複数の独立したローカル作業を安全性と依存関係を確認して実行計画にまとめ、対応環境では並列化やAppleシリコン向け処理の選択を支援します。
- ビルドやテストの独立した処理を並列化したいとき
- 複数段階の作業計画を安全に実行したいとき
- Appleシリコン向け処理の候補を確認したいとき
商品動画の冒頭を引きつける短い映像として企画・制作します。既存の人気広告の構成を参考にしたり、商品写真から使用場面を生成したりできます。
Makes the first 5-15 seconds of a product video, the part that stops the scroll: either finds a hook that already went viral for this kind of product (TikTok Creative Center, ad spy tools, Reels search) and remakes the idea, or generates a realistic UGC-style clip of the product in use with Seedance 2.5 on varg from the product's own photos (a backyard-camera fail, a torture test, problem then solution, a satisfying close-up, a gift reaction), or the whole "The problem → Vs → The solution" music ad. Use when someone selling a product (dropshipping, Shopify, TikTok Shop, Amazon) asks for a viral hook, a scroll-stopper, a "TikTok made me buy it" clip, product b-roll that looks like real footage, or "a video of my product in use" made from product photos. Not for the full guide reel with the laptop and pen (use varg-dropship-reel, which calls this skill for its hook), talking-head UGC ads with an actor reading a script (find one with video-skills), or studio packshot spins (use varg-ai).
実行スコア 未検証
インストール方法を見る商品動画の冒頭5〜15秒にあたる、視聴者の目を引く短い映像を企画・制作します。同じ種類の商品で反応のよかった広告を調べ、そのアイデアを自社商品向けに組み直す方法と、商品写真からUGC風の使用シーンを生成する方法があります。耐久テスト、問題から解決への展開、満足感のある接写、贈り物へのリアクションなどを扱います。構成によっては「問題→比較→解決」を音楽に合わせた短い広告全体として作る手順もあります。生成後は商品形状や動作、手や動物の描写を確認します。
Shopify、TikTok Shop、Amazonなどで商品を販売し、商品を使う様子が伝わる短い映像や、動画の冒頭に置くフックを作りたいときに向いています。商品写真しかなく、実際の使用場面を見せる映像が必要な場合にも使えます。
生成にはVARG_API_KEYとcurl、またはvarg MCP接続が必要です。商品写真を用意し、他者の映像を使う場合は書面で許可を得てください。実在の顧客や出来事としてAI映像を扱わず、媒体が求めるAI表示を付けます。
この紹介文は、公開されている SKILL.md をもとに AI(OpenAI Codex)が作成しました。正確な仕様は下の原文を確認してください。
インストールする前に、エージェントに与えられる指示の中身を確認できます。
The hook is the first 5-15 seconds of a product reel: real-looking footage of the product doing the one thing that makes people stop. In the dropshipping reels this skill was built from (@tradelle, Sept-Oct 2026), the hook carried the video: a snake biting a leg gaiter, a worker getting sand and grinder sparks thrown at his face mask, a cat's claw lit up by an LED clipper. The voiceover over the hook is a plain product line ("These are the leg guards I wear when I'm outdoors"). Two more reels (a cat runner, drink-test nail stickers) show the other use: the hook stretched into the whole 15-22 s ad, "The problem → Vs → The solution", set to music with no voice (section 3).
Generation runs on varg: the varg-ai skill (REST at https://api.varg.ai/v2) or the varg MCP server (https://mcp.varg.ai/mcp, tools call_tool + get_job). Every call returns a job; poll GET /v2/jobs/{id} until completed, the file is at output.outputs[0].url.
