generated: '2026-08-11'
method: derived
source: >-
  openapi/anyimagedetector-ai-image-detector-openapi.yml, enriched from the published
  interpretation rules at https://imagedetector.online/methodology
docs: https://imagedetector.online/methodology
description: >-
  The AI Image Detector API has no persistent object graph — no ids, no id prefixes, no
  resources to fetch, list or relate. It is a stateless transform: an image goes in, a scored
  judgement comes out. Two schemas exist and neither references the other. What IS worth
  modelling is the SEMANTICS of the one returned object, because the provider publishes the
  exact score bands that produce `verdict` and `confidence` — so a consumer can reproduce those
  labels from `ai_score` alone, and can reason about the uncertain band rather than treating the
  verdict as a black box.
stateless: true
identifier_scheme:
  ids_returned: false
  id_prefixes: []
  note: >-
    No detection id is returned, so a caller cannot reference, re-fetch or dispute a specific
    result through the API. Signed-in users see results in a web "detection history"; there is
    no API to read it.
entities:
  - name: DetectionResult
    schema: DetectionResult
    kind: value-object
    persisted_for_caller: false
    operations: [detectImage]
    fields:
      - name: verdict
        type: string
        enum: [likely_ai, likely_human, uncertain]
        derived_from: ai_score
      - name: ai_score
        type: number
        range: 0.0-1.0
        description: Detected likelihood of AI generation. The primitive; everything else is a label on it.
      - name: confidence
        type: string
        enum: [low, medium, high]
        derived_from: distance of ai_score from the uncertain band
      - name: source_breakdown
        type: array
        state: reserved
        description: Documented as "reserved for future per-generator scores; currently always []".
  - name: ImageInput
    kind: request-payload
    persisted: false
    mutually_exclusive_fields: [image, imageUrl]
    fields:
      - name: image
        type: file
        constraints: JPG, PNG or WebP; max 8 MB; multipart/form-data
      - name: imageUrl
        type: string
        constraints: publicly reachable HTTP/HTTPS URL; application/json
    retention: >-
      "The application does not add the submitted image file to detection history." Result data
      (score, verdict) may persist for signed-in users; the image file does not.
  - name: Error
    schema: Error
    kind: value-object
    detail: errors/anyimagedetector-problem-types.yml
relationships: []
relationships_note: >-
  None. No entity references another; there are no has_one/has_many/belongs_to links because
  there are no identified resources.
derivation_rules:
  name: Verdict and confidence thresholds
  source: https://imagedetector.online/methodology
  published: true
  note: >-
    Published verbatim by the provider. These bands let a consumer recompute verdict and
    confidence from ai_score, or apply its own stricter cutoff for a high-stakes workflow.
  bands:
    - ai_score_min: 0.75
      ai_score_max: 1.00
      verdict: likely_ai
      confidence: high
    - ai_score_min: 0.55
      ai_score_max: 0.74
      verdict: likely_ai
      confidence: medium
    - ai_score_min: 0.46
      ai_score_max: 0.54
      verdict: uncertain
      confidence: low
    - ai_score_min: 0.26
      ai_score_max: 0.45
      verdict: likely_human
      confidence: medium
    - ai_score_min: 0.00
      ai_score_max: 0.25
      verdict: likely_human
      confidence: high
  caveat: >-
    The provider states these thresholds "make the output easier to interpret; they do not turn
    a probabilistic result into certainty," and instructs that a single detector score must not
    be the sole basis for disciplinary action, legal claims, payment refusal, account suspension
    or public accusations.