Tmall Product Images

Overview

Turn original product photos and optional product notes into a Tmall/Taobao-ready product image set. Preserve product truth and rebuild the ecommerce presentation so the output is useful, differentiated, and suitable for platform listing review.

Use this skill with $gpt-image-2 when generating or editing actual image files.

Workflow

Gather inputs.
- Require at least one original product image.
- Use product notes when supplied: product name, brand authorization, selling points, material, size, SKU, audience, category, certificates, usage scenarios, and forbidden claims.
- If product notes are missing, infer only visible product facts and ask for missing high-risk facts when needed.
Build a product fact sheet.
- Separate confirmed from user, visible from image, reasonable inference, and to confirm.
- Identify the likely category and subcategory from the image and notes. If confidence is low, use the general strategy.
- Do not invent certifications, ingredients, material grades, efficacy, safety claims, brand authorization, test reports, awards, sales data, or category-specific regulated claims.
- For cosmetics, food, health, baby, medical, electrical, or safety-related products, be extra conservative and mark proof-dependent claims as to confirm.
Load references before drafting image outputs.
- Always load references/tmall-image-guidelines.md.
- Load references/category-strategies.md and apply the closest matching category strategy.
- If category confidence is low, use the general strategy and avoid category-specific claims.
Create a dedicated output folder before generating final assets.
- Use a clear folder such as ~/Downloads/tmall-product-images/<short-product-name>-<timestamp>/.
- Use exactly two top-level subfolders by default: 来源/ and 产出/.
- Put original product images and any supplied notes or source files in 来源/.
- Put generated main images and detail images in 产出/.
- Do not write prompts, payload JSON files, or Markdown planning files into the final folder unless the user explicitly asks for traceability/debugging.
Generate or plan a complete image set.
- Main images: default to at least 5 square images.
- Detail images: default to 4-8 modular images when enough product information exists.
- If information is sparse, generate a smaller safe set and clearly list missing facts.
Generate images with $gpt-image-2 when requested.
- Use image editing when original product photos are supplied and product fidelity matters.
- Use text-to-image only for supporting scenes, abstract detail backgrounds, or mockups where exact product fidelity is not required.
- Preserve product shape, proportions, color, material, logo placement, packaging, and distinctive physical details.
- Change ecommerce composition, background, props, text layout, cropping, lighting, and design system enough to avoid copying competitor creative identity.

Default Deliverables

For a full run, create:

01-主图-首图.png: clean hero image.
02-主图-卖点.png: core selling point or category benefit.
03-主图-细节.png: material, structure, texture, function, or craftsmanship.
04-主图-SKU规格.png: colors, sizes, variants, or package set.
05-主图-场景.png: realistic usage or lifestyle scenario.
06-详情-参数.png: factual product parameters when confirmed.
07-详情-卖点拆解.png: modular benefit explanation.
08-详情-使用场景.png: use case image.
09-详情-售后或注意事项.png: only when facts are supplied.

Adjust count by category and available facts. Do not fabricate detail-page claims just to fill images.

Output Folder Rules

The final folder should normally look like:

<run-folder>/
  来源/
    original-01.png
    product-notes.txt
  产出/
    01-主图-首图.png
    02-主图-卖点.png
    03-主图-细节.png
    04-主图-SKU规格.png
    05-主图-场景.png
    06-详情-参数.png

Keep implementation artifacts outside this folder unless requested.

Image Strategy

Prefer platform-safe, product-first images over busy advertising posters.
Use clean product photography for the first image.
Use text overlays sparingly and only for stable, confirmed claims.
Keep text readable on mobile.
Avoid excessive borders, badges, fake discounts, noisy backgrounds, and contact information.
Use white or light neutral backgrounds for white-background/transparent-style assets.
For branded goods, include logos only when the user provides the logo or confirms authorization.
For competitor/source images, preserve product facts but rebuild the commercial presentation.

Compliance Rules

No QR codes, phone numbers, WeChat IDs, external URLs, or off-platform contact prompts.
No watermarks, unrelated store names, platform badges, or competitor marks.
No absolute advertising claims such as 最, 第一, 顶级, 100%, 永久, unless the user supplies proof and explicitly asks for inclusion.
No fake certificates, test reports, patents, awards, factory claims, endorsements, or sales data.
No medical, health, whitening, slimming, child-safety, fireproof, food-contact, electrical safety, or other regulated claims without supplied proof.
If a claim cannot be proven from input, mark it as a missing fact instead of placing it on an image.