arazzo: 1.0.1
info:
  title: OpenAI Moderate then Generate Image
  summary: Screen an image prompt, then generate an image only when it is allowed.
  description: >-
    Runs the image prompt through the moderation classifier and branches: when
    the prompt is flagged the flow ends, and when it is allowed it calls the
    image generation endpoint. Every step spells out its request inline so the
    flow can be read and executed without opening the underlying OpenAPI
    description.
  version: 1.0.0
sourceDescriptions:
- name: openaiApi
  url: ../openapi/openai-openapi-master.yml
  type: openapi
workflows:
- workflowId: moderate-then-generate-image
  summary: Moderate an image prompt and generate an image only when allowed.
  description: >-
    Classifies the supplied prompt for policy violations, and only when the
    prompt is not flagged does it request an image from the image generation
    endpoint.
  inputs:
    type: object
    required:
    - apiKey
    - imageModel
    - prompt
    properties:
      apiKey:
        type: string
        description: OpenAI API key used as a Bearer token.
      imageModel:
        type: string
        description: The image model id (e.g. dall-e-3 or gpt-image-1).
      prompt:
        type: string
        description: A text description of the desired image.
  steps:
  - stepId: moderate
    description: Classify the image prompt for policy violations.
    operationId: createModeration
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        input: $inputs.prompt
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      flagged: $response.body#/results/0/flagged
    onSuccess:
    - name: promptAllowed
      type: goto
      stepId: generateImage
      criteria:
      - context: $response.body
        condition: $.results[0].flagged == false
        type: jsonpath
    - name: promptFlagged
      type: end
      criteria:
      - context: $response.body
        condition: $.results[0].flagged == true
        type: jsonpath
  - stepId: generateImage
    description: Generate an image from the moderated, allowed prompt.
    operationId: createImage
    parameters:
    - name: Authorization
      in: header
      value: "Bearer $inputs.apiKey"
    requestBody:
      contentType: application/json
      payload:
        model: $inputs.imageModel
        prompt: $inputs.prompt
        n: 1
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      created: $response.body#/created
      firstImage: $response.body#/data/0
  outputs:
    flagged: $steps.moderate.outputs.flagged
    firstImage: $steps.generateImage.outputs.firstImage