brighter AI · Redact

Anonymization QA — Upload

Drop images with faces or licence plates. They run through brighter AI's pipeline and a quality & compliance report is generated.

MOCK (local Pillow effects — no API key)

Click or drop images here

JPG / PNG · up to 20 MB each · faces & licence plates

Plate region profile

Threshold sweep (optional)

Comma-separated. Empty = no sweep (fewer API calls). Each step ≈ 2 calls.

Ground-truth count (optional)

Real faces+plates in the sample → enables measured recall.

Volume to forecast (optional)

Customer frame count → sampled cost forecast from the local pre-screen.

€ / GPU-hour (optional)

Adds a € figure to the forecast.

Presentation

Live stages actually runs one stage per click: see the local detector first, then watch brighter AI run, then the QA comparison, then the decision. Walks one image.

Local pre-screen sensitivity

Lower = the YuNet pre-screen catches more small/distant faces (more false positives). Tunes Step 1 + the cost forecast — not what brighter actually redacts.

Full intelligence — detection counts, data-utility check, and per-image risk signals

⚠︎ The trial has a call-volume quota that replenishes every few minutes. Full intelligence ≈ 3 calls/image (anonymize) + 2 (counts) + 2 (utility) + 2 per sweep step. For a smooth live demo, start with 1–2 images and a short or empty sweep.

Running live anonymization — a few seconds per image…

View run history → · solution-engineering POC, not affiliated with brighter AI / Milestone Systems