How to unpixelate an image

A mosaic of small amber and grey tiles resolving into a smooth curved surface across a dark background

To unpixelate an image, upscale it with AI instead of sharpening it. Sharpening only hardens the blocks that are already there. An upscaler generates new pixels at up to 4x and reconstructs the shapes those blocks were standing in for.

Pixelation is a different problem from blur

A blurry photo contains the detail, badly distributed. A pixelated photo does not contain it at all: the file was saved at a size too small to hold it, and the encoder averaged whole regions into single squares. That difference decides which tool works.

Sharpening filters assume there is an edge in there somewhere to find. Applied to pixelation, they find the boundaries between blocks and make those crisper, which is why a sharpened pixelated image looks like a stained-glass window.

Upscaling generates what compression removed

An AI upscaler works differently. It has seen enough images to know what a human eye, a brick wall and a car headlight look like at high resolution, so given a low-resolution version it produces a plausible high-resolution one. UnblurApp does this at up to 4x, so a 500-pixel image comes back at 2000.

The word doing the work in that sentence is plausible. The tool is not recovering the original pixels. It is proposing detail consistent with what it sees, which is exactly what you want for a family photo and exactly what you must not rely on for anything forensic.

UnblurApp tool list showing Colorize, Restore Faces and Repair cards, each with a before and after thumbnail — The Repair tool increases photo resolution up to 4x instead of sharpening what is already there.

Where pixelated images usually come from

Messaging apps are the biggest source. A photo sent through several chats gets recompressed at each hop, and each pass discards a little more. By the fourth forward, a picture that started at 12 megapixels is a few hundred kilobytes of approximation.

The others are old web downloads sized for a 2006 monitor, screenshots taken on a small display and then enlarged, and profile pictures pulled from a social network at thumbnail size. All of them respond to upscaling because the subject is still recognisable, even if the detail has gone.

The cases nothing will fix

Deliberate pixelation, the kind applied to blur a licence plate or a face in a news photo, is designed to destroy information. Every video-drama enhance scene you have watched is fiction. If a plate has been reduced to eight grey squares, the number is gone, and any tool producing a legible result has invented one.

The same applies to heavily censored regions in leaked documents. A model can generate text-shaped marks there, and they will be wrong.

Get the best input you can find first

Before you upscale, spend two minutes looking for a better copy. The version in the original chat thread is usually less compressed than the one someone forwarded to you. A photo saved from a website may exist at full size behind the thumbnail. An email attachment often kept the original where the messaging app did not.

Ten seconds of searching beats any amount of processing, because a better source gives the model more to work with.

Questions people also ask

Can you unpixelate an image for free?

UnblurApp is free to download and lets you try the upscaler. Free web tools also exist, though most cap the output resolution, which defeats the point of upscaling.

How much can you upscale before it looks wrong?

Up to about 4x holds together well on most photos. Beyond that the model is guessing more than it is reconstructing, and faces in particular start to look synthetic.