Also collect: 1-3 product photos (a clean packshot plus one in-use photo if there is one), the product name, and the one benefit to show.
| Where | What to look at |
|---|---|
| TikTok Creative Center → Top Ads / Top Products | Filter by your country and category; sort by CTR. Note the first 3 seconds of the top 5 ads. |
| Kalodata, FastMoss | TikTok Shop products with rising sales and the creator videos that sold them. |
| Minea, PiPiADS | Ad spy tools: the ad, how long it has been running, engagement. An ad running 30+ days is profitable. |
| Instagram / TikTok search | tiktok made me buy it, amazon finds, backyard camera caught this, <product> test, "where can I buy this" in comments. |
| Product research tools (Tradelle, Sell The Trend, AutoDS) | The product page often embeds the video ads that sell it. |
Save the 2-3 best references with the media-download skill (vdl <url>) and break each one down: every shot with its time, the text on screen, the voiceover word for word, the structure. The fastest way is one video_analyze call on varg (~5 credits, MCP analyze_video); for exact frames, a 1-frame-per-second contact sheet with FFmpeg plus Whisper word timings. Commands, the table to fill and what to look for: references/breakdown.md. End with the idea in one line ("snake strikes the gaiter in a swamp, the gaiter holds"); you remake the idea, not the clip.
Every prompt has five blocks (the order matters for Seedance 2.5):
| Archetype | What happens | Prompt core (blocks 2-4) |
|---|---|---|
| Backyard camera caught this | Security-cam angle: something goes wrong, the product saves the day | "Fixed security-camera view from a house corner, slightly wide lens, timestamp-free. A man in overalls clears tall grass with a leaf blower; a large snake lunges at his leg and bites the <product from @image1> on his shin; he jumps back unhurt." |
| Torture test | The product takes abuse and holds | "Front-facing phone selfie video by a worker in a garage. He wears the <product from @image1>. A friend throws a handful of sand at his face, then holds an angle grinder near him so sparks spray across the visor; he doesn't flinch." |
| Problem → solution | 2-3 s of the annoying problem, then the product fixing it | "Shot 1 (0-3s): backyard camera view, an orange cat climbs the fence and escapes while its owner chases it. Shot 2 (3-8s): handheld, the same cat happily rolls around the lawn inside the <product from @image1>." |
| Satisfying close-up | Macro of the product doing its job | "Extreme close-up, hands hold a fluffy cat's paw; the <product from @image1> switches on its LED, lights up the claw so the pink quick shows, snips just the tip; the clipping drops into the guard." |
| Gift reaction | Someone unwraps it, puts it on, reacts | "Living room, Father's Day: a dad opens a gift box, pulls out the <product from @image1>, puts it on and laughs; his son films on a phone." |
| Before / after | Same spot, two states | "Shot 1: a grimy oven door, a hand sprays the <product from @image1>. Shot 2: same framing, the hand wipes once and the glass is spotless." |
For a hook with a person who must stay the same across several clips, make a character sheet first (varg-character-design).
Two of the reels this skill was built from aren't hooks for a longer video. They are the entire ad: a cat-runner reel (@rollymewofficial, 14 s, 160k likes) and a German drink-test nail-sticker reel (NailSense, 22 s). Both use the same skeleton:
| Time | Text on screen | Picture |
|---|---|---|
| 0-5 s | The problem 🥺 / Das Problem 🥺 | The pain, shot as real life: a cat escaping over the fence on a backyard camera; a man leaning over a woman's drink at a club bar |
| 5-9 s | (same text) | The stakes, on a security-cam angle with an on-screen timestamp and camera name ("CAM 04 · HINTERER FLUR · 17.10.2026 02:47"): the cat gone, the man walking her down a back corridor |
| 9-10 s | Vs... | A hard cut, sometimes with the product already coming into frame |
