Amazon Rekognition Review

Amazon Rekognition is a set of pretrained image and video analysis APIs inside AWS. For faces it covers detection and attributes (DetectFaces), one-to-one comparison of two images (CompareFaces), and one-to-many search: you add faces to a collection with IndexFaces, optionally group them into users, then query with SearchFacesByImage or SearchUsersByImage. Face Liveness checks a short selfie video for spoofs such as printed photos or masks. The same service also does labels, text, moderation, celebrity recognition and stored or streaming video.

You call it through the AWS SDKs or CLI with an AWS account and IAM permissions; there is nothing to host. Collections hold face vectors that you control, not the images themselves. Under the AWS Service Terms (section 50), AWS keeps a moratorium on police use of face comparison in criminal investigations, bars sustained surveillance of a specific person without independent review such as a court order, and applies separate Biometric Notice and Consent terms.

Rekognition Image is billed per image processed, in monthly volume tiers. Face APIs fall into two groups: Group 1 (CompareFaces, IndexFaces, SearchFacesByImage, SearchFaces, SearchUsersByImage, SearchUsers, AssociateFaces, DisassociateFaces) and Group 2 (DetectFaces, plus labels, text, moderation, celebrities and PPE). Running two APIs on one image counts as two images.

US East (N. Virginia) rates:

What that means: 1,000 face searches cost 1,000 x $0.001 = $1.00. Building a 10,000-face collection costs 10,000 x $0.001 = $10 to index, then 10,000 x $0.00001 = $0.10 a month to store.

Free tier: for 12 months from account creation, 1,000 Group 1 and 1,000 Group 2 images a month, 1,000 face vectors and 1,000 user vectors stored, and 60 minutes of video a month. New AWS customers since July 15, 2025 also get up to $200 in Free Tier credits. Prices vary by region.

Prices checked on the vendor's pricing page on November 12, 2026. See Amazon Rekognition pricing.

Face-related capabilities listed on AWS's product and pricing pages:

Rekognition suits teams already on AWS that want face verification or search on their own users' photos with no servers to run and a per-image bill: identity checks at onboarding, photo de-duplication, access control with opted-in users. At $1 per 1,000 searches it is among the cheapest managed options at low volume.

If you cannot send face images to a US cloud, or want to own the model, self-host CompreFace or DeepFace. On Azure, Azure AI Face is the equivalent but identification and verification require Limited Access approval. For a flat monthly plan with a small quota, Luxand.cloud is simpler to start with.

Reviewers on G2 (4.3 out of 5 from 28 reviews at the time of checking) praise recognition accuracy and few false positives, and mention a learning curve and price as drawbacks. A developer who compared five face APIs on Medium describes Rekognition as strong for large-scale stored-image and video work but tied to AWS. FAUN.dev scores it highest on task coverage and integrations and lowest on portability, since every call runs through an AWS account and IAM.