AWS Rekognition

Amazon Rekognition is AWS's computer vision service, offering pretrained APIs for image and video analysis rather than requiring customers to build or train their own models. It handles object and scene detection, facial analysis and comparison, celebrity recognition, text extraction (OCR), unsafe content moderation, and video-specific features like activity detection and person tracking across frames. Everything runs through simple API calls and is billed per image processed or per minute of video analyzed.

It's aimed at developers and businesses that need vision capabilities baked into an application without hiring a machine learning team: media companies moderating user uploads, retailers tagging product images, security teams flagging faces or objects in security footage, and marketing teams cataloging visual content at scale. Because it's a managed AWS service, it integrates directly with S3 for storage, Lambda for event-driven processing, and other AWS tools, which makes it a natural fit for teams already building on that stack.

What sets it apart from open source or DIY computer vision is that there's no model training or infrastructure to manage: you send an image or video and get structured results back. The tradeoff is less customization than a purpose-built model would offer, and accuracy on facial recognition in particular has drawn scrutiny and some public controversy over bias and law enforcement use, which pushed AWS to pause facial recognition sales to police for a period. It remains one of the most widely used off-the-shelf vision APIs given AWS's market reach.