Amazon Outdoor Product Lifestyle Video · skill Open Skill công khai

Amazon Outdoor Product Lifestyle Video

Amazon Outdoor Product Lifestyle Video là một Open Skill công khai trên Digen. Mở ra để xem workflow này tạo gì.

@taozi0413

营销广告

0 3 11

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/102982/2026-08-19/wcnq5udc.jpg

# Amazon Scene Showcase

Turn an Amazon product image set into **one finished video ≤15 seconds**.

Keep it simple: **cover → showcase → priced ending**.  
Default audio: **silent** or **BGM only**. Do not force narration or digital-human VO.

**Cover preference (multi-model mix):** do **not** open on a plain white-bg catalog plate — it looks weak after model handoff. Open on the **best in-scene product showcase** instead. White shots are for **identity lock** only (and fallback if no usable scene exists).

This Skill file is Agent-facing English.  
**User-facing replies** follow `user_lang` (language of the first user message).  
**Model prompts** for image/video/music stay **English** (no CJK inside generation prompts).  
When showing tips to non-English users, **translate the English tip below** into `user_lang` — do not paste other-language examples from this file (there are none).

---

## When to use

- Amazon / ASIN / listing product showcase clips
- Uploaded scene (+ white/detail/lifestyle) image sets
- User wants a short silent or BGM product video with price/brand on the end card
- User already has shot clips and only needs the **final mix**

## Do not use / push back

- Long cinematic ads, multi-episode stories, talking-head UGC scripts
- Requiring static cover PNG / end-card PNG as a hard gate (video cover/end is enough)
- Inventing regulated claims not in the listing

---

## Public skill tip (what to ask users for)

On a thin first turn, tell the user what to send.  
Lock `user_lang` from the first message. Show **one** tip in `user_lang` only (translate from English if needed). Never bilingual concat.

### Canonical tip (English — translate when `user_lang` is not English)

```text
For a better Amazon showcase video (≤15s), please send:

1) Product images (3–8, clear):
   - At least 1 strong in-scene / lifestyle shot (product fully visible) → used for the opening
   - Optional: white-bg shot (identity only), detail/feature close-ups
2) Product title or short name (brand + model)
3) Price + currency (e.g. $129.99) — or say “price TBD”
4) Brand name or logo file (if you have one)
5) Optional: marketplace (US/UK/DE/…) · vertical 9:16 (default) or landscape · silent / BGM only

Tips for best results:
- Prefer a sharp scene photo where the product is the hero (not tiny in frame)
- Avoid heavy watermarks, extreme blur, or cropped-off products
- One clear scene beat beats many similar angles
```

### If the user already pasted enough

Do **not** re-ask the full tip. Proceed. Only ask for missing blockers (e.g. no image at all, or no way to infer a short name).

### Skill store blurb (English)

> Upload 3–8 Amazon product photos (include one strong in-scene shot), plus title, price, and brand. Get a ≤15s silent/BGM showcase: scene opening, mid demo, ending with price + brand.

---

## Global hard rules

1. **One final file** ≤15s is the only `completed` deliverable. Shot clips alone = `partial`.
2. **Must call platform mix** (`mix_clips` or equivalent timeline merge). Do not stop after `generate_video` segments. Do not rely on local `ffmpeg` unless Shell is explicitly available.
3. **Non-stop default:** if the user did not ask to pause/confirm, run intake → videos → (optional BGM) → **mix** without waiting.
4. **Layout lock:** product subject sits **lower-center** on cover and ending; ending **top** reserved for price + brand/logo overlays (composite after video; do not bake fragile text into the generative frame).
5. **Cover = best scene hero, not white plate.** Prefer the strongest in-scene product frame for COVER. White/seamless catalog shots are identity references only.
6. **Ref diversity:** BODY and END must not be identical clones of COVER; vary angle, distance, or detail while locking product identity.
7. **Video model contract** (when this runtime requires it):  
   `model=rm3.5-ref` · `gen_mode=reference_to_video` · `generation_type=reference`  
   Forbidden: swapping to unrelated models for this skill’s video