generated: '2026-06-20'
method: searched
source: >-
  https://docs.aws.amazon.com/rekognition/latest/APIReference/Welcome.html,
  https://docs.aws.amazon.com/rekognition/latest/dg/error-handling.html, and
  derived from openapi/amazon-rekognition-openapi.yml. Cross-cutting request/
  response semantics that apply across every Amazon Rekognition operation.
description: >-
  How the Amazon Rekognition API behaves across operations: protocol, auth,
  idempotency, pagination, request tracing, versioning, error envelope, and
  throttling signals.
base_url: https://rekognition.amazonaws.com
api_style: >-
  AWS JSON 1.1 RPC over HTTPS. Every call is an HTTP POST with an
  X-Amz-Target: RekognitionService.<Action> header and a JSON body; the operation
  is selected by the target header rather than the URL path.
content_type: application/x-amz-json-1.1
authentication:
  scheme: AWS Signature Version 4 (SigV4) over IAM credentials
  header: Authorization (AWS4-HMAC-SHA256)
  docs: https://docs.aws.amazon.com/rekognition/latest/dg/security_iam_service-with-iam.html
  detail: authentication/amazon-rekognition-authentication.yml
idempotency:
  supported: partial
  mechanism: ClientRequestToken request field
  applies_to: >-
    Asynchronous video jobs (StartLabelDetection and other Start* operations) and
    CreateFaceLivenessSession. Reusing a token with different parameters raises
    IdempotentParameterMismatchException. Synchronous analysis calls have no
    global idempotency key.
pagination:
  style: cursor
  request_params:
    MaxResults: Maximum number of results per page.
    NextToken: Opaque token returned by the prior page; omit for the first page.
  response_fields:
    NextToken: Present when more results are available; pass it on the next call.
  note: Invalid or expired tokens raise InvalidPaginationTokenException.
field_expansion:
  supported: false
metadata:
  supported: true
  mechanism: Tags (key/value) on collections, stream processors, and Custom Labels projects.
  note: External identifiers can be attached to indexed faces via ExternalImageId.
request_tracing:
  request_id_header: x-amzn-RequestId
  description: Every response carries an x-amzn-RequestId used for AWS Support and CloudTrail correlation.
  audit_log: AWS CloudTrail records Rekognition API calls.
versioning:
  scheme: fixed-service-version
  current: '2016-06-27'
  model_versions: >-
    Deep-learning model versions are returned per response (LabelModelVersion,
    FaceModelVersion, ModerationModelVersion, TextModelVersion).
  detail: lifecycle/amazon-rekognition-lifecycle.yml
  changelog: changelog/amazon-rekognition-changelog.yml
error_envelope:
  media_type: application/x-amz-json-1.1
  rfc9457: false
  shape: '{ "__type": string, "Code": string, "Message": string, "Logref": string }'
  detail: errors/amazon-rekognition-problem-types.yml
  docs: https://docs.aws.amazon.com/rekognition/latest/dg/error-handling.html
rate_limits:
  signal_exceptions: [ProvisionedThroughputExceededException, ThrottlingException]
  guidance: Retry retryable throttling errors with exponential backoff and jitter (built into the AWS SDKs).
  detail: rate-limits/amazon-rekognition-rate-limits.yml
async_jobs:
  pattern: >-
    Video and bulk operations are asynchronous: a Start* call returns a JobId,
    completion is signaled via an Amazon SNS NotificationChannel, and results are
    retrieved with the matching Get* operation.
other_conventions:
  - name: Image input
    detail: Images are supplied as base64 Bytes (up to 5 MB) or as an S3Object {Bucket, Name, Version}.
  - name: Video input
    detail: Videos are supplied as an S3Object; results delivered asynchronously.
  - name: Confidence scores
    detail: Detection results carry a Confidence value between 0 and 100.
  - name: Regional endpoints
    detail: Rekognition is a regional service; use rekognition.<region>.amazonaws.com in practice.