Censor anything in an image

Upload the image above, drag a box over the part you want gone, and pick Pixelate, Blur or Black bar. Add as many boxes as you need, then download the censored PNG.

Black bar, always. Text is a short string in a known font, so pixelation can sometimes be worked backwards. A solid bar removes the pixels entirely and leaves nothing to reconstruct.

Yes, if you go heavy enough. Aim for a face region that ends up no more than about 8–12 blocks across. If you can still see where the eyes are, increase the strength.

No. All processing happens locally in your browser — the uncensored original never leaves your device.

Yes. Draw as many boxes as you need, each with its own style, and use Undo or Clear to fix mistakes.

The exported PNG is redrawn from a canvas, so it does not carry the original photo’s EXIF metadata across — but do not rely on that as your only privacy step if location data is the concern.

Yes — completely free, no sign-up, no watermark.

Often, yes — which is why it should never be pixelated in the first place. Text is drawn from a small alphabet in a predictable font, so a pixelated block can be compared against every candidate string rendered the same way until one matches. Public tools that do exactly this to screenshots have existed since 2020. Blur has the same weakness. Use Black bar for anything made of characters, and pixelation only for faces, bodies and scenery.

Not in the sense people usually mean. Pixelation throws the original pixels away, so there is nothing left to restore. What the “unpixelate with AI” tools do is generate a plausible-looking face or texture that would pixelate down to the same blocks — an invention, not a recovery. A widely reported 2020 face-upscaling demo made the point badly and publicly when it turned a pixelated photo of a well-known Black politician into a white man. Treat their output as a guess, never as evidence of what was underneath.

For faces and general detail in a photo you are posting publicly, yes — if you go heavy enough that the region is only a handful of blocks across. Two real caveats, though. Researchers have shown that pixelated faces can often be matched back to the right person when the attacker already has a small set of candidate photos — that is re-identification, not reconstruction, but the outcome is the same. And pixelated text is genuinely recoverable: published tools can reconstruct pixelated screenshot text when the font and rendering are known. For names, numbers, addresses and licence plates, use a solid black bar instead.

No. The classic eye bar is a convention, not a protection — jawline, hairline, ears, build, tattoos, clothing and the setting all identify people, and anyone who already knows the person will recognise them instantly. If the goal is genuine anonymity, hide the whole face rather than a strip of it, and think about the badge, the number plate and the house number in the background too.

It depends where you are and why you are publishing, and this is not legal advice — but the direction of travel is clear. Under the GDPR a recognisable face counts as personal data, so publishing an identifiable stranger without a lawful basis can be a violation, including for photos taken in public. Anonymising the image is the usual answer when consent is not practical. Plenty of places outside the EU have narrower rules, and journalism and personal use are often treated differently. If in doubt, blur them — it costs you nothing and it is what most publishers do.

Most of them do — they receive your file, process it on their servers and hand back a link, which means a copy exists on someone else’s machine, usually under terms that let them retain it. That is a poor trade when the reason you are pixelating is that the image shows something private. Pixelateme is built the other way round: the work is done by your own browser with the Canvas API, so nothing is transmitted. If you want to prove it, load this page, switch off your network, and it still works.