Covering something in a photo, and actually covering it
There is a difference between hiding something and removing it, and it is the difference between a screenshot being safe to post and a lawsuit. Many tools draw a shape on top of the image and export a file where the original is still underneath. Here the effect is written into the pixels: once you download, there is no layer to peel off, because there is no layer.
Pixelating text does not work, and this is not an opinion
Pixelation is reversible for text, and the attack is simple enough to be a weekend project. The attacker takes the font and size from the uncensored text around it, generates every plausible candidate, applies the exact same pixelation to each, and compares. When a candidate produces the identical blocks, the hidden text is recovered. Published tools do this today. Blur behaves the same way for the same reason: it is a deterministic transform of a small set of possibilities.
So when is blur fine?
When what you are hiding is not drawn from a small set. A face, a room, a logo, a street behind someone — blur and mosaic work there because an attacker has nothing to enumerate. What they cannot survive is anything with a known alphabet and a fixed shape: document numbers, card numbers, plates, addresses, names. For those, the black bar is the only honest answer, and it is the default here for exactly that reason.
The metadata nobody remembers
A photo from a phone usually carries EXIF: the camera model, the exact timestamp, and very often the GPS coordinates where it was taken. Covering a face in an image and then posting it with the coordinates of the house attached defeats the purpose. Rebuilding the file from the canvas discards all of it, so the download is clean whichever mode you pick.
Why it runs in your browser
The pictures worth censoring are the ones you least want on a stranger’s server: documents, children, medical screens, a card in your hand. Uploading them to an online censoring tool to make them private is a contradiction. Here the file is read into memory, drawn onto a canvas and exported locally; no request carries the image.
A short checklist before you post
Zoom in before you drag. A box that looks like it covers a plate at 30% zoom often leaves two readable characters at full size, and the person checking will be at full size. Cover the whole object, not the readable part of it: half a badge, a reflection in a window, a name on a lanyard behind the subject. Remember that context identifies people too — a covered face next to an uncovered street number is not anonymous. And if the image is a screenshot of a document rather than a photo, consider whether you should be sharing the screenshot at all instead of a version you produced on purpose.
Want to go deeper? Read our guide: How to hide sensitive information in a screenshot (the safe way)