ecommerce-images

E-commerce main image/detail image workflow (script-free version)

Execute according to the following process, and do not create custom code scripts for this skill.

Input specification

• Required input:
  • Product original image path (local file, can be used as a reference image)
• Optional input:
  • mode：main | detail | both(default both）
  • mainStyle: main image style name (recommended)
  • detailStyle: detail image style name (recommended)
  • detailCount: number of detail images (when mode=detail|both , must ask; if the user has not provided it, ask first)
  • quality、imageSize、model(passed through to the image generation skill)
  • ar(optional override; if not passed, use the default ratio)

Execution order

1. Verify that the input image exists and is readable.
2. Determine the generation target based on mode :
   • main: generate only the main image
   • detail: generate only the detail image set
   • both: generate the main image first, then the detail image set
3. When mode=detail|both , first ask the user "How many detail images are needed?", and after obtaining detailCount , execute.
4. Assemble Chinese prompts for each target image according to the style name.
5. First call banana-proxy to generate:
   • Use the reference image (--ref）
   • Default ratio:
     • Main image:1:1
     • Detail image:3:4
   • Suggested output paths:
     • Main image:out/ecommerce-images/<原图名>_main_<风格key>.png
     • Detail image i:out/ecommerce-images/<原图名>_detail_<风格key>_n<i>.png
6. If banana-proxy fails, immediately fall back to baoyu-image-gen：
   • keep the same prompt, reference image, and output path conventions
7. Return a structured result, including at least: mode, style name, number of detail images, final output path of each image, and the provider actually used.

Provider strategy

• Default:banana-proxy
• Fallback:baoyu-image-gen
• Reverse order is prohibited (unless the user explicitly requests it)

Style selection (extensible)

Main image styles (mainStyle）：

• 白底极简(default)
• 卖点文案
• 轻场景

Detail image styles (detailStyle）：

• 卖点总览(default)
• 参数规格
• 场景对比

When an undefined style name is received, fall back to the default style, and state in the result that a fallback occurred.

Detail image set rules

• detailCount Suggested values for3-6 : images (adjustable according to user requirements).
• Each image in the set needs a clear division of roles to avoid repeated compositions. Example division of roles:
  • Image 1: overview of core selling points
  • Image 2: partial close-up
  • Image 3: parameters/specifications
  • Image 4: usage scenario
  • Image 5: comparison or additional details
• If the user only says "generate detail images" but does not give a quantity, you must first ask for the quantity, then start generation.

Prompt template

Main image 白底极简：

基于参考图片中的商品，生成一张电商平台可用的商品主图。
要求：
1) 商品为唯一主体，完整不截断、无遮挡。
2) 使用干净白色或浅色背景，画面简洁。
3) 不添加Logo、水印、价格角标、二维码。
4) 光线均匀，突出材质与轮廓，符合主流电商审核要求。

Detail image 卖点总览：

基于参考图片中的商品，生成一张电商详情页模块图。
要求：
1) 以商品为主体，可展示2-3个局部特写。
2) 可添加少量简短中文卖点，排版清晰易读。
3) 背景可轻场景化，但不喧宾夺主。
4) 不添加Logo、水印、二维码或平台强绑定UI元素。

Environment variables

• When using banana-proxy :
  • LNAPI_KEY(required)
• When using baoyu-image-gen fallback:
  • Prefer GOOGLE_API_KEY, or OPENAI_API_KEY

Output convention

• mode=main: return the main image path
• mode=detail: return the detail image path
• mode=both: return the main image and detail image paths
• Detail images return an array:savedDetailImages[]
• Include for each image the provider（banana-proxy or baoyu-image-gen) and styleKey

user-facing wording

• First confirm the style in natural language, for example:
  • "Should the main image be minimalist with a white background, or include selling-point copy?"
  • "For the detail images, do you want a selling-point overview, parameter specifications, or scenario comparison?"
• If the user says "default style", then use for the main image 白底极简, and for the detail images 卖点总览。