Anonymise photos without sending them anywhere
Most online face-blurring tools require you to upload your photo to their servers — which is awkward when the whole point of blurring the photo is that you don't want it out in the world. This tool flips that around: the face-detection neural net runs in your browser via TensorFlow.js, the blurring is done by your own browser's canvas, and your photo never travels over the network. We don't even log the filename.
Auto, manual, or both
Auto mode uses the TinyFaceDetector model — a fast, lightweight neural network that finds faces at a wide range of scales and angles. After detection, every face shows up as a red dashed box with an X handle to remove false positives.
Manualmode lets you click-and-drag rectangles over anything that needs to disappear — license plates, T-shirt graphics, badges, signs, screens, bystanders. Manual regions show up in blue so you can tell them apart from the AI's detections.
Both mode is what most people want: run auto detection, then paint extra rectangles over whatever the AI missed.
Tuning the blur intensity and shape
The intensity slider controls the pixelation block size. Low values (5–10) produce a subtle pixelation that still suggests a face shape; high values (20+) reduce the area to a few coarse colour blocks — the look you typically see on the news. Pick high values for true anonymisation; the lower end can sometimes be reversed by modern AI tools, so don't rely on it for sensitive material.
The shape control toggles between an oval blur (clipped to an ellipse — the natural choice for faces) and a rectangle (the choice for licence plates, signs and other rectangular objects).
Privacy-first journalism, parenting and street photography
Use cases this tool was built for:
- Journalists protecting sources — blur a source's face before publishing without ever sending the original to a remote service.
- Parents sharing kid photos — anonymise other children in school events, parties or sports matches.
- Street photographers — comply with GDPR when including identifiable bystanders in a publishable shot.
- Real-estate listings — blur cars, doorbells with names, and overheard street numbers.
- Surveillance/CCTV stills — blur faces before forwarding stills to insurance, building managers or social networks.
- Sensitive corporate screenshots — quickly cover faces in employee photos when sharing externally.
What the AI can and can't do
TinyFaceDetector is good but not perfect. It handles the standard cases (front-facing portraits at decent resolution, groups in ordinary lighting) very well. It struggles with very small faces (under ~40 px wide), strong profiles, partial occlusion (sunglasses + masks together can confuse it), and anything heavily stylised like sketches or anime. When you're relying on this for real privacy work, do a visual pass after the auto detection to confirm every face is covered — Manual mode is one click away.