| 10-12 s | The solution 😍 / Die Lösung 😍 | The product revealed in a clean, warm setting (a vanity with a candle; an autumn lawn) |
| 12-16 s | One feature line (Drink-Test als Sticker 💖) | Hands use it, step by step |
| 16-20 s | One benefit line (Süß & kinderleicht 🥰) | The proof moment: the star sticker turns from pink to green in the drink; the cat rolls safely through a store |
| 20-22 s | Link in der Bio für deine 🛡 | The product held up in the problem setting |
varg-dropship-reel's assemble.mjs (overlays), with the music as the only soundtrack.Upload the product photos (POST /v2/files, MCP create_upload → PUT to proxy_url), then:
curl -s -X POST https://api.varg.ai/v2/video \
-H "Authorization: Bearer $VARG_API_KEY" -H "Content-Type: application/json" \
-d '{
"model": "seedance_2_5",
"prompt": "Vertical UGC product clip shot handheld on an iPhone in a cozy living room, natural window light, slight hand shake. Extreme close-up: a person'\''s hands hold a fluffy orange cat'\''s front paw on a grey blanket. They bring in the pet nail clipper from @image1 (exact same product, same white and pastel-blue colors, same LED light at the blade). The LED switches on and lights up the claw so the pink quick is visible through it; the clipper snips the very tip of the claw cleanly, the clipping drops into the transparent guard. The cat stays calm. Real footage look, no music, no text, no subtitles.",
"files": [{ "url": "PRODUCT_PACKSHOT_URL" }],
"aspect_ratio": "9:16",
"resolution": "720p",
"duration": 6
}'
files is the first frame on most models, but seedance_2_5 (≤9 images), kling_v4 (≤7) and wan_3 (≤10) take it as a reference and build the scene around it. Name each one in the prompt (@image1, @image2).POST /v2/estimate (same body) or MCP estimate_cost. A 6 s 720p seedance_2_5 clip cost 231 credits in our test (Oct 2026). Drafts: wan_3 at about a quarter of that, then the keeper on seedance_2_5.audio off when a voiceover goes on top. Turn it on ("audio": true) for a hook that plays with its own sound (a scream, a snap, a crash), as in "Backyard camera caught this".@image2, and add "do not change the product's shape, color or size".Tested 2026-10-28, four more hooks from one packshot each, all on seedance_2_5: a backyard security camera with a solar flood light (night IR, the light flares on), a torture test with cut-resistant gloves (knife across the glove, grabbing the blade, a mandoline), a dog version of the nail clipper, and a Father's Day gift reaction with a neck massager. All read as phone or CCTV footage on the first take; the full reels are in varg-dropship-reel. A 21 s "The problem → Vs → The solution" ad for car seat gap fillers (2 kling_v4 dashcam clips + 2 seedance_2_5 product clips, PNG text overlays, a mureka_9_5 music bed): https://s3.varg.ai/files/194b94d0-58ff-4e8b-91de-4f1959d04013/file_a2aa0853b683.mp4. Its reveal clip drifted from long foam wedges to boxier blocks; pass a second product angle as @image2 for products with an unusual shape.
Tested 2026-10-27: an LED pet nail clipper packshot (gpt_image_2_5) + the prompt above on seedance_2_5 at 6 s 720p gave a usable handheld close-up on the first try: the product stayed recognizable, the LED lit up, and it looked like phone footage.
varg-dropship-reel (the full 60-second guide reel this hook opens), varg-ai (models, uploads, prices), media-download (save reference clips), varg-character-design (a person who stays the same across clips), video-skills (other ad formats).
まだレビューはありません。使ってみた感想をお寄せください。
ログインなしで書き込めます。使ってみた感想、質問、動かなかった環境などを気軽にどうぞ。
まだコメントはありません。最初のコメントを書いてみませんか。
複数の独立したローカル作業を安全性と依存関係を確認して実行計画にまとめ、対応環境では並列化やAppleシリコン向け処理の選択を支援します。
TikTokやMeta広告の勝ちパターンを分解し、異なる切り口のテスト案や予算、評価時期を整理して、数値から継続・修正・停止を判断するスキルです。
将来の読み手が理由を知りたくなる未決定や暫定対応を、技術的負債として登録し、後から項目ごとに整理できるスキルです。
LangChain、EINO、AutoGen、AgentScope、CrewAI向けのコード雛形やAPI対応、プロジェクト構成を参照し、バージョン確認を踏まえてAIエージェント開発を支援します。
エラーやテスト失敗、画面の異常などを再現して原因を特定し、テストで確認しながら影響を絞った修正を進めるデバッグ用スキルです。
対話を通じてAI Agentの要件整理から設計、ツール定義、コード生成、動作確認まで6段階で進め、実行可能なプロジェクトを作ります。