Can blurred or pixelated text be recovered?

Sometimes, yes. Blur and pixelation are transformations of the original pixels, not deletions, and a transformation can be undone when the attacker knows enough about the input. That is why every serious redaction guideline, from courts to security teams, says the same thing: cover, don't smudge.

Pixelation

Pixelation replaces each block of pixels with their average. For text in a known font and size, an attacker can render candidate characters, pixelate them the same way, and compare against your image block by block. Short, high-value strings such as passwords, account numbers and license plates are the easiest targets because the search space is small. Open-source tools have demonstrated this against pixelated text since 2020, and the technique gets stronger every year as the models improve.

Blur

Gaussian blur spreads each pixel's value over its neighbors. Mathematically it's a convolution, and convolutions can be partly inverted (deconvolution) when the blur radius is guessable, which it usually is, because most tools use a handful of presets. Real-world recovery is less reliable than with pixelation, and heavy blur on long text is often unrecoverable in practice. But "often" is a bad word to build a privacy process on, and a blurred number is also a signal that there is a number there worth attacking.

What is actually safe

• Solid boxes. A filled rectangle contains zero information about what was underneath. Black is conventional; any opaque color works.
• Removing the text before capture. If the text never made it into the image, there is nothing to recover. This is what Cloakshot does: it covers the text on the web page first, then captures.
• Flattening. If you redact in a layered editor (Photoshop, Figma, PDF annotation), make sure the export flattens the layers. A box on a separate layer can simply be moved aside in the original file. This has caused real document leaks in courts and government releases.

Three mistakes that survive even a solid box

1. Metadata. Screenshots usually carry little, but exported PDFs and edited images can keep the original in an embedded thumbnail or undo history. Export as a fresh PNG.
2. The unredacted copy. The original is still in your clipboard manager, screenshot folder, or chat "recently sent" list. Redacting before capture avoids this by never creating it.
3. Context. A box over the account number doesn't help if the next line says "Account for Maria Oliveira, São Paulo". Redact identifiers together.

So why do tools still offer blur?

Because it looks nicer, and because for low-stakes content (a stranger's face in a crowd, a random background window) it's fine. Cloakshot's editor offers blur and pixelate for exactly that, and uses solid black by default for everything it detects, because the things it detects are precisely the things worth attacking.