---
name: cross-border-ecommerce-detail-page-prompts
description: Generate delimiter-separated standalone JSON AI image prompt templates for cross-border ecommerce product detail pages aimed at English-speaking shoppers. Use when the user mentions 详情页提示词, 电商主图提示词, listing图片提示词, 产品详情页模版, 跨境电商图片生成, hero image prompt, e-commerce detail page, 卖点图, 主图, 场景图, Amazon, Shopify, eBay, Etsy, or asks to create prompts for product listing/detail page images. Supports free-form product info, design style, user-defined image task lists, repeated page attributes with different layouts, platform/aspect ratio, text density, optional forbidden on-image text, and 【【【【 group splitting.
---

# Cross-Border Ecommerce Detail Page Prompt Skill

Generate AI image prompts for a single cross-border ecommerce product detail page. You output `N` JSON objects (one per image task), separated by `【【【【`. Each image is free to have its own visual mood — Hero dramatic, Spec quiet, Lifestyle scenic — and is only loosely tied together by a single shared **brand anchor** (one primary color + one headline font).

The skill focuses on **what each image should communicate** (page attribute) and **how it should be laid out** (composition + copy). The product itself is never described in detail — refer to it briefly as "the product" or by its short name.

## ⚠️ MANDATORY DELIMITER CONTRACT — READ FIRST

This rule overrides every other formatting instinct. It is not a suggestion.

- If you output `N` JSON objects, you MUST output exactly `N - 1` standalone `【【【【` delimiter lines.
- Every `【【【【` sits on its own line, with a blank line above and below. The delimiter is the 4 characters `【【【【` — not 3, not 5, not `[[[[`, not `---`.
- Do NOT wrap output in a JSON array `[...]` or an outer object `{...}`.
- Do NOT put `【【【【` before the first object or after the last object.
- Do NOT add any preface, summary, explanation, Markdown heading, or code fence.

Correct shape (2 objects → 1 delimiter):

```
{ ...first object... }

【【【【

{ ...second object... }
```

Wrong shapes — never emit any of these:

```
[ {...}, {...} ]              ← outer array, FORBIDDEN
{...}{...}                    ← no delimiter, FORBIDDEN
{...} 【【【【 {...}             ← delimiter not on its own line, FORBIDDEN
【【【【\n{...}\n【【【【           ← leading/trailing delimiter, FORBIDDEN
```

## Input

Accept free-form input. Do not require a fixed table. **Never ask the user clarifying questions** — if a field is missing, fall back to the default.

- `product_name`: short product name in any language. Used only as a short reference in the prompts. If missing, infer a simple English category name.
- `product_info`: optional short selling points / specs / scenes. Used to inform headlines and layout choices, never to describe the product's appearance.
- `product_image`: optional reference image the user attaches for the downstream image model. **Its presence or absence does not change the prompts** — `visual_presentation` never describes the product's appearance either way.
- `image_task_list`: list of image tasks the user wants. Do not deduplicate.
- `design_style`: visual style keyword. Default: `modern minimal ecommerce`.
- `platform_aspect_ratio`: e.g. `Amazon 1:1`, `Shopify 16:9`, `4:5 mobile`. Default: `1:1 square ecommerce listing image`.
- `text_density`: `normal` / `simple` / `no_text`. Chinese inputs map as follows:
  - `正常` / `一般` / `标准` → `normal`
  - `少` / `简单` / `简洁` / `一点点` / `少量` / `少一点` → `simple`
  - `无文字` / `没有文字` / `不要文字` / `纯图` / `无字` → `no_text`
- `forbidden_text`: optional single string or list of exact words, phrases, names, or symbols that must not appear visibly anywhere in any generated image. Recognize free-form labels including `禁止图片中出现的文字`, `禁止图片里出现的文字`, `禁止文字`, `禁用文字`, and `不要出现的文字`. For example, `禁止图片中出现的文字：hello、SALE` maps to `["hello", "SALE"]`. Default: an empty list.

**Zero-question rule:** if the user provides only a product name (or any single field), do not ask follow-up questions. Immediately apply all defaults — default 10 modules, default style, default ratio, default text density `normal` — and produce the full output.

Free-form parsing rules:

- Each line, comma-separated task, numbered task, or colon-expanded layout is one independent image task.
- Repeated page attributes are allowed and must be output as separate JSON objects (e.g. three different "卖点图" layouts → three objects).
- `1 / 2 / 3` is ordering, not "number of selling points", unless the user explicitly says "2 selling points / 三大卖点".
- If the user gives only a layout name, infer the closest page attribute.
- **If the user does not provide `image_task_list`, output the Default 10 Modules below.**
- If the user provides a task list, output exactly that many objects. Do not auto-fill to 10.

