AI Product Ad Studio · Open Skill 공개 스킬

AI Product Ad Studio

One product photo in. A finished 15–45 second ad out — cut for the platform you sell on. It does not invent a different product. It keeps your item exactly as it is and builds the angles a real shoot would get: hero, macro texture, real context, detail, scale, motion, benefit, variants, logo card. WHAT IT DOES • Analyses your photo and locks the product's shape, colour and label as unchangeable • Builds a brand kit first — three colours, one surface, one lighting language — so the ad looks art-directed, not assembled • Generates every frame with i2i against your own photo, so the product is identical in every shot • Adds commercial motion: light sweeps, slow orbits, a pour or a drop — subtle, premium, never cheap whips • Writes a voiceover under 35 words for 30 seconds: hook, one benefit, one proof, brand • Designs a music bed that lands a soft hit on the logo card • Delivers the main cut AND a 15-second cut, because advertisers always need both WHAT MAKES IT GOOD • The hero frame is treated as the whole ad — if it is weak, nothing downstream saves it • At least two macro shots, because texture is what makes a product look expensive • One shot establishes real scale, which is the buyer's first question • It remembers your product, so the next variant does not restart from zero GOOD FOR Amazon and Shopify listing videos, TikTok Shop, Etsy, launch teasers, UGC-style ads, premium brand films, and A/B palette variants for testing. NOT FOR Invented claims, health or medical benefits, before/after body content, or recreating another brand's campaign. Original work only. Upload one photo and say what it is — it will build the angles, the motion, the voice and the cut.

@omegabubble

Product

0 0 3

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/4267104/2026-10-19/n5gt7t5b.png

---

allowed-tools:
  - generate_image
  - generate_video
  - generate_tts
  - generate_music
  - recognize_image
  - recognize_video
  - split_image
  - mix_clips
  - create_document
  - update_document
  - memory_whoami
  - memory_query
  - memory_store
---

# AI Product Ad Studio

One product photo in, one finished 15–45 second ad out. This skill keeps the product **exactly
as it is**, builds the angles a real shoot would get, adds motion, a voiceover, music and brand
text, and cuts it for the platform the user is selling on.

## When to use

- The user has **product photos** and wants an ad.
- The user is selling on **Amazon / Shopify / TikTok Shop / Etsy** and needs a listing video.
- The user wants a **UGC-style** ad, a **premium brand film**, or a **launch teaser**.
- The user wants **several variants at once** to test which converts.
- The user has one photo and needs a full set of angles and a video from it.

Do **not** use this for: inventing product claims, medical or health benefits, before/after body
content, or recreating a real brand's campaign — see Notes.

## Steps

### 0. Intake — the product must not change

Call `write_todos`: intake → product analysis → brand kit → shot plan → stills → motion →
voice/music → cut → deliver.

Ask at most **three** questions:

1. **What is it, and what does it do?** — plain description, no marketing.
2. **Where will this run?** — 9:16 (TikTok/Reels/Shorts), 1:1 (feed), or 16:9 (site/YouTube).
3. **What is the one feeling?** — premium, playful, clean, urgent, cozy.

If the user only uploads a photo, **infer and proceed**. State what you inferred in one line.

### 1. Product analysis — lock what may not change

```
recognize_image(image_url=<url>, prompt='Describe this product for a video shoot: exact form and
proportions, colour and finish, material, the label/branding position, any text on it, and what
lighting would flatter it. Do not invent ingredients, claims or certifications.')
```

Then persist the constraints so later runs stay consistent:

```
memory_store(key='product_<slug>', value='<name>; <form>; <colour/finish>; <label position>;
<what must never change>; <brand palette>')
```

**The product's shape, colour and label are locked.** Everything else is yours to art-direct.

### 2. Brand kit — decide before you generate

Three colours, two type sizes, one surface material. State them in one line and save:

```
create_document(title='<product> — brand kit', content='palette: ... | type: ... | surface:
... | lighting: ... | motion: ...')
```

Rules: one hero colour + one neutral + one accent. **Never more than three.** A surface (stone,
linen, acrylic, water, brushed metal) gives the ad its physical credibility.

### 3. Shot plan — 8 to 12 shots for 30 seconds

| # | Shot | Length | Function |
|---|---|---|---|
| 1 | ヒーロー | 3s | Product alone, ideal light. The money shot |
| 2 | 素材感 | 3s | Macro: texture, edge, finish |
| 3 | 使用文脈 | 4s | The product in its real environment |
| 4 | ディテール | 3s | A mechanism, a seam, a control, a pour |
| 5 | スケール | 3s | In a hand, on a table — how big is it really |
| 6 | 動き | 4s | The product's own motion, or motion around it |
| 7 | ベネフィット | 4s | What it does, shown not claimed |
| 8 | パレット | 3s | Colour/flavour/size variants together |
| 9 | ロゴカード | 3s | Brand text on a clean field |

**9 shots ≈ 30 seconds.** For a 15-second cut, keep 1/2/4/6/9 only.

Craft rules:

- **Shot 1 is the ad.** If the hero frame is weak, nothing downstream saves it.
- **At least two macro shots.** Texture is what makes a product look expensive.
- **Show scale.** One shot with a hand or a known object answers the buyer's first question.
- **The last card is quiet.** Brand, product name, one line. No busy background.

### 4. Stills — the product must be identical in every frame

```
generate_image(
  model='i2i',
  input_urls=[<the user's product photo>],
  prompt='the exact same product, unchanged shape, colour and label, <new angle/lighting/
  surface>, <p