Django Camera Kit¶

Document scanning and KYC face verification for Django.

Two problems come up constantly in Django apps that deal with real-world documents and identity: camera-based document capture (edge detection, multi-page flow, PDF export) and face verification (selfie vs. ID photo, liveness). Neither has a solid, drop-in Django package. Django Camera Kit is both, as two independent modules.

Document scanning¶

DocumentScannerWidget replaces a plain <input type="file"> in any existing form:

from django import forms
from django_camera_kit.widgets import DocumentScannerWidget

class ContractForm(forms.ModelForm):
    class Meta:
        model = Contract
        fields = ["signed_document"]
        widgets = {"signed_document": DocumentScannerWidget()}

It opens the camera, detects the document's edges live (OpenCV.js, fully client-side), lets the user drag the corners when auto-detection misses, chains multiple pages, and assembles a single PDF — written back into the same file input. No new models, no endpoints, no migrations. See Document Scanning.

KYC & identity verification¶

KYCVerificationWidget drives a live selfie + ID capture flow with a head-movement liveness challenge, then posts to a server-side endpoint that extracts face embeddings (YuNet + insightface) and compares them — matching only ever happens server-side.

from django import forms
from django_camera_kit.kyc.widgets import KYCVerificationWidget

class SignupForm(forms.Form):
    kyc_verification_id = forms.IntegerField(
        widget=KYCVerificationWidget(verify_url="/kyc/verify/")
    )

Requires pip install django-camera-kit[kyc], PostgreSQL with the pgvector extension, and django_camera_kit.kyc in INSTALLED_APPS. See KYC & Identity Verification.

Why not Haar cascades?¶

The face detector is YuNet (cv.FaceDetectorYN), not the classical Haar-cascade approach most tutorials use. Haar cascades detect faces from brightness-contrast patterns and are well documented to have significantly higher miss rates on darker skin tones and in low-contrast lighting — a bias traced back to the demographic composition of their historical training sets (see Buolamwini & Gebru, "Gender Shades", 2018). YuNet is a small CNN trained on WIDER FACE, a much larger and more diverse dataset, and doesn't carry that specific failure mode.

Installation¶

pip install django-camera-kit          # scan module only
pip install django-camera-kit[kyc]     # scan + KYC

See Installation for the full setup, including the vendor files the scan module needs.