arazzo: 1.0.1
info:
  title: Adobe Photoshop Product Photo Pipeline
  summary: Cut out a product, wait for the cutout, then smart-crop it to the subject.
  description: >-
    The e-commerce catalog pattern. A raw product photograph is stripped of its
    background by the Sensei background removal service, and once that job lands
    the resulting cutout is fed straight into the AI product crop endpoint, which
    detects the subject and crops to it with configurable padding. Each stage is
    asynchronous, so the flow polls the matching status endpoint between
    submissions and passes the finished asset forward by href. Every step spells
    out its request inline so the flow can be read and executed without opening
    the underlying OpenAPI.
  version: 1.0.0
sourceDescriptions:
- name: photoshopApi
  url: ../openapi/adobe-photoshop-api-openapi-original.yml
  type: openapi
workflows:
- workflowId: product-photo-pipeline
  summary: Chain background removal into product crop for a catalog-ready image.
  description: >-
    Submits the source photo for background removal, polls the Sensei job, then
    submits the cutout for AI product crop and polls the PSD service job until
    the cropped asset is written to the output URL.
  inputs:
    type: object
    required:
    - imageUrl
    - cutoutOutputUrl
    - croppedOutputUrl
    properties:
      imageUrl:
        type: string
        description: Pre-signed GET URL of the raw product photograph.
      cutoutOutputUrl:
        type: string
        description: Pre-signed PUT URL where the background-removed cutout is written.
      croppedOutputUrl:
        type: string
        description: Pre-signed PUT URL where the final cropped image is written.
      padding:
        type: number
        description: Padding to leave around the detected product subject.
        default: 20
      unit:
        type: string
        description: Unit for the padding value, pixel or percent.
        default: pixel
  steps:
  - stepId: submitRemoveBackground
    description: >-
      Remove the background from the raw product photograph, returning a PNG
      cutout with a transparent background.
    operationId: removeBackgroundV2
    requestBody:
      contentType: application/json
      payload:
        image:
          source:
            url: $inputs.imageUrl
        mode: cutout
        output:
          mediaType: image/png
        trim: true
    successCriteria:
    - condition: $statusCode == 202
    outputs:
      statusHref: $response.body#/_links/self/href
  - stepId: awaitRemoveBackground
    description: >-
      Poll the Sensei status endpoint until the cutout is ready. The submit call
      returns the absolute status URL in _links.self.href; the jobId this
      operation requires is the final path segment of that URL, so a runner takes
      that segment from statusHref before issuing the request.
    operationId: getSenseiJobStatus
    parameters:
    - name: jobId
      in: path
      value: $steps.submitRemoveBackground.outputs.statusHref
    successCriteria:
    - condition: $statusCode == 200
    - context: $response.body
      condition: $.outputs[0].status == 'succeeded'
      type: jsonpath
    onFailure:
    - name: keepPolling
      type: retry
      retryAfter: 3
      retryLimit: 20
      criteria:
      - context: $response.body
        condition: $.outputs[0].status == 'pending' || $.outputs[0].status == 'running'
        type: jsonpath
    - name: cutoutFailed
      type: end
      criteria:
      - context: $response.body
        condition: $.outputs[0].status == 'failed'
        type: jsonpath
    outputs:
      cutoutHref: $response.body#/outputs/0/_links/self/href
  - stepId: submitProductCrop
    description: >-
      Feed the finished cutout into the AI product crop endpoint, which finds the
      main subject and crops to it with the requested padding.
    operationId: productCrop
    requestBody:
      contentType: application/json
      payload:
        inputs:
        - href: $steps.awaitRemoveBackground.outputs.cutoutHref
          storage: external
        options:
          padding: $inputs.padding
          unit: $inputs.unit
        outputs:
        - href: $inputs.croppedOutputUrl
          storage: external
          type: image/png
          overwrite: true
    successCriteria:
    - condition: $statusCode == 202
    outputs:
      statusHref: $response.body#/_links/self/href
  - stepId: awaitProductCrop
    description: >-
      Poll the PSD service status endpoint until the crop job reports succeeded.
      The jobId is the final path segment of the returned status URL.
    operationId: getPsdJobStatus
    parameters:
    - name: jobId
      in: path
      value: $steps.submitProductCrop.outputs.statusHref
    successCriteria:
    - condition: $statusCode == 200
    - context: $response.body
      condition: $.outputs[0].status == 'succeeded'
      type: jsonpath
    onFailure:
    - name: keepPolling
      type: retry
      retryAfter: 3
      retryLimit: 20
      criteria:
      - context: $response.body
        condition: $.outputs[0].status == 'pending' || $.outputs[0].status == 'running'
        type: jsonpath
    - name: cropFailed
      type: end
      criteria:
      - context: $response.body
        condition: $.outputs[0].status == 'failed'
        type: jsonpath
    outputs:
      status: $response.body#/outputs/0/status
      croppedHref: $response.body#/outputs/0/_links/self/href
  outputs:
    cutoutHref: $steps.awaitRemoveBackground.outputs.cutoutHref
    croppedHref: $steps.awaitProductCrop.outputs.croppedHref
    status: $steps.awaitProductCrop.outputs.status