AWS Rekognition

Analyze images for faces, labels, text, celebrities, and moderation content

Overview

The AWS Rekognition tool sends images to the Amazon Rekognition API for computer vision analysis. It supports five analysis types: Detect Faces (facial landmarks, emotions, attributes), Detect Labels (objects, scenes, concepts), Detect Text (OCR), Recognize Celebrity (identify famous people), and Detect Moderation Labels (unsafe content). Images can be provided as binary data from an upstream node or referenced via an S3 bucket and object key. The tool reads the binary image, sends it to the Rekognition API, and returns the analysis results as JSON.

Category: Utility
Tool Name: aws_rekognition
Version: 1

Appearance: Icon: lucide-Server | Color: #FF9900

Node Type

Action (Binary) — handles file/binary data operations

Input / Output

Direction	Port(s)
Input	Input
Output	Output, Error

Credentials

This tool requires AWS credentials. See the Credentials Guide for setup instructions.

Resources

Resource	Value
Image	image

Operations

Operation	Value	Description
Analyze	analyze	The operation to perform.

Parameters

Image: Analyze

Parameter	Type	Required	Default	Description
Type	options	No	detectFaces	The type of image analysis to perform.
				Options: detectFaces (detect faces and facial attributes in an image), detectLabels (detect objects, scenes, and concepts in an image), detectModerationLabels (detect inappropriate or unsafe content), detectText (detect and extract text from an image — OCR), recognizeCelebrity (identify celebrities in an image)
Binary File	boolean	Yes	false	Whether the image to analyze should be taken from a binary field. When false, provide an S3 bucket and object key.
Input Binary Field	string	Yes	data	The name of the binary property containing the image to analyze. Names are case-sensitive — see the upstream node’s Binary Data panel for the exact names to use. (shown when Binary File is true)
Bucket	string	Yes	—	Name of the S3 bucket containing the image. Supports expressions like {{ $json.bucket }}. (shown when Binary File is false)
Name	string	Yes	—	S3 object key name of the image. (shown when Binary File is false)
Additional Fields	collection	No	{}	Optional settings that refine the analysis. Which of them apply depends on the Type you chose.
— Attributes	multiOptions	No	[]	An array of facial attributes you want to be returned. “All” returns all attributes; “Default” returns a subset (BoundingBox, Confidence, Pose, Quality, Landmarks). (shown when Type is detectFaces)
				Options: all, default
— Max Labels	number	No	0	Maximum number of labels to return. The service returns the specified number of highest-confidence labels. Set to 0 for no limit. (shown when Type is detectLabels, detectModerationLabels)
— Min Confidence (minConfidence)	number	No	0	Minimum confidence level for returned labels. Labels below this threshold are excluded. Value between 0 and 100. (shown when Type is detectLabels, detectModerationLabels)
— Version	string	No	—	If the bucket is versioning-enabled, specify the object version to analyze. (shown when Binary File is false)
— Regions of Interest	fixedCollection	No	{}	Bounding box regions to restrict text detection analysis to specific parts of the image. Add more than one entry to cover several regions. (shown when Type is detectText)
— — Height	number	No	0	Height of the bounding box as a ratio of the overall image height.
— — Left	number	No	0	Left coordinate of the bounding box as a ratio of overall image width.
— — Top	number	No	0	Top coordinate of the bounding box as a ratio of overall image height.
— — Width	number	No	0	Width of the bounding box as a ratio of the overall image width.
— Word Filter	collection	No	{}	Filters to apply to detected text words. (shown when Type is detectText)
— — Min Bounding Box Height	number	No	0	Minimum height of the word bounding box. Words with bounding box heights less than this value are excluded. Value is relative to image height.
— — Min Bounding Box Width	number	No	0	Minimum width of the word bounding box. Words with bounding box widths less than this value are excluded. Value is relative to image width.
— — Min Confidence (MinConfidence)	number	No	0	Minimum confidence for word detection. Words with confidence below this value are excluded. Values should be between 50 and 100.

All Operations

Parameter	Type	Required	Default	Description
Max Concurrency	number	No	10	Maximum number of items to process concurrently.

Output Data

One output item per input item. The Amazon Rekognition response is merged into the item JSON at the top level — the incoming fields stay addressable alongside the analysis, and the SDK’s transport metadata ($metadata) is stripped before the merge. Binary data on the input item is forwarded unchanged, and it is forwarded on error items too, so a retry branch still has the image.

