Blur or redact a screenshot so the hidden part stays hidden

Most advice about hiding information in a screenshot stops at "drag the blur tool over it". That is the part that looks right rather than the part that is right. Blurring and redacting make two different promises, only one of them survives someone who actually tries, and the way your editor writes the change into the file decides whether either promise holds at all.

Blurring and redacting are two different promises

Redacting says: this area is gone, replaced by something that carries none of the original. A solid block of colour has no relationship to what was underneath it. There is nothing left to reconstruct because the pixels that held the information no longer exist in the file.

Blurring and pixelating say something weaker: this area has been degraded so that a human reading at normal size cannot make it out. That is a claim about legibility, not about the information being absent. The pixels are still there, they have just been averaged together, and averaging is a lossy transformation rather than a deletion.

For a face in the background of a screenshot, or a colleague's avatar, degrading legibility is usually the honest amount of protection to want. For a password, a bank account number, a home address, an API key or a customer name, it is not. Those need the stronger promise.

The mistake that hands the information straight back

The single most common failure has nothing to do with blur strength. It is covering something with a marker, highlighter or brush that is not fully opaque.

Highlighter tools in general-purpose image editors are usually semi-transparent by design, because a highlighter that hides the text underneath would be useless as a highlighter. Drag one over an account number and it looks convincingly covered on your screen. Anyone who receives that file can open it, push the brightness and contrast sliders, and read the number straight through the ink. No specialist software and no skill required.

The same problem appears with a dark rectangle drawn at 80 or 90 percent opacity, which is a common default. Ninety percent hidden is zero percent hidden once someone stretches the remaining ten percent across the full contrast range.

If you take one thing from this page: whatever you cover it with must be fully opaque, and it must be flattened into the image rather than saved as a separate layer. A PDF or a layered file that still holds the original underneath is not redacted, it is decorated.

Why pixelation is weaker than it looks

Pixelation replaces each block of pixels with the average colour of that block. How much survives depends entirely on the block size relative to what you are hiding.

Text is the difficult case. Characters are drawn from a small, known alphabet in a small, known set of fonts. If the blocks are small compared to the letters, each block keeps enough signal that the original characters can be narrowed down by trying candidate strings, pixelating each one the same way, and comparing the result to what is in the image.

This is not theoretical. Depix, released in 2020, does exactly that search, and Unredacter, published by security researcher Dan Petro at Bishop Fox in 2022, improved on it and successfully recovered text from a pixelated sample that had been posted as a public challenge.

It is worth being accurate about how strong the attack is, because overstating it is its own kind of misinformation. These tools need favourable conditions. Depix in particular assumes the pixelation was produced by a specific kind of averaging filter, and independent attempts to use it on arbitrary real-world images have often failed. Recovering pixelated text is not a button anyone can press on any image.

But the conclusion the researchers themselves reached is the practical one, and it does not depend on how reliable the tooling is today: when the thing being hidden is text, cover it with a solid bar. Not pixelation, not blur, not a swirl. The reason is simply that a solid bar has no failure mode, and every degradation-based approach has one that improves over time as the tools do.

You do not have to judge that in the moment, and you should not have to. The rule that always holds: if the thing you are hiding is text and it matters, use a solid redaction rather than pixelation. Pixelation is the right tool for faces, logos, background windows and anything where you want the shape to remain visible so the screenshot still makes sense.

How Screenshot Files applies it

The editor has both tools, and both write their result directly into the image rather than sitting on top of it as an annotation you could later remove.

Redact fills the area you drag with a solid, fully opaque colour. There is no transparency to recover and no layer to peel off. Once applied, the pixels that were there have been overwritten in the image data itself.

Pixelate reads the area, averages it in blocks, and writes the averaged values back over the original pixel data. The block size scales with the size of the area you selected, with a floor so that a small selection still gets meaningful blocks rather than a cosmetic softening.

Because both are destructive at the moment you apply them, the copy you export, copy to your clipboard or share as a link carries no trace of the original in the covered area. This is also why the editor asks you to be deliberate: you cannot recover a redacted area later either. Keep the original if you might need it.

All of this runs in your browser on the image you pasted. A screenshot you are redacting because it contains something sensitive does not have to be uploaded to a service in order to be redacted, which would be an odd order to do things in.

Check it before you send it

Two habits catch almost every mistake that survives the steps above.

First, look at the whole screenshot, not just the field you were thinking about. Sensitive information leaks through browser tabs, window titles, notification banners, autocomplete dropdowns, breadcrumb trails, a visible email address in an account menu, and the reflection of a filename in a recent-files list. Crop aggressively before you redact. The information you never captured cannot be recovered from anything.

Second, after you export, open the exported file and try to defeat your own redaction. Increase the brightness, increase the contrast, zoom in. It takes ten seconds. If anything shows through, you have learned that before the recipient did.

Common questions

Is blurring ever good enough?

For anything where you want the shape to stay readable as context but the detail should not be, yes. Faces, background windows, a colleague's name in a sidebar. For credentials, account numbers, addresses and anything a person could act on if they recovered it, use a solid redaction instead.

Can a redacted area be recovered from the exported file?

Not when the redaction was written into the image data before the file was created, which is what happens here. The risk you read about with redacted documents comes from formats that keep the original content and draw a black box over it as a separate object. A flattened image has no such second copy.

Does the metadata still hold anything?

Screenshots carry far less metadata than photographs, and cloud copies here are converted to WebP during processing, which does not preserve the original file's metadata block. The thing worth worrying about in a screenshot is almost always what is visible in the pixels rather than what is recorded around them.

What if I redacted the wrong area?

Undo while you are still in the editor. Once you have saved the edited version, the covered pixels are gone from that copy, which is the entire point. If the original still matters, keep it as a separate screenshot before you start editing.