Can you unpixelate an image?

Not truly. Pixelation discards the detail in each block, so the original can’t be restored. AI tools can generate a convincing guess, but it is invented, not recovered.

No. They produce a plausible face that matches the blocks, which may look nothing like the real person. Treat the result as a guess, never as evidence.

Often, yes. Because fonts are predictable, candidate words can be pixelated the same way and matched. Tools like Depix recovered pixelated passwords this way. Use a solid black bar for text.

No better than a photo, but video leaks more: as the subject moves, the blocks change slightly from frame to frame, which can give away more detail. Use strong pixelation or a solid mask.

Neither is safe for text at low strength. For faces, both work if strong enough. The only option that removes information completely is a solid black bar.

Use an AI upscaler; it will sharpen and invent plausible detail, which is fine for your own photos. Just remember the new detail is generated, not recovered.

For faces and general detail in a photo you are posting publicly, yes — if you go heavy enough that the region is only a handful of blocks across. Two real caveats, though. Researchers have shown that pixelated faces can often be matched back to the right person when the attacker already has a small set of candidate photos — that is re-identification, not reconstruction, but the outcome is the same. And pixelated text is genuinely recoverable: published tools can reconstruct pixelated screenshot text when the font and rendering are known. For names, numbers, addresses and licence plates, use a solid black bar instead.

No. The classic eye bar is a convention, not a protection — jawline, hairline, ears, build, tattoos, clothing and the setting all identify people, and anyone who already knows the person will recognise them instantly. If the goal is genuine anonymity, hide the whole face rather than a strip of it, and think about the badge, the number plate and the house number in the background too.

It depends where you are and why you are publishing, and this is not legal advice — but the direction of travel is clear. Under the GDPR a recognisable face counts as personal data, so publishing an identifiable stranger without a lawful basis can be a violation, including for photos taken in public. Anonymising the image is the usual answer when consent is not practical. Plenty of places outside the EU have narrower rules, and journalism and personal use are often treated differently. If in doubt, blur them — it costs you nothing and it is what most publishers do.

“Pixelate” and “mosaic” are the same effect under two names: the area is divided into a grid and each cell is flattened to one average colour, giving hard-edged blocks. Blur instead smears each pixel into its neighbours, producing a soft smudge with no edges. Pixelation looks deliberate and is easier to judge at a glance, which is why it is the default for censoring faces; blur looks gentler and less like a redaction. Neither is safe for text.