type: image
name: translateDiagram
image_model:
  model: google/gemini-3-pro-image-preview
  imageConfig:
    imageSize:
      $get: size
    aspectRatio:
      $get: ratio

instructions: |
  You are a multilingual diagram translator specializing in technical diagrams such as flowcharts, architecture diagrams, network topologies, and mind maps.

  Your task is to regenerate the provided image by translating all label-style text into {{locale}}.

  Requirements:
  - Retain all command-style, technical, or protocol-specific terms (e.g., "API", "TCP/IP", "POST") in English;
  - Maintain the original diagram’s structure, layout, visual style, and spatial relationships between elements;
  - Only translate label or annotation text — all graphical elements should remain unchanged;
  - **If {{locale}} is Simplified Chinese, Traditional Chinese, Japanese, or Korean, use the font "Microsoft YaHei" (微软雅黑)** to ensure optimal readability, proper character rendering, and consistent typographic style across Latin and CJK scripts;
  - For other languages, use a clean sans-serif font that best fits the visual style of the original image;

  Your final output should be a regenerated image identical in design but localized in text according to {{locale}}.

input_file_key: existingImage

input_schema:
  type: object
  properties:
    existingImage:
      type: array
      description: Array of mediaFile objects for the source diagram image to translate
      items:
        type: object
        properties:
          type:
            type: string
            description: File type, should be "local"
          path:
            type: string
            description: Absolute path to the image file
          filename:
            type: string
            description: Image filename
          mimeType:
            type: string
            description: MIME type of the image (e.g., "image/png", "image/jpeg")
    ratio:
      type: string
      description: Aspect ratio of the image
    size:
      type: string
      description: Size of the generated image
      default: "1K"
    locale:
      type: string
      description: Target language code for translation
  required:
    - existingImage
    - ratio
    - locale
include_input_in_output: true