## Default 10 Modules (used only when no task list is provided)

1. `Hero Image (主图/首图)`
2. `Key Selling Points (核心卖点图)`
3. `Product Showcase (产品展示图)`
4. `Lifestyle Shot (模特/场景图)`
5. `Detail Close-up (细节特写图)`
6. `Comparison (对比优势图)`
7. `Spec Infographic (参数信息图)`
8. `Usage Scenarios (使用场景图)`
9. `Social Proof (信任背书图)`
10. `Call to Action (促销行动图)`

## Page Attribute Mapping

- `主图 / hero / first image` → `Hero Image (主图/首图)`
- `卖点 / selling points / benefits` → `Key Selling Points (核心卖点图)`
- `单卖点 / strongest benefit` → `Single Benefit Highlight (单卖点强化图)`
- `功能 / how it works` → `Function Explanation (功能说明图)`
- `效果 / result` → `Effect Demonstration (功效/效果图)`
- `痛点 / problem solution` → `Problem Solution (痛点解决图)`
- `before after / 前后对比` → `Before And After Comparison (使用前后对比图)`
- `尺寸 / size / dimension` → `Size Ratio Infographic (尺寸比例图)`
- `参数 / specs` → `Specification Infographic (参数规格图)`
- `材质 / material / texture` → `Material Detail (材质细节图)`
- `安装 / assembly` → `Installation Guide (安装组装图)`
- `步骤 / how to use` → `Usage Steps (使用步骤图)`
- `操作 / control / adjust` → `Operation Demo (操作演示图)`
- `模特 / model / wearing` → `Model Display Shot (模特展示图)`
- `真人使用 / lifestyle use` → `Real User Lifestyle Shot (真人使用图)`

## Brand Anchor (Soft Consistency Across All N Objects)

The visual system is **not** locked across all `N` images. Each image can have its own background, accent colors, light direction, and atmosphere — driven by the `page_attribute × layout × design_style` combination. A Hero image can be dramatic; a Spec Infographic can be quiet and clean; a Lifestyle Shot can borrow scene colors. The goal is rhythm, not uniformity.

Only one thing is locked across all `N` objects: a single `brand_anchor` string in the form `Primary #XXXXXX + [Font]`, where:

- `Primary` is one brand color hex chosen from the `design_style` direction. It is the recurring anchor — used at least as accent, CTA, or emphasis in every image, but never required as the dominant page color.
- `[Font]` is a single headline font from the whitelist: `Montserrat`, `Inter`, `DM Sans`, `Playfair Display`, `Raleway`, `Bebas Neue`.

`brand_anchor` must be byte-for-byte identical across all `N` objects. Everything else (secondary colors, backgrounds, atmosphere, accents beyond the brand color) is decided per image by the layout and page attribute.

## Text Density Rules

- `normal`: emit a clear `main_title`. The image may include the title plus short benefit labels or icon notes.
- `simple`: emit a short `main_title`. The image uses one main headline and at most a few short labels.
- `no_text`: `main_title` = `""`. The image must contain no on-image text in any language.

When `no_text` conflicts with a size / spec / certification / social proof image, keep `no_text`: convey the idea with composition, icons without letters, arrows, scale, and lighting only.

## Forbidden On-Image Text Rules

Treat `forbidden_text` as an optional global visible-content constraint independent of `text_density`.

- Parse one value, comma-separated values, Chinese-list-delimited values, or a line-by-line list. Trim surrounding whitespace and remove only exact duplicates.
- Prohibit every entry and its recognizable case, spacing, or punctuation variants from headlines, labels, captions, badges, logos, watermarks, packaging, props, signs, and backgrounds.
- Never use a forbidden entry or recognizable variant in `main_title` or any suggested supporting copy. Rewrite or remove conflicting copy while preserving the page attribute.
- For `normal` or `simple`, include the complete list in every `visual_presentation` using this final sentence: `Use English Latin-script text only, and do not render these forbidden strings or recognizable case, spacing, or punctuation variants anywhere in the image: "[item 1]", "[item 2]".`
- JSON-escape quotes, backslashes, and control characters inside forbidden entries. Preserve the exact entries; quoted non-English forbidden entries are the only exception to the English-only output rule.
- For `no_text`, keep the stricter `No on-image text in any language.` rule and do not enumerate the list because all visible text is already prohibited.
- Do not add a new output field. Enforce this input through copy selection and `visual_presentation` so the seven-field schema remains unchanged.

## Compliance

- Do not invent certifications, ratings, reviews, awards, sales numbers, or medical claims unless the user provides them.
- Comparison images with no real data must use generic visual contrast (e.g. `ordinary bulky alternative`), not invented numbers like `3x faster`.
- Consumer-facing on-image text is English Latin script only (in `no_text` mode, no text at all).

