Pixelation is not always redaction
For a face or a house, a coarse mosaic is fine — the information is genuinely destroyed. For a short, predictable string it is not. Researchers have recovered pixelated text by generating every candidate, pixelating each the same way, and matching. Anything with few possibilities — a six-digit code, a car registration, a date of birth — is vulnerable to that.
If what you are hiding must stay hidden, draw a solid block over it rather than blurring it. A black rectangle has nothing to reverse.
Choosing the block size
The mosaic has to be coarse relative to what it covers. A face needs blocks big enough that eyes and mouth become single squares — roughly a tenth of the face width. Too fine and the features are still legible at full zoom, even if they look hidden in the preview.
Zoom in on the result before you publish it. A preview scaled down to fit the screen hides detail the full-size file still contains.
The part people forget
Covering the pixels does not remove the metadata. A photograph can carry the exact coordinates it was taken at, which may give away the address you just blurred out of the picture. The saved file here is written fresh and drops EXIF, but if you edit elsewhere, check with the EXIF viewer before sharing.
Frequently asked questions
Can pixelated text be recovered?
Sometimes, when there are few possible answers — a PIN, a plate, a short code. The attack regenerates candidates and matches the pattern. Use a solid block for anything that must stay secret.
How coarse should the blocks be?
Coarse enough that the features become single squares — around a tenth of the width of whatever you are hiding. Check at full zoom, not in the scaled preview.
Does it remove location data too?
The file saved here is written fresh, so EXIF including GPS does not carry over. Worth checking if the image passed through other software first.
Is my image uploaded?
No — which is the point, given that the whole reason you are here is that the picture contains something you do not want seen.
Something wrong with this tool, or an idea for it? Tell us