How to Create Custom Prompt Templates for E-commerce Product Images

You can create custom prompt templates for e-commerce product images by selecting a base skeleton from the documentation, populating placeholders with product-specific data, and submitting the rendered string through the API payload builder to the generation service.

The awesome-gpt-image-2 repository provides a modular, three-layer architecture for building reusable prompt templates that drive AI-generated e-commerce visuals. By leveraging structured templates from docs/templates.md and the API utilities in shared/apimart.js, developers can standardize product imagery across platforms while maintaining flexibility for individual product attributes.

Understanding the Template Architecture

The repository implements a clear separation of concerns across three distinct layers that handle template definition, prompt generation, and API communication.

Template Definition Layer

The template definition layer stores reusable "fill-in-the-blank" prompt skeletons in docs/templates.md. According to the freestylefly/awesome-gpt-image-2 source code, this file contains specific e-commerce sections with both plain-text and JSON-structured templates that describe product type, platform, visual style, and constraints. These templates define placeholders for platform names, color modes, aspect ratios, and content sections.

Prompt Generation Layer

The prompt generation layer handles the replacement of placeholders with actual product data. The file src/image25/cases.js demonstrates concrete implementations, containing sample prompts for product images such as ceramic mug descriptions. This layer optionally serializes the result to JSON for programmatic API consumption, transforming human-readable templates into machine-parsable payloads.

API Payload Construction Layer

The API payload construction layer packages the final prompt into the format expected by the backend generation service. The shared/apimart.js file exports buildApimartGenerationPayload(prompt, …), which normalizes the prompt string by trimming whitespace and adding metadata such as language codes. Subsequently, api/generate-image.js validates prompt length and case identifiers before forwarding the request to the generation engine.

Step-by-Step Guide to Creating E-commerce Prompt Templates

Select a Base Template from the Documentation

Begin by opening docs/templates.md and locating the e-commerce template block. For product-detail cards or social commerce posts, use the plain-text template structure:

生成一张[平台，如 X/抖音/小红书/微信朋友圈]内容截图，[深色/浅色]模式。
整体比例：[9:16 / 3:4 / 1:1]，手机截图风格。

核心内容：
- 账号信息：[头像描述 / 用户名 / 认证标识]
- 正文内容：[具体文本内容，包含指定中文]
- 互动数据：[点赞/评论/转发/收藏数量]

This template originates from the repository's documentation and supports popular Chinese social commerce platforms including Douyin (抖音) and Xiaohongshu (小红书).

Structure Your Product Data as JSON

For applications requiring programmatic template processing, convert the plain-text skeleton into a JSON schema. Create a structured object that maps template sections to data fields:

{
  "type": "e-commerce landing page hero section",
  "product": "Organic Face Cream",
  "features": ["天然配方", "适合干性皮肤", "含维生素E"],
  "style": {"theme":"Minimalist","primary_color":"#FF5733"},
  "layout": "Hero + Product Card",
  "constraints":"High-resolution, text legible, 1:1 ratio"
}

This JSON structure mirrors the pattern found in the repository's infographic templates and enables type-safe validation before API submission.

Populate Placeholders with Product Details

Replace each bracketed placeholder with specific product information. For example, rendering a summer dress listing for Douyin:

生成一张抖音内容截图，浅色模式。整体比例：9:16，手机截图风格。

核心内容：
- 账号信息：头像为圆形，用户名 “ShopifyStore”，已认证。
- 正文内容：全新夏季连衣裙，轻盈面料，颜色：海蓝。
- 互动数据：点赞 2.1k，评论 184，转发 57，收藏 312。

This approach aligns with the implementation style demonstrated in src/image25/cases.js, where concrete product attributes replace template variables.

Serialize and Prepare the API Payload

When using the JSON template approach, serialize the object and prepare it for transmission:

const template = {
  type: "e-commerce product detail",
  product: "Summer Dress",
  colors: ["#00ADEF", "#FFFFFF"],
  layout: "Full-width image with price tag",
  constraints: "No watermarks, 4K resolution"
};
const prompt = JSON.stringify(template);

The buildApimartGenerationPayload function in shared/apimart.js automatically handles the prompt field normalization, ensuring the string is properly trimmed and formatted before API submission.