## JSON Schema

Each object must include exactly these 7 fields, in this order:

```json
{
  "slide_number": 1,
  "page_attribute": "Key Selling Points (核心卖点图)",
  "layout": "Five short benefit callouts surrounding a centered product",
  "main_title": "Why Pros Choose This",
  "text_density": "normal",
  "brand_anchor": "Primary #1F3A5F + Montserrat",
  "visual_presentation": "Anchored to the #1F3A5F brand color, the wireless headphone sits centered with five compact callouts radiating outward. Generous whitespace, soft top-down studio light, headline left-aligned in Montserrat at the upper edge. English Latin-script text only."
}
```

Field rules:

- `slide_number` follows output order, starting from 1.
- `page_attribute` uses bilingual English + Chinese labels from the Page Attribute Mapping.
- `layout` is one natural English sentence covering both **what** the composition is and **how** it is laid out (no separate composition/logic fields).
- `main_title` is an idiomatic English ecommerce headline that excludes every `forbidden_text` entry and its recognizable variants; empty string for `no_text`.
- `text_density` is one of `normal` / `simple` / `no_text`.
- `brand_anchor` is byte-for-byte identical across all `N` objects, in the form `Primary #XXXXXX + [Font]`.
- `visual_presentation` is a short 1–2 sentence English description of the image. See the writing rules below.

## visual_presentation Writing Rules

Write 1–2 short, natural English sentences. No fixed step order, no required opening clause, no minimum sentence count. The goal is a quick, evocative directive — not a detailed spec.

What to include (organically, in any order that flows):

- A reference to the brand color (the `Primary` hex in `brand_anchor`), positioned as anchor / accent / emphasis somewhere in the image. The rest of the palette is free.
- Where the product sits in the frame and roughly how large.
- The defining composition idea (e.g. "five callouts radiating outward", "two halves split diagonally", "product silhouette on a clean stage").
- The atmosphere or light direction if it matters for this page attribute.

What to **never** include:

- Product appearance details — color, material, finish, buttons, ports, seams, texture, shape, proportions, structural elements. Refer to the product only by its category or short name (`the wireless headphone`, `the insulated tumbler`, `the product`). This rule holds whether or not the user provided a reference image.
- A fixed opening clause like "Maintaining the brand palette of…" — write naturally instead.
- Long descriptions of typography sizes, alignment values, or padding numbers — those belong in `text_attributes` of richer schemas, not here.

End-of-text language rule:

- For `normal` / `simple` without `forbidden_text`: end with `English Latin-script text only.` (or weave the same constraint into the sentence naturally).
- For `normal` / `simple` with `forbidden_text`: make the final sentence the exact combined language-and-prohibition instruction defined in Forbidden On-Image Text Rules, keeping the complete `visual_presentation` within 1–2 sentences.
- For `no_text`: end with `No on-image text in any language.` (or weave it in).

## Silent Preflight Check

Before emitting, silently confirm:

- Object count = user's task count, or 10 if no task list was provided.
- Repeated page attributes were not merged.
- `brand_anchor` is byte-identical across all `N` objects.
- Every `visual_presentation` is 1–2 short English sentences and references the brand color.
- No `visual_presentation` describes the product's color, material, buttons, ports, seams, finish, or structural detail.
- `no_text` objects have an empty `main_title`.
- No `main_title` or suggested supporting copy contains a `forbidden_text` entry or recognizable case, spacing, or punctuation variant.
- When `forbidden_text` is non-empty and `text_density` is `normal` or `simple`, every `visual_presentation` contains the complete escaped prohibition list.
- No invented certifications, ratings, reviews, or unprovided data.
- **Delimiter count check (HARD GATE):** count JSON objects as `N`, count `【【【【` lines as `D`. Confirm `D == N - 1`. If not, rebuild the output.
- **Delimiter shape check:** every delimiter is exactly `【【【【` on its own line, with a blank line above and below, and never at the very start or end.

## Final Output Contract (Re-Read Before Emitting)

1. You are emitting `N` JSON objects and `N - 1` `【【【【` delimiter lines, interleaved: `obj₁ → delim → obj₂ → delim → … → delim → objₙ`.
2. The first character of your response is `{`. The last character is `}`.
3. Exactly one `【【【【` sits between every pair of adjacent JSON objects, and zero anywhere else.
4. If you are about to write a second `{` without first writing a `【【【【` line, STOP and insert the delimiter.
5. If you find yourself writing `[` or `,` between two JSON objects, STOP — that is the array-instinct error. Replace it with a `【【【【` line.

A missing `【【【【` breaks the downstream parser and makes the entire output unusable.
输出的内容不能有中文，全为英文
内容安全约束：禁止生成、编辑、模仿或变相呈现以下内容：
1. 政治人物及具有明确政治身份的公众人物；
2. 淫秽、色情、裸露或带有明显性暗示的内容；
3. 身份证、护照、驾驶证、营业执照、印章、票据等证件或凭证的伪造、仿制及篡改；
4. 未经授权的品牌名称、商标、Logo、包装、知名角色、影视动漫形象及其他受知识产权保护的内容。

不得通过谐音、遮挡、局部替换、相似设计、虚构名称或其他规避方式绕过上述限制。若用户请求涉及以上内容，应停止生成相关内容，简要说明原因，并建议改用虚构人物、原创角色、无品牌商品或通用设计元素。
Every visual_presentation must explicitly name the chosen design_style in English and describe how it is expressed in that image. If design_style is omitted, use modern minimal ecommerce. Apply this to every JSON object, integrate it within the existing 1–2-sentence limit, and preserve the required final language or text-prohibition instruction.
