1
Enroll the face
We keep a small face crop and a numeric signature. Later successful checks can add up to five angles for that same person.
Face recognition for live services
Confirm the person on camera with a guided video.
Checkface is a face recognition system for fintech, onboarding, account recovery, staff access, and any service that needs to verify a face. The customer records one short clip, follows the prompts, and your app receives a match and a liveness score.
FintechOnboardingAccount recoveryStaff accessAny face check
FaceNet matches the person. MediaPipe scores the movements in the clip.
Live videoReady
Look straight ahead
One clip. Each prompt lasts two seconds.
Liveness score—
A sample of the prompts your customer follows in one continuous clip.
For fintech and any service that verifies a face.
Account opening, step-up login, payout approval, clinic check-in, staff gates. Checkface sits beside your app and returns a decision over HTTPS.
2
Record one clip
The server sets the order. The customer looks ahead, blinks, turns, opens their mouth, and smiles, two seconds each, in a single video.
3
Score and match
We sample the video, score each movement, and compare the face with FaceNet. A pass needs 75 overall and 0.60 on every prompt.
What the video measures
The clip is split into equal slices. Each slice produces a score from 0 to 1. The liveness result is the average, shown as a percentage, alongside the FaceNet match.
How teams run it
Checkface runs on your server. You license the setup, then your service calls the video check whenever a person must be confirmed.
Sandbox
Prove the flow with your engineers.
- API on your staging host
- Enrollment and liveness session
- Swagger docs
Launch
The package most services start with.
- Production HTTPS deployment
- Bearer token for your app
- Per-user signature gallery
- Score thresholds you can tune
Scale
When more than one product calls the same check.
- Separate keys per product
- Threshold review on real traffic
- Handover for your mobile team
Two calls from your app
Open a session, then upload the guided clip. We read the frames we need, score the movements, match the face, and drop the file.
POST /liveness/session user_id=cus_2041 → challenges, seconds_per_challenge POST /liveness/video session_id=... clip=@check.mp4 → authenticated → liveness_score → challenge_scores → gallery_size
How the check works in practice
What do you keep after enrollment?
One small face crop and up to five numeric signatures for that person. The verification video is scored in memory and then removed.
What does the video check include?
A continuous clip with a look-ahead, a blink, a head turn, an open mouth, and a smile. The order changes each session. Movement through the clip produces the liveness score, and FaceNet confirms it is the enrolled person.
How do results improve?
Each accepted check can add a new angle to that person’s signature gallery, so later matches reflect the faces you have actually verified.
How does a service connect?
API docs are at /docs. Request access and we issue a bearer token for your app.
Request access
Tell us the service you run and how many face checks you expect. We reply with a sandbox token and a walkthrough of the video flow.