Blur, Pixelate & Black Out Screenshots

Hiding something is not the same as destroying it

Every mode here makes the covered area unreadable to you, looking at the screen. Only two of them make it unrecoverable. That difference does not show up in the preview, which is why it is worth stating before you pick a mode.

Blur and pixelate both work by averaging. Pixelate divides your selection into blocks and paints each block one flat colour; the blur here does the same thing and then smooths the result so the edges stop looking blocky. In both cases the colour that gets painted is derived from the pixels that were there. The detail is gone, but a summary of it remains — and a summary is enough when there are only a few thousand possible answers.

Blackout and Whiteout do not average anything. They fill the selection with one fixed colour, so every pixel in the area ends up the same value regardless of what it was. Reading the exported file back gives a single flat colour and no variation whatsoever. There is no signal left to attack.

How a blurred phone number gets read back

The attack needs no special software and works on any screenshot whose font and layout are recognisable — which is most of them, because they come from apps everyone has seen.

Say you blur the last five digits of a mobile number. Those digits have 100,000 possible values, and a screenshot from a known app tells the attacker the font, the size, and the position. So they render +91 98765 00000, apply the same blur setting, and measure how far the result sits from your blurred patch. Then the next candidate, and the next. The real number produces the closest match, usually by a wide margin, because it is the only one that averaged to exactly those colours. On a laptop this is seconds of work, not a research project.

The reason this keeps surprising people is that the blurred image genuinely looks destroyed. Nothing about it hints that the numbers are still recoverable, and the attack does not undo the blur — it re-does the blur on guesses until one lines up. That is also why turning the strength up does not straightforwardly fix it: a stronger setting collapses more candidates to the same output, but only once enough of them collide does any single answer stop standing out.

The blur in this tool is built to collapse candidates on purpose. It quantises the selection to a coarse mosaic first — throwing the fine detail away — and only then smooths that mosaic back up for the soft look. The smoothing runs on data that is already gone, so it cannot bring anything back. The minimum block size is fixed at 12 for the same reason: below that, too much structure survives.

What each mode is actually good for

Use Blackout when being read costs you something. Passwords, API keys and tokens, OTPs, card and account numbers, government IDs, email addresses, precise home addresses, anything in a support ticket that a stranger could act on. These all share the property that makes the averaging attack easy — a rigid format and a small set of possible values — and they are exactly the fields where a near-miss guess is as good as a correct one.

Use Whiteout for scanned paperwork, where a solid black bar reads as a redaction and draws the eye. It is the same operation with a different fill colour and offers the same protection.

Blur and pixelate are the right choice for faces, logos, background clutter, and the contents of other windows. A face has no format to enumerate — there is no list of candidate faces to test against — and you often want the shape and skin tone to stay, so the image still reads as a photo of a person rather than a photo with a hole in it. Same for wallpaper or an unrelated document on the desk behind the subject.

Limits worth knowing before you rely on this

It only touches what you cover. There is no detection here — nothing scans for numbers or faces. If a detail is visible in a corner you did not look at, it ships. The usual misses are the browser tab title, an autocomplete dropdown, a notification banner, and a reflection in a dark UI panel.

It does not clean metadata. The pixels are rewritten; the EXIF block is a separate concern. Use the image metadata viewer and remover if GPS coordinates or a device name are what you need gone, or the bulk metadata remover for a folder of them.

Crop instead, when you can. If the sensitive part is at an edge, cutting it off with the crop tool removes those rows of pixels from the file entirely. That is a stronger guarantee than covering them, and it usually makes the screenshot easier to read too.

Check the export, not the preview. The preview draws selection outlines and handles that are not written into the exported file, and the export is the thing you send. Open it once before it leaves your machine.

Choosing an output format

PNG is the default and the right pick for a screenshot. It is lossless, so a hard-edged blackout bar stays hard-edged and UI text stays crisp. JPG re-compresses in 8 × 8 blocks and puts visible ringing around high-contrast edges — the artefacts collect exactly along the border of a black redaction bar, which looks like a mistake even though the redaction itself is intact. WebP sits between the two and is worth it mainly when you are attaching to something with a size cap.

One quirk: exporting to JPG flattens transparency onto white, because JPEG has no alpha channel. That only matters if your source PNG had transparent regions. If you are converting for size rather than redacting, the image compressor gives you a quality slider with a live size readout.