What is face similarity?
A face similarity checker turns each face into an embedding, a short list of numbers that describes its features, and measures how close the two embeddings are. SightRadar reports that as cosine similarity on a 0 to 1 scale. Photos of the same person land close together even across glasses, beards, lighting and a few years of ageing. Different people land further apart, including lookalikes.
When an image holds more than one face, the largest, most confidently detected face is used. The images are fetched, compared in memory and discarded. Nothing is written to a collection.
How to read the score and threshold
The score is a measure of resemblance, not a verdict. The tool also shows a match decision made at a default threshold, so you can see which side of the line a pair falls on.
- High score, above the threshold: very likely the same person.
- Near the threshold: treat it as uncertain. Ask the person to confirm, or send it to review.
- Low score: very likely different people.
The right threshold depends on your photos and on what a mistake costs you. Strict for identity checks, looser for photo discovery. The thresholds guide shows how to pick accept, review and reject bands from your own labelled pairs. Scores from other vendors do not carry over: AWS Rekognition reports 0 to 100, and its operating point is different.
What people use face comparison for
- Identity verification: match a selfie to the photo on an ID during onboarding, alongside document and liveness checks.
- Account recovery: confirm the person regaining access is the person who enrolled.
- Marketplace and dating trust: check that profile photos show the verified account holder.
- Attendance and access: confirm a check-in photo matches the enrolled person.
- Photo de-duplication: spot when two uploads show the same person.
Compare is for one-to-one checks. To find a person across thousands of photos, use face search on a collection instead; see use cases. Use face comparison only with a lawful basis and, where required, consent; see trust and responsible use.
Compare faces with the API
This page calls the same endpoint you would: POST /v1/compare, the SightRadar equivalent of AWS Rekognition CompareFaces. Two images cost 186 credits (93 per image processed). See the compare endpoint reference for the full request.
curl -X POST "https://api.sightradar.com/v1/compare" \
-H "Authorization: Bearer $SR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"sourceUrl":"https://example.com/a.jpg","targetUrl":"https://example.com/b.jpg"}'Frequently asked questions
- What is face similarity?
- Face similarity is a number that says how alike two faces are. Each face is turned into an embedding, a list of numbers that summarises its features, and the two embeddings are compared with cosine similarity. The result runs from 0 to 1: scores near 1 mean the faces are very likely the same person, and low scores mean they are likely different people.
- Does a high score prove two photos show the same person?
- No. A high score means the faces are very similar, which is strong evidence but not proof. Identical twins and close relatives can score high, and poor photos can lower the score for the same person. For identity verification, combine the face match with document checks and liveness.
- What similarity threshold should I use?
- Each result shows the default threshold used to decide match or no match. For a real product, pick your own threshold from labelled pairs of your own photos: strict for identity verification, looser for photo discovery where a person confirms the suggestion. The thresholds guide walks through it.
- Are my images stored?
- No. Compare processes both images in memory to compute the score and then discards them. No photo, face crop or faceprint is saved, and nothing is used to train models.
- Why does the tool ask for image URLs?
- This tool calls the public compare endpoint, which fetches each image from a public https URL. Use links to JPG, PNG or WebP files that open without a login. The sample pair button loads two synthetic faces if you just want to see it work.
- Is the face compare tool free?
- Yes. Sign in with a free account and each comparison is charged against the free credits on that account (186 credits, one 93-credit charge per image). There is a fair-use limit per visitor to keep the tool available for everyone.
- Can I run face comparison from my own app?
- Yes. The same operation is available as POST /v1/compare in the SightRadar API, the equivalent of AWS Rekognition CompareFaces. Create a key in the console and call it from any HTTP client or the official Python and Node SDKs.