AWS Rekognition
AIAnalyze images with Amazon Rekognition: detect objects, text, faces, unsafe content, celebrities, and protective equipment, and manage face collections for search and comparison.
What This Integration Enables
Rekognition reads what is in an image. It labels objects and scenes, extracts text, detects and compares faces, screens for unsafe content, and checks for protective equipment. Face collections let an agent index known faces and search or match against them, which is useful for access and identity workflows. These are consequential detections. A face match or a safety flag can trigger a real-world action, so the model's confidence and a person's judgment both matter. An orchestration layer is what routes high-confidence detections straight through and holds the borderline ones for review, and FlowRunner is built for that layer. The model does the looking; a person owns the calls that carry weight.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Claims Image Intake
Photos submitted with an insurance claim need to be catalogued and checked. The agent runs label and text detection on each image, reads the visible serial number, and matches it against the policy record. It attaches a structured summary to the claim. The adjuster reviews a catalogued file instead of opening every photo, and anything the model reads with low confidence is flagged for a person.
Workplace Safety Monitoring
Site photos are submitted for a safety audit. The agent runs Detect Protective Equipment on each image to check whether required gear is present. Images that pass are logged automatically. Any image where equipment is missing or the detection is uncertain is routed to a safety officer with the photo and the model's read, so a person confirms every potential violation.
Content Screening
User-uploaded images need screening before they go live. The agent runs moderation detection on each upload. Clean images publish. Anything the model flags as potentially unsafe is held and routed to a moderator with the model's labels attached, so the borderline calls are made by a person rather than a threshold.
Human-in-Loop Highlight
A face match or a safety flag is exactly the kind of detection that should not act on its own. When Rekognition returns a face match, a missing-equipment result, or an unsafe-content flag, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, assembles the image and the detection with its confidence score, and sends it to the responsible person via Slack. They confirm or reject. The model surfaces the candidates; a person owns every consequential match.
Agent Capabilities
13 actionsImage Analysis
4- Detect Labels Detects real-world objects, scenes, concepts, and activities in an image, each with a confidence score, and returns bounding boxes for detected instances.
- Detect Text Detects text in an image and returns each detected line and word with its confidence and bounding geometry.
- Detect Moderation Labels Detects unsafe or inappropriate content in an image (such as explicit or suggestive nudity, violence, drugs, hate symbols, and more), returning hierarchical moderation categories with confidence scores.
- Detect Protective Equipment Detects personal protective equipment (PPE) worn by people in an image, reporting per-person body parts and whether each is covered by a face cover, hand cover, or head cover, plus an overall summary of persons with and without required equipment.
Face Analysis
3- Detect Faces Detects faces in an image and returns per-face details such as bounding box, quality, pose, and (when All Attributes is enabled) age range, emotions, facial landmarks, gender, and attributes like smile, eyeglasses, and beard.
- Recognize Celebrities Recognizes well-known people in an image and returns matched celebrities with their name, ID, match confidence, and known URLs, plus a count of unrecognized faces.
- Compare Faces Compares the largest face in a source image against faces in a target image and returns matches above a similarity threshold, along with unmatched faces.
Collections
6- Create Collection Creates a face collection, a server-side container used to store searchable face vectors indexed from images.
- List Collections Lists the face collection IDs in the configured region and account, along with the face model version used by each.
- Delete Collection Permanently deletes a face collection and all of the faces indexed within it. This action cannot be undone.
- Index Faces Detects faces in an image and adds them to the specified collection as searchable face vectors, optionally tagging them with an external image ID.
- Search Faces by Image Searches a collection for faces matching the largest face detected in the supplied image, returning matches above a similarity threshold.
- List Faces Lists the faces indexed in a collection, returning each face's ID, bounding box, external image ID, and confidence.
Start building with AWS Rekognition
$100 in credits. No card required. Connect in minutes.