SightRadar documentation
A face recognition API you can integrate in an afternoon. Index faces, search a collection with a selfie, compare two faces, or detect faces without storing anything. Billed per photo, on prepaid credits.
SightRadar is a REST API for face recognition. You send photos, it detects every usable face, stores a compact faceprint in a collection, and lets you find every photo a person appears in from a single selfie. It is request-and-response compatible with AWS Rekognition's face operations, so an existing integration can move over with an endpoint change.
Quickstart
Key, collection, index, search, wallet. Four calls, ten minutes.
Core concepts
Collections, faceprints, similarity scores, and the four operations.
API reference
Every endpoint with parameters, bodies, and status codes, generated from the OpenAPI spec.
Playground
Call the live API from your browser with your own key.
What you can build
| Operation | Endpoint | What it does | Stores data? |
|---|---|---|---|
| Index | POST /v1/collections/{id}/index | Detects every quality-gated face in a photo and stores its faceprint. | Yes, the faceprint. Never the photo. |
| Search | POST /v1/collections/{id}/search | Finds the photos a person appears in from one selfie. Ranked by similarity. | No |
| Compare | POST /v1/compare | One-to-one similarity between two faces, for identity verification. | No |
| Detect | POST /v1/detect | Locates and quality-gates faces without storing anything. | No |
Everything else is control plane: collections, batches, webhooks, keys, wallet and usage. Control-plane calls are free.
How the pieces fit
Create a collection
A collection is the namespace your faceprints live in. One per event, per venue, per tenant. Creating one is free.
Index photos
Send each photo by URL, as multipart, or as raw bytes. For thousands of photos, submit a batch and receive one webhook per photo.
Search or compare
Search a collection with a selfie to get ranked photo_ids back. Or compare two images directly for a yes/no with a calibrated score.
Set your own threshold
Scores are cosine similarity from 0 to 1. Pick the cut-off from your own matched and unmatched pairs. Read choosing a threshold before shipping a hard number.
Coming from AWS Rekognition?
The request and response shapes for CreateCollection, IndexFaces, SearchFacesByImage, SearchFaces, CompareFaces, DetectFaces, ListCollections and DeleteCollection are supported. On Python and boto3, the Rekognition shim makes it a two-line change. The migration guide covers everything else, including the one step nobody can skip: re-indexing.
Where to get help
- Something unclear in these docs? Email support@sightradar.com.
- Live incidents and uptime: status page.
- Consent, retention and lawful use: trust and responsible use.