AWS Rekognition

Amazon Rekognition is AWS's fully managed computer vision service that adds image and video analysis to applications through simple API calls, without requiring any machine learning expertise or infrastructure to build and train models from scratch. It analyzes millions of images and hours of video within seconds, making deep learning-based visual analysis accessible to developers who just want to call an API rather than build a computer vision pipeline.

The service covers a wide range of use cases through distinct capabilities: face detection and analysis (identifying attributes like emotion, age range, and whether eyes are open), face comparison and search against a private collection, Face Liveness to detect spoofing attempts during remote identity verification, label detection for objects and scenes, text detection in images, celebrity recognition, and content moderation that flags unsafe or inappropriate visual content. For teams with specialized needs, Rekognition Custom Labels uses AutoML to train a model that recognizes business-specific objects, such as a company logo or a piece of equipment on a factory line, using as few as 10 to 20 training images. Amazon Rekognition Video extends these capabilities to both stored video in Amazon S3 and real-time streaming video from connected cameras via Amazon Kinesis Video Streams.

Rekognition is built for developers and businesses already working within or adopting the AWS ecosystem: security and identity teams building onboarding verification flows, media and advertising companies automating content tagging and moderation, retailers cataloging product imagery, and connected-device makers building smart home or workplace safety alerts. It is not a no-code tool or a consumer app — it is an API-first service that expects integration through the AWS SDKs, CLI, or console, and pricing scales with usage rather than following flat subscription tiers.

Amazon Rekognition uses pay-as-you-go pricing with no upfront commitment across four usage categories. Image analysis: Group 1 APIs (face search/compare/index) and Group 2 APIs (labels, moderation, text, celebrities) are billed per image on a tiered scale starting at $0.0010 per image for the first million images per month, dropping to $0.0004 per image at higher volumes; Image Properties is billed separately starting at $0.00075 per image. Face metadata storage costs $0.00001 per face or user vector per month. Video analysis: stored video is billed per minute (for example $0.10/min for Label Detection, $0.05/min for Shot Detection), while streaming video events cost around $0.00817 per minute processed. Custom Labels charges $1 per training hour and $4 per inference hour (inference must be manually deprovisioned to stop billing). Face Liveness checks start at $0.015 per check for the first 500,000 checks per month, decreasing at higher volumes. Custom Moderation adds $5 per training hour plus a tiered per-image inference cost starting at $0.0012 per image. New AWS accounts get a 12-month Free Tier (1,000 images/month, 60 video minutes/month, 2 free training hours) plus up to $200 in AWS Free Tier credits.