@verifner/web

Guided capture for the VerifNer verification API: document front (and back for a CNI), selfie, quality gates in the browser, compression for 3G, upload with a short-lived client token. No framework, no dependencies, one file.

Flow

1. Your server calls POST /v1/sessions with the secret key and gets { id, client_token } (15 minutes, single use).

2. Your page mounts the SDK with those two values. The secret key never reaches the browser.

3. The SDK uploads to POST /v1/sessions/:id/verify and calls onComplete.

4. Your server trusts the webhook or GET /v1/sessions/:id, not the browser.

Or npm install @verifner/web, then import { create } from '@verifner/web'.

Screens

Welcome (steps, duration, consent, language) → document (skipped when documentType is set) → how to prepare it, with tips and an illustration → camera (the front, then "turn your card over" and the back for a CNI) → a checklist to review each photo → selfie tips → selfie → sending → result. Text is set in Satoshi, loaded from verify.verifner.com (font changes it). By default the person sees "verification sent"; showVerdict: true shows the verdict (verified, in review, not confirmed), never its reasons. Your server decides from the webhook either way.

Quality gates

Each frame under the guide is scored: Laplacian variance (blur), blown-out pixel ratio (glare) and mean luma (dark). Six clean frames in a row (about 1.5 s) trigger auto-capture for documents; the selfie is taken with the button. Captures are cropped to the guide, scaled to 1600 px on the long edge and re-encoded to about 300 KB. Without a camera the file picker takes over, through the same resize.

Develop

npm run build --workspace=@verifner/web      # dist/verifner.js, dist/verifner.iife.js, types
node --import tsx --test packages/sdk-web/test/*.test.ts
VERIFNER_API_KEY=vn_test_… node packages/sdk-web/example/server.mjs   # http://localhost:3000

Face detection

The selfie guide finds the face with a TypeScript port of pico.js and its facefinder cascade, both by Nenad Markus under the MIT licence (https://github.com/nenadmarkus/picojs). The cascade (about 240 KB) is served from verify.verifner.com; set faceModel to host it yourself, or false to turn the face checks off.

Licence

MIT.