok Shop video | TikTok ad | Meta ad |
|---|---|---|---|
| AI voice as the presenter | Banned | Label required | OK (auto "AI info") |
| Realistic AI people | High risk; needs a visible real person | Label (AIGC toggle) or rejected | Auto-labelled |
| Still-image slideshow | Banned (≥3 s moving footage, a person visible) | Weak | Allowed |
| Price or deal said in video | Must match the product page exactly, or a violation | Must match the landing page | Must match |
| Before/after | Typical results only; none for weight | No false-impression before/after | Cosmetics 18+, no weight loss |
| Music | Commercial Music Library | Commercial Music Library | Original audio or Meta Sound Collection |
| Near-duplicate reposts | Suppressed; 5+ low-quality Shop videos in 7 days hide links | Grouped | Grouped as one ad (Andromeda) |
Everywhere: never write a first-person testimonial for an AI person ("I used this for two weeks…"). The FTC rule bans fake testimonials, AI ones included, at about $51,744 per violation. New York has required a conspicuous synthetic-performer disclosure in ads since 9 June 2026. An AI presenter may demo and state facts; results and reviews must be real. Full table with sources: references/platform-rules.md.
Route by what the person has:
| They have | Route | Skill |
|---|---|---|
| Real footage | Recut hook-first, captions, CTA, variants | ecommerce-video-edit |
| Product photos only, no person needed | Product shots: ASMR, unboxing hands, in use, macro | product-photo-to-video |
| Product photos, want a person | AI actor UGC (not for TikTok Shop) | ai-ugc-ad-video |
| A TikTok Shop listing | Shop-compliant script and shot list, real presenter | tiktok-shop-video |
| A winning ad | New hooks on the same body, new concepts | ad-creative-testing |
| A competitor's viral video | Copy the structure, never the footage or actor | remake-reference how-to on varg how to |
| A viral-style hook clip from photos | Backyard-camera fail, torture test, problem → solution as phone footage | varg-product-hook |
| A faceless dropshipping "guide" reel | Product, cost vs price on a laptop, steps, "Comment GUIDE" | varg-dropship-reel |
Product fidelity (non-negotiable): start every shot from the real product photo, never from text. Keep clips at 5 s or less with a slow or locked camera ("no fast spins, that's where products melt"). Spell the label text out in the prompt. Composite the real cutout or logo back onto hero shots. Reject any clip whose silhouette, label text, colour or part count differs from the photo.
Character lock: one identity sheet, the same reference on every shot, one video model per character (no model-hopping between shots), one voice, the exact transcript with the audio.
No video tool? Never ship a Ken Burns slideshow with a robot voice as the ad. Deliver the script, a shot list and a 5-shot phone filming checklist, and label any FFmpeg animatic as a placeholder.
Models on varg (prices only from estimate_cost): stills nano_banana_2_1 / gpt_image_2_5; product in hand bria_fibo_edit_1_5_product_holding; clips seedance_2_5 (most real, ~3 tries per keeper), kling_v4 (cheaper, often right first time), kling_v4_flash for drafts, veo_3_1; presenter with speech in one call creatify_boreal; voice eleven_v4; lip-sync sync_v3. Pipelines and prompts: references/models-and-prompts.md.
ecommerce-video-edit §6).<product>_<angle>_<format>_<hook-id>_<creator>_<len>s_v<N>.mp4.ad-creative-testing.| Read | When |
|---|---|
| references/platform-rules.md | Before any script: TikTok Shop, TikTok ads, Meta, FTC, NY, music, with sources |
| references/formats.md | Choosing the format: 30 named formats, when each works |
| references/characters.md | Casting the person on camera, by category and claim |
| references/structure.md | Second-by-second templates: TikTok Shop 18 s, UGC 15 s, yapper, VSL |
| references/models-and-prompts.md | Pipelines, varg model ids, copyable prompts |
| references/questions.md | The 25 questions sellers ask most, answered |
| references/sources.md | Where every number comes from |
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Plan worthwhile local tasks by checking dependencies, effects, resources, and platform boundaries, then execute the exact approved plan. It can use safe parallel work or suitable Apple-silicon backends when the project exposes them.
Turn a winning TikTok or Meta ad into a structured creative test plan. Identify what made it work, plan distinct concepts, set test budgets and review timing, and use funnel metrics to decide whether to stop, iterate, or scale each ad.
Record deferred engineering decisions and temporary workarounds in a debt registry, with context such as effort, ongoing cost, affected area, and payoff trigger. Entries receive short labels so they can be reviewed or dropped later.
Generate or compare AI agent code for LangChain, EINO, AutoGen, AgentScope, and CrewAI using framework-specific templates, API mappings, conventions, and project structures. Reference files include verification dates to help identify potentially outdated guidance.
Investigate a reproducible bug, test failure, or unexpected page behavior, then make a narrowly scoped fix. The workflow calls for a regression test before changing code, verifies the fix, and reports remaining risks and evidence while avoiding unrelated refactoring or feature work.
Build an AI agent project through six collaborative stages, from requirements and architecture to tools, prompts, code generation, and verification. Start framework-independent, then generate a runnable project for frameworks such as LangChain, EINO, AutoGen, AgentScope, or CrewAI.