Blur faces

Detecting faces automatically

Face detection uses the YuNet model, which Verpixle integrates through OpenCV (cv2.FaceDetectorYN). It is built to find faces that are only a few pixels wide – passersby in the background, people in the second row, spectators in a crowd. It works just as reliably on half-turned heads and pure profile shots, which is exactly where retouching by hand eats up the most time. Partially covered faces – a cap, glasses, a mask or hair in the picture – are usually caught as well. Its limits show with heavy motion blur and harsh backlighting: what the eye itself can barely read as a face may be missed by a model too.

This grading follows how the YuNet model is designed and how typical shots behave; it is not a guarantee for any individual image. Look at the result before you publish, and when in doubt choose the strongest level.

How it works

1. Select a file. Drag your photo or video into the box on the homepage. JPG, JPEG, PNG, MP4 and MOV up to 200 MB are supported – no account needed.
2. Tick “Faces”. Detection starts at the push of a button and marks every face it finds. Three levels – light, medium, strong – set how coarse the blocks turn out.
3. Check the result and save it. The anonymized image keeps its resolution. The uploaded file and the result disappear from the server automatically after no more than 24 hours.

When do faces have to be made unrecognizable?

There is no duty to black out every face – it depends on whether a person is recognizable and whether you distribute the image. In Germany, § 22 KUG (Kunsturhebergesetz, the German act on the right to one's own image) generally requires the consent of the person portrayed before their likeness is distributed. § 23 KUG names exceptions, for example likenesses from the sphere of contemporary history, or shots in which people appear merely as incidental extras alongside a landscape or another location. Whether someone in the background really is a mere incidental extra is decided case by case.

The GDPR applies alongside it across the EU: a recognizable face is personal data, and anyone who processes or publishes it needs a legal basis under Art. 6 GDPR – usually consent, or a legitimate interest that outweighs the interests of the person portrayed. Anyone who has other people's faces made unrecognizable before an image moves into a blog, a shop, a marketplace or a social network usually never has to make that assessment in the first place. The pragmatic route is therefore: anonymize people first, publish afterwards.

Note: This is not legal advice. For a binding assessment of your specific case, consult a lawyer.

Group photos, events and club photos

The more heads there are in a picture, the more tedious manual work becomes. That is exactly where automatic detection pays off: on a group photo, Verpixle covers every face it finds in a single pass – whether that is ten people lined up as a team or several dozen in a hall. So you do not have to blur one face after another; you handle the whole image at once.

• Club and team. If not everyone involved has given consent – for minors, that means their legal guardians – the picture can still go on the club website in anonymized form.
• Company party, conference, trade fair. Event photos almost always capture guests nobody asked. For recaps, newsletters and social media posts, blurred is by far the simpler option.
• School, daycare, youth work. The standard is especially strict for children and teenagers – here it is worth making faces unrecognizable as a matter of course.
• Vacation, street, town festival. Share a photo from a pedestrian zone and you share the people in it too. One pass is enough to anonymize every passerby.

Blurring faces in video

The same principle applies to moving footage, just more often: Verpixle evaluates every single frame anew and places the blocks over the faces found in it. If someone walks through the shot, their head stays covered throughout, without you dragging areas around by hand. MP4 and MOV files are accepted up to 200 MB, up to 120 seconds of running time and up to Full HD (1920 × 1080); resolution, image orientation and audio track are left untouched. Trim longer clips to the relevant section beforehand. The Anonymize video page explains how that works in detail.

Blur or black out?

Both methods pursue the same goal and still look different. Blurring breaks the area down into coarse blocks: colors and outlines roughly remain, the photo looks less cut up and still shows that a person stood there. Blacking out replaces the area with an opaque bar – as clear as it gets, but visually a foreign body. Verpixle works with the first variant and offers three levels, from light to strong.

What decides the effect is the strength you choose. If the blocks stay too fine, the eye region and the shape of the face can still be guessed – for large, frontal faces the strongest level is therefore the safe choice, while for small heads in the background the medium one usually does. Whatever the method, one rule also holds: pass on the anonymized version only. Sending the original alongside it undoes the whole effort.

Common questions

Blur a face now

Blurring faces takes only a few seconds: select a file, tick “Faces”, save the result – no account and no installation.

Last updated: July 26, 2026