AI Auto-Redact - Censor Faces Automatically
Automatic face detection, then a permanent blur
Face detection here runs on BlazeFace, a compact neural network built for real-time use, executed through TensorFlow.js on your GPU. It scans the image for faces and returns their positions, and the tool then rasterises a heavy blur over each region it found. The result is a one-click way to anonymise a group photo instead of drawing rectangles by hand.
The important property is that the blur is destructive. Once you download the file, the obscured pixels have genuinely been replaced, so the faces cannot be recovered from the exported image by adjusting contrast or running an unblur filter. That is a meaningful difference from a black box drawn on top in a PDF or a slide, where the original picture often still sits underneath the annotation and can be extracted.
It suits classroom and school photos shared with parents, event and conference pictures posted publicly, street photography where passers-by were caught incidentally, property listings and internal documentation, and anything going onto a public site where the people in frame did not agree to appear. After the automatic pass, glance over the preview and confirm every face you want covered has a blur on it, then download.
Everything, including the model itself, runs in your browser. The photos you are anonymising are by definition the ones you would least want sitting on someone else's server, so nothing is uploaded and there is no account or usage log. Pair it with the metadata remover to strip GPS and timestamp data from the same image before you share it.
How to blur faces in a photo
- Upload the photo. All the usual formats work. The first run loads the detection model, which then stays cached.
- Click Detect and Censor Faces. The model scans the image and blurs every face it locates.
- Check the preview. Confirm each face you want covered has a blur over it. Faces at a steep angle, very small in frame, or partly hidden are the ones worth looking at twice.
- Download. The blur is baked into the pixels of the exported file, so it cannot be reversed afterwards.
For anything the automatic pass does not cover, or to obscure things that are not faces such as a licence plate, house number, name badge or screen, use the manual privacy blur to paint over regions yourself. There is also a dedicated licence plate blur. See how to blur faces and text in photos for the full guide.
Frequently Asked Questions about Image Redaction
Is the blur permanent?
Yes. Once you download the result the obscured pixels have genuinely been replaced, so the faces cannot be recovered by adjusting contrast or running an unblur filter. This differs from a black box drawn over a PDF or slide, where the original image often still sits underneath.
What model does the detection use?
BlazeFace, a compact real-time face detector, run through TensorFlow.js on your GPU. It loads once and is then cached in your browser.
Which faces should I double-check?
Faces at a steep angle, very small in the frame, or partly hidden behind something. A quick look over the preview before downloading catches these, and the manual blur covers anything left.
Can I blur things that are not faces?
Use the manual privacy blur to paint over licence plates, house numbers, name badges or screens, or the dedicated licence plate blur.
Does it work on a phone?
Yes. The page is responsive and the model runs on the phone's own hardware, so you can redact straight from your gallery.
Are the photos uploaded?
No. The model and the blurring both run in your browser, with no account, no usage log and no server copy. Consider also stripping GPS data with the metadata remover before sharing.
🔎 Verify it yourself: nothing uploads
Don’t take our word for it. Open your browser’s DevTools (F12 → Network tab), then run this tool on any image. The only download you’ll see is the open-source AI model coming to your browser — your image itself is never sent anywhere. Every tool on this site works the same way — all processing happens on your device.