SightRadar is a cloud face recognition API that indexes face embeddings, searches a photo collection from a single selfie, compares two faces 1:1, and detects faces — returning JSON in the same request and response shapes as AWS Rekognition.
What is SightRadar?
SightRadar is a hosted face recognition API that detects faces in photos, stores their embeddings in collections, and returns similarity scores you set your own threshold on. It takes photo bytes as input and returns Rekognition-shaped JSON containing values such as Similarity, FaceId, and SearchedFaceConfidence. It runs as a cloud endpoint at api.sightradar.com, called with the AWS SDK you already use — the page's example uses boto3 with the rekognition client. SightRadar also powers live event galleries, and the same engine is described as running on millions of photos in production.
What makes SightRadar stand out?
- Rekognition-compatible API — the request and response shapes match AWS Rekognition, so changing a single
endpoint_urlkeeps existing IndexFaces, SearchFacesByImage, and CompareFaces calls working. - Eight supported operations — CreateCollection, IndexFaces, SearchFacesByImage, SearchFaces, CompareFaces, DetectFaces, ListFaces / DeleteFaces, and ListCollections / DeleteCollection.
- Scores you threshold on — searches return per-match Similarity values (98.1, 97.4, 96.7 in the demo response) alongside SearchedFaceConfidence for the query image, so the cutoff is yours to set.
- Zero-face photos are valid — a photo containing no faces returns a valid result rather than an error.
- Billed per photo, never per face — control-plane calls are free and there are no minimums or subscriptions.
- Detect and compare stay in memory — those operations do not persist images; only indexed faces live in collections, which you can delete via API or console.
- Per-account isolation — each account's face vectors are namespaced and never cross-searchable, encrypted in transit, with scoped API keys.
Who is it for?
- Teams already running AWS Rekognition: swap the endpoint URL to reduce per-photo cost without rewriting index and search code.
- Event photographers and gallery operators: index a shoot so guests can find their own photos from one selfie.
- Identity verification developers: run 1:1 CompareFaces checks and enforce a fixed similarity threshold.
- Product engineers adding photo organization: index a library and retrieve every photo in which a person appears.
What can you do with SightRadar?
- Live event galleries: index photos as they arrive and let guests search an entire gallery from a single selfie; the page's demo returns matches in about 41ms.
- Migrate off AWS Rekognition: point the AWS SDK at https://api.sightradar.com and keep existing collection, indexing, and search calls unchanged.
- 1:1 verification: compare a live selfie against a stored face and accept or reject on a similarity threshold you configure.
- Quality gating: run DetectFaces to locate and quality-check faces without storing anything.
How does SightRadar work?
- Open the console, add credits, and copy your API key.
- POST photos to a collection — SightRadar detects, embeds, and stores every face it finds.
- Send a selfie to search, or compare two faces with the same calls you already use.
Pricing
SightRadar is pay-as-you-go: you are billed per photo processed, never per face, with no subscription, no minimums, and free control-plane calls. Real-time processing is $0.00093 per photo, batch is $0.00062 per photo, and search by face-id is $0.00062 per photo with storage at $6 per million faces per month. You can start free without a card, and a calculator on the pricing page compares rates against AWS Rekognition.
Alternatives
- AWS Rekognition — the API SightRadar is shape-compatible with; SightRadar states its batch rate is half the Rekognition rate.
- Azure Face API
- Face++
FAQ
Is SightRadar compatible with AWS Rekognition?
Yes. SightRadar uses the same request and response shapes as AWS Rekognition for operations including IndexFaces, SearchFacesByImage, SearchFaces, CompareFaces, and DetectFaces. You keep your existing AWS SDK code and change only the endpoint URL to https://api.sightradar.com.
How is SightRadar billed?
Billing is per photo processed, not per face, with batch at $0.00062 per photo, real-time at $0.00093 per photo, and search by face-id at $0.00062 plus $6 per million faces per month for storage. Control-plane calls are free and there are no minimums.
Does SightRadar store my images?
Detect and compare process images in memory and do not persist them. Indexed faces live in collections you can delete at any time through the API or console, and each account's vectors are namespaced and never cross-searchable. SightRadar states it does not train general-purpose models on your data.
What happens if a photo has no faces?
A photo containing zero faces returns a valid result rather than an error, so batch pipelines do not have to special-case empty or face-free images.
Are the demo faces real people?
No. The faces shown in the live demo are AI-generated, and the matched shots are the same generated identity with real appearance changes such as glasses, a clean-shaven face, aging, and different lighting. The page notes that published benchmarks on public datasets are coming.
