Azure AI Face Review

Azure AI Face (part of Azure AI Vision, now under Microsoft Foundry Tools) detects faces in an image and returns a face ID used by every other operation. From there it can verify whether two faces belong to the same person (1:1), identify a face against a person group of up to 1 million persons with up to 248 faces each (1:N), find similar faces, group unknown faces, and run liveness detection against photos, screens and masks. You call it through the REST API or client SDKs, or try it in Vision Studio.

Access is restricted. Per Microsoft's Limited Access page, identification and verification require a registration form, are available only to customers managed by Microsoft (working directly with Microsoft account teams) for approved use cases, and do not run on the Free (F0) tier. Face detection works without registration. Microsoft has retired emotion and gender inference and limited age, smile, facial hair, hair and makeup attributes, and US police departments may not use the service.

Face is billed per transaction: each API call counts as one, and training a large person group or face list counts one transaction per 1,000 images trained. US East list prices:

What that means: Microsoft's docs say detection is the first step in every other scenario, so a 1:N identification of a new photo is at least a Detect call plus an Identify call: 1,000 x 2 x $0.001 = $2.00 per 1,000 searches. Storing 10,000 faces costs 10 x $0.01 = $0.10 a month.

The pricing page also lists enterprise commitment tiers; check it for your region and agreement. Prices checked on the vendor's pricing page on November 12, 2026. See Azure Face API pricing.

Operations described in Microsoft's Face documentation:

Azure AI Face fits organisations already working with a Microsoft account team that need identity verification with liveness, touchless access control or check-in for opted-in users, and can pass Microsoft's use-case review. Microsoft puts consent duties on you: its docs say you must give notice, obtain consent and delete biometric data as data protection law requires.

If you cannot get Limited Access, or want to start today with a credit card, Amazon Rekognition offers comparison and search with no approval step. To keep faces on your own servers, self-host CompreFace or DeepFace. For emotion or gender attributes, which Azure retired, Face++ and Luxand.cloud still list them.

Azure Face API has a small review base: 12 reviews on G2 at 3.8 out of 5 at the time of checking, with praise for the range of attributes and complaints about accuracy, pricing and a confusing portal. The most common developer questions on Microsoft Q&A are about whether comparing two images needs Limited Access (it does) and about slow responses to access requests.