Can someone recover text from a blurred screenshot?

Last updated: August 6, 2026

Often, yes. In most photo and markup apps, the blur or black box you draw is a separate editable layer sitting on top of the original image — remove the layer, or revert the edit, and the original text is back, unchanged. A tool that flattens the blur into the actual pixels closes that specific hole. It does not, on its own, make blur as safe as a solid fill: pixel-level blurring and pixelation are known to be reversible in principle, and screenshots of text are an easier target than most images because the alphabet is small and predictable.

Why doesn't blurring in most apps actually delete anything?

Because most apps treat it as an edit, not a deletion, and modern photo editors are built to be reversible on purpose. Apple's own Markup tool in Photos is a clear example: when you draw a black rectangle or a highlight over part of a photo, that markup is stored as a non-destructive edit. Anyone with access to the photo can open it in the Photos app, tap Edit, and choose "Revert to Original" — which removes the markup entirely and restores the unredacted photo, in one tap, with no special tools. The same shape of problem shows up in many blur and annotation apps: the blur is a filter applied to a layer, the layer is separate from the base image, and the layer can be hidden, deleted, or never actually merged into the exported file at all.

This is the failure mode that matters most in practice, because it takes zero technical skill to exploit. You don't need to break an encryption scheme or train a model — you need to know that an "undo" or "revert" button exists.

What's different about a destructive flatten?

RedactOffline renders every redaction — solid fill, blur, or pixelate — through Core Image's CIContext.createCGImage, which produces a brand-new, single-plane bitmap. There's no redaction layer sitting above the original in the exported file, because there's no layer at all — the pixels in the redacted region are simply different values now, permanently, in the one image that gets saved. See how it works for the specific APIs involved. The unredacted original exists only in memory for the length of the editing session and is never written into the export.

That closes the "just hit revert" attack completely. It does not automatically make every redaction style equally safe.

If the blur is flattened, is it unrecoverable?

Not necessarily — and this is the honest part most redaction tools skip. Solid fill replaces pixels with a single constant color, which discards the underlying information; there's nothing left to statistically recover. Blur and pixelation are different: they're both linear, well-understood transformations that preserve some structure of what was underneath — low-frequency shape, edges, rough color distribution. A blurred region still correlates with the original content, even though a human eye can no longer read it.

Researchers have demonstrated exactly this. In "Defeating Image Obfuscation with Deep Learning" (McPherson, Shokri, and Shmatikov), neural networks trained on obfuscated images were able to recover faces, objects, and handwritten digits that had been protected with mosaicing (pixelation) and blurring — the same two techniques most redaction tools, including RedactOffline, offer as visual styles. The attack doesn't need the original photo; it needs enough examples of the same kind of content to learn what the obfuscation is hiding.

Screenshot text is, if anything, an easier target than faces. A face has enormous natural variation; a line of text on an iPhone screen does not — it's rendered in one of a small number of system fonts, in a narrow range of sizes, against a mostly uniform background. That's exactly the kind of narrow, structured problem a model can learn well. A blurred or pixelated area of text is not the same risk as a blurred face in a crowd photo — it's closer to a puzzle with very few possible solutions.

So does flattening the blur help at all, or should I just always use solid?

It helps, it just doesn't help as much as solid fill. Flattening removes the zero-effort attack — no one can revert or peel off a baked-in blur the way they can with Markup's black box. What's left is a much harder, model-based attack that requires someone to specifically target your image with the right tools and, in most real-world sharing situations, the motivation to bother. That's a meaningfully smaller attack surface than an editable overlay, but it is not the same guarantee as "the original pixels are gone," which is what solid fill actually delivers.

Recommending against the two paid redaction styles, in favour of the free one, isn't a hedge — it's the accurate technical position. Use blur or pixelate when you want to obscure something that doesn't carry real risk if it's ever partially reconstructed — a busy background, a logo, something you'd rather not show but wouldn't be harmed by. Use solid fill for anything that actually matters: card numbers, ID numbers, addresses, phone numbers, full names tied to sensitive content, MRZ lines on passports or ID cards.

What should I actually do before sharing a screenshot?

1. Let RedactOffline detect the text automatically, or select it manually.
2. For anything sensitive — financial, identity, medical, legal — choose solid fill, not blur or pixelate.
3. Reserve blur and pixelate for content you want to obscure but that isn't harmful if imperfectly hidden.
4. Export and share the redacted copy; the original stays only on your device unless you separately choose to share it.

For the full detection and export pipeline, see how it works. For the tap-by-tap process, see how to redact a screenshot on iPhone.