The merged payload is the raw response of the Amazon Rekognition operation that matches your Type, so its shape changes with that selection:

Type	Rekognition operation called	What comes back
detectFaces	DetectFaces	One entry per detected face, each with its bounding box, confidence, pose, quality and landmarks — plus emotions, age range and the other facial attributes when Attributes is set to all.
detectLabels	DetectLabels	The objects, scenes and concepts found in the image, each with a confidence score, trimmed by Max Labels and Min Confidence.
detectModerationLabels	DetectModerationLabels	The unsafe-content categories the image matched, each with a confidence score and its parent category.
detectText	DetectText	Detected words and lines, each with its text, confidence and bounding geometry, narrowed by Regions of Interest and Word Filter.
recognizeCelebrity	RecognizeCelebrities	The recognized celebrities with their names, match confidence and known URLs, alongside the faces that could not be matched.

The nested detection arrays keep Amazon’s own PascalCase field names, so downstream expressions address them exactly as the Rekognition API reference documents them for the operation you called. Because a whole detection array lands on one item, use a Split Out node when you want one workflow item per face, label or word.

Nothing is written when an image contains no matches — the response simply carries an empty array, which is a successful result rather than an error.

Usage Examples

• Detect faces and emotions in a photo
• Identify objects and scenes in an image using label detection
• Extract text from an image using OCR
• Recognize celebrities in a photograph
• Check an image for inappropriate content with moderation labels
• Analyze an image stored in S3 without downloading it first

Example Configuration

Detect faces, with the full attribute set, on an image supplied as binary data:

{
  "type": "aws_rekognition",
  "parameters": {
    "resource": "image",
    "operation": "analyze",
    "type": "detectFaces",
    "binaryData": true,
    "binaryPropertyName": "data",
    "additionalFields": {
      "attributes": ["all"]
    }
  }
}

Label an image that already lives in S3, keeping only high-confidence results:

{
  "type": "aws_rekognition",
  "parameters": {
    "resource": "image",
    "operation": "analyze",
    "type": "detectLabels",
    "binaryData": false,
    "bucket": "{{ $json.bucket }}",
    "name": "photos/sample-image.jpg",
    "additionalFields": {
      "maxLabels": 10,
      "minConfidence": 75
    }
  }
}

Run OCR over the top half of an image and drop small or unreliable words:

{
  "type": "aws_rekognition",
  "parameters": {
    "resource": "image",
    "operation": "analyze",
    "type": "detectText",
    "binaryData": true,
    "binaryPropertyName": "data",
    "additionalFields": {
      "regionsOfInterestUi": {
        "regionsOfInterestValues": [
          { "top": 0, "left": 0, "width": 1, "height": 0.5 }
        ]
      },
      "wordFilterUi": {
        "MinBoundingBoxHeight": 0.02,
        "MinConfidence": 85
      }
    }
  }
}

Screen user uploads for unsafe content:

{
  "type": "aws_rekognition",
  "parameters": {
    "resource": "image",
    "operation": "analyze",
    "type": "detectModerationLabels",
    "binaryData": false,
    "bucket": "user-uploads",
    "name": "content/user-image-123.jpg",
    "additionalFields": {
      "minConfidence": 70
    },
    "maxConcurrency": 20
  }
}

Error Handling

Mode	Behavior
stop	Halts workflow on first error
continue	Skips failed items, passes successful ones through
errorPort	Routes failed items to Error output port

Tips

Analyze images with AWS Rekognition to detect faces, labels, text, celebrities, or moderation content.

• Two ways in, one node. Turn Binary File on to analyze a file already on the item; turn it off and give a Bucket and Name to analyze an object in S3 without downloading it first. Version applies to the S3 route only, for versioning-enabled buckets.
• Additional Fields are scoped to the Type. Attributes belongs to detectFaces; Max Labels and Min Confidence to detectLabels and detectModerationLabels; Regions of Interest and Word Filter to detectText. Fields set for a different Type are simply not sent.
• all costs more than default. The default attribute set returns bounding box, confidence, pose, quality and landmarks; all adds the emotion, age and appearance attributes and is billed accordingly.
• Min Confidence is a percentage. Values run 0–100, and Amazon recommends staying at or above 50 for word-level text filtering.
• Region comes from the credential. The node analyzes images in the region recorded on the AWS credential, so an S3 bucket in a different region needs a credential for that region.
• One node, one analysis type. To collect faces and labels for the same image, chain two nodes rather than trying to combine types.