MZSoft Studio
What PrivShot Can and Can't Detect: An Honest List
Detected automatically: faces, phone numbers, mainland-China ID numbers, bank card numbers, company credit codes, addresses, names (opt-in) and whole cards/documents (opt-in). Not detected: passport numbers, emails, usernames, plates, signs, tattoos, reflections. Use the review screen and the brush.
The problem
An app that says “automatically blurs personal data” invites you to trust it to find everything. No detector does. If you don’t know what a tool can see, you can’t know what you still have to check, and the thing you assumed was covered is the thing that gets posted.
Detected automatically
| Category | How it’s found | Default |
|---|---|---|
| Faces | Face detection, scanned in tiles at full resolution to catch small faces | On |
| Phone numbers | System phone-number detection on recognised text | On |
| ID numbers | Mainland-China 18-character resident ID format, accepted when the checksum or the embedded date is plausible | On |
| Bank cards | 15 to 19 digits that pass the Luhn checksum | On |
| Company credit codes | 18-character unified social credit code, validated by its check character | On |
| Addresses | System address detection on recognised text | On |
| Names | On-device language tagging with a confidence floor, plus labelled fields (“Name”, “Cardholder”) | Off (more false alarms) |
| Whole documents | Card/document shape detection | Off (very aggressive) |
Text recognition is tuned for Chinese (Simplified) and English.
Not detected
- Passport numbers (use whole-document blur or paint over)
- Email addresses and usernames
- Licence plates (not built yet)
- Signs, house numbers, badges, tattoos, logos
- Faces in reflections, or heavily hidden, tiny or shadowed faces
- Text OCR can’t read: glare, blur, odd fonts, handwriting
- Legacy 15-digit ID numbers (deliberately skipped, because they collide with ordinary numbers)
Why the gaps are honest
Detection errs on the side of covering more, and the review screen exists so you can fix errors in either direction. But a detector can’t flag what it doesn’t recognise, so the rule is simple: the review screen is the safety check, and you are the last step.
A routine that works
- Run the scan.
- Zoom in and look at the whole picture, not just the highlighted parts.
- Check each category’s detections, ticking and unticking.
- Switch to Manual for anything the list above says isn’t covered.
- Save, then glance at the saved copy before you send it.
Related
FAQ
Does it detect passport numbers? Not automatically. Use whole-document blur and/or paint over the number.
Does it detect licence plates? Not yet. Cover them with the brush or rectangle.
What languages does text recognition use? It is tuned for Chinese (Simplified) and English text.
Why are names and whole documents off by default? They produce more false alarms, since they cover a lot. They are detected anyway so you can switch them on per photo.