Submit to the Generation Service

Invoke either submitPersonalGeneration for client-side requests or submitPlatformGeneration for server-side processing, passing the rendered prompt string:

import { submitPersonalGeneration } from "./shared/apimart.js";

const apiKey = process.env.APIMART_KEY;
const language = "zh";
await submitPersonalGeneration(prompt, apiKey, language);

As implemented in api/generate-image.js, the server validates the prompt length and case ID at lines 115-119 before queuing the generation job. The endpoint returns a URL to the created image, which you can embed in your e-commerce front-end or marketing assets.

Complete Implementation Example

The following example demonstrates the complete workflow from template selection to API submission:

// 1️⃣ Load a plain-text template and fill it
function renderEcomPrompt({ platform, mode, ratio, account, copy, stats }) {
  return `
生成一张${platform}内容截图，${mode}模式。整体比例：${ratio}，手机截图风格。

核心内容：
- 账号信息：${account}
- 正文内容：${copy}
- 互动数据：${stats}
`.trim();
}

// Example usage for a fashion product
const prompt = renderEcomPrompt({
  platform: "抖音",
  mode: "浅色",
  ratio: "9:16",
  account: "头像圆形，用户名 “FashionHub”，已认证",
  copy: "新品夏季连衣裙，轻盈面料，颜色：海蓝",
  stats: "点赞 2.1k，评论 184，转发 57，收藏 312"
});

// 2️⃣ Send to backend using the repository's client helper
import { submitPersonalGeneration } from "./shared/apimart.js";

await submitPersonalGeneration(prompt, process.env.APIMART_KEY, "zh");

For JSON-based workflows, structure and submit the template as follows:

import { submitPersonalGeneration } from "./shared/apimart.js";

const jsonPrompt = JSON.stringify({
  type: "e-commerce product detail",
  product: "Smartwatch Pro",
  features: ["OLED display", "Heart-rate monitor", "Water-resistant 5ATM"],
  style: {theme:"Tech", primary_color:"#222222"},
  layout: "Hero image + specs overlay",
  constraints:"4K resolution, text legible, no brand logos"
});

await submitPersonalGeneration(jsonPrompt, process.env.APIMART_KEY, "en");

Summary

• Template Storage: Reusable e-commerce prompt skeletons reside in docs/templates.md, supporting both plain-text and JSON formats for different integration scenarios.
• Data Population: Replace bracketed placeholders with specific product attributes such as platform names, color values, and feature lists to create concrete generation prompts.
• API Integration: Use buildApimartGenerationPayload from shared/apimart.js to normalize prompts before submission, and submitPersonalGeneration to transmit requests to the generation service.
• Validation Pipeline: The server endpoint in api/generate-image.js validates prompt lengths and case identifiers at lines 115-119 before processing.
• Platform Flexibility: Templates support multiple e-commerce platforms including X (Twitter), Douyin, Xiaohongshu, and WeChat Moments, with configurable aspect ratios (9:16, 3:4, 1:1).

Frequently Asked Questions

What file contains the base e-commerce prompt templates in awesome-gpt-image-2?

The base e-commerce prompt templates are stored in docs/templates.md, specifically within the e-commerce section that contains both plain-text and JSON-structured examples. This file includes placeholders for platform specifications, visual modes, and content sections that you can customize for specific products.

How does the repository handle prompt validation before image generation?

According to the source code in api/generate-image.js (lines 115-119), the server validates the prompt length and checks case identifiers before forwarding the request to the generation engine. Additionally, shared/apimart.js normalizes the prompt string through buildApimartGenerationPayload, which trims whitespace and ensures proper formatting.

Can I use JSON format instead of plain text for my e-commerce templates?

Yes, the repository supports both formats. While docs/templates.md provides plain-text templates with bracketed placeholders, you can structure your data as JSON objects containing fields like type, product, features, style, layout, and constraints. Serialize these objects using JSON.stringify() before passing them to submitPersonalGeneration or submitPlatformGeneration.

What function should I call to submit a completed prompt to the image generation API?

Use submitPersonalGeneration(prompt, apiKey, language) for client-side submissions or submitPlatformGeneration for server-side processing, both imported from shared/apimart.js. These functions utilize buildApimartGenerationPayload internally to construct the proper API request format required by the generation service.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}