Consent is now a legal posture
Children's biometric and image data sits squarely inside modern privacy law. A single missed consent isn't an oversight any more — it's a notifiable issue.
On-device photo consent screening
Know who’s in every photo before you publish.
FaceGate screens your photos against an enrolled consent list — entirely on the device. No cloud. No third parties. No surprises.
- No cloud calls
- Runs on the phone
- Apache 2.0 model
Photo
Source image
Detect
Find faces
Embed
112 × 112 chip
Match
Cosine score
- Safe
Classify
Verdict
- 100% on-device
- No cloud calls
- Apache 2.0 production model
- Full audit trail
- Material 3 design
Why this matters
Every photo a school publishes might contain someone who didn’t consent.
Manual review doesn't scale
A sports day or production can produce hundreds of photos in an afternoon. Eyeballing each one for non-consenting faces is slow, tiring, and error-prone.
Cloud tools are the wrong trade
Most off-the-shelf face recognition uploads photos to a vendor's servers. Using one means sending children's faces to a third party to solve a privacy problem.
The risk isn’t a bad photo getting through — it’s not knowing you let one through.
The solution
A phone app that screens photos against a consent list — before they go anywhere.
Stays on the phone
Faces never leave the device. No cloud. No third parties.
Consent-aware
Tag people as Include or Exclude. The app handles the rest.
Fast on a batch
Scans dozens of photos and gives you a verdict per image.
Publish-ready
Cover faces with an emoji, export, done. Full audit trail.
Built so you can replace manual review with a structured, auditable workflow.
How it works
Same workflow you’d do by hand — just structured and fast.
1
Enrol
Add the people whose consent you're tracking. A few photos each is enough.
2
Scan
Point the app at a folder or take a photo. It detects and recognises every face.
3
Classify
Each photo lands in one of three buckets: Safe, Unsafe, or Review.
4
Publish
Cover non-consenting faces, export, post.
Every step runs locally on the phone — no internet required.
See the verdict
Flip a person’s status. Watch the photos re-classify.
Three illustrated photos, three enrolled people. Toggle Include ↔ Exclude and see how the verdict shifts — the same logic the real app applies on-device.
1 face
SafeWhy? All faces are on the consent list as Include.
2 faces
UnsafeWhy? Blake is Excluded and confidently matched.
3 faces
ReviewWhy? An unknown face was detected — flagged for human review.
Consent list
Illustrative. The same logic runs on your device with no network calls.
Alex
Blake
Casey- Not on list
Unknown
Confidence model
Tune how confident the app needs to be to call a match.
Below the uncertain threshold, the app routes to human review instead of guessing. The defaults below are tuned for AuraFace v1 — the production model.
0.30Uncertain0.35Confident
Cosine similarity ranges from 0 (different) to 1 (identical). The two handles split it into three bands: confident match, uncertain (human review), and no match.
Where it fits
A few of the verticals we work with.
Compliance posture
Designed against modern Australian privacy law.
FaceGate is built against the Australian Privacy Principles and the WA Privacy and Responsible Information Sharing Act 2024. All biometric data stays on-device. No stratified demographic cohort evaluation has been published yet — that’s a Q4 2026 deliverable, and we publish the first model card before any commercial sale.
APP-alignedST4S Readiness — in progressAuraFace v1 · Apache 2.0F1 = 0.920 · precision = 0.992
1FP / 1568 negatives
Internal 18-person ablation
0cloud calls / scan
All inference on-device
0.920F1 (AuraFace v1)
Apache 2.0 commercial model
≈ 5ms model inference
Lightweight, on a phone
Want to see it on your photos?
We're working with WA schools on early deployments. Tell us what you need.