Images tower · floor

How to fix pixelated photos: three causes, one hard limit

The same appearance has more than one cause, and only some of them are in the file. Sorting out which one you have is the whole job, because two of the three are fixable and the third is not.

That four-way split is a useful piece of thinking to borrow, because “this photo is pixelated” is a statement about how something looks, and the look can come from any of those places. A picture that is perfectly sound in its file can be pixelated on a screen simply by being shown larger than it is.

The first case is the cheerful one, and it is worth ruling out before touching anything. If the image looks fine at a smaller size on screen, the file is fine and the problem is presentation: a container too wide, a zoom too far in, a slideshow filling a display the picture was never sized for. Nothing needs editing.

The second and third look identical and are not. Both give you a file without enough detail — but in the second, a better version existed at some point. That makes the first move the same in both cases and free to try: go looking for the original. The copy in a message thread, the version still on the camera, the upload before the platform resized it.

If there is no original, you arrive at the question people are really asking, and GIMP answers it in its own manual, in one sentence.

It is worth sitting with what that rules out, because it is a lot. It rules out fixing a pixelated photograph by making it bigger. It rules out the idea that a better algorithm would have worked — the manual says no matter which one. And it explains the disappointment of the obvious attempt: enlarging does something, the something is blur, and blur is not detail.

What it does not rule out is the opposite direction. Reducing an image is documented as having a cost too — GIMP says quality is lost when pixels are dropped — but the result of showing a small image at its own size is a sharp small image. If the picture only needs to be small, that is a complete fix — and it is a real one, even though it works by lowering the requirement rather than by improving the file.

A related point on the same page is worth knowing before you enlarge anything for a practical reason: GIMP warns that scaling changes “the number of pixels (the amount of information) the image contains, so it directly affects the amount of memory the image needs (in RAM or in a file)”. Past a size set in the preferences — the default is 128 MiB — “you are warned and asked to confirm the operation”. So an enlargement that fixes nothing can still cost you a much larger file.

Where this page’s evidence stops. Everything quoted above is about interpolation — arithmetic on neighbouring pixels. A tool that produces detail which was not in the file is doing something other than interpolation: it is generating it. That is a different operation, and GIMP’s sentence is not a verdict on it. Nor is this page: the site does not publish judgements on tools it has not run against real work for long enough to have one. What can be said is narrower and still useful — anything that adds detail is adding invention, and whether invented detail is acceptable depends entirely on what the picture is for. On a decorative background, fine. On a photograph of a person, a document or a number plate, that is a different question and worth asking deliberately.

The practical order that falls out of all this is short. Check whether it is only the display. Then look for an original. Then decide whether the picture can simply be used smaller. Only after those three is it worth reaching for anything that changes the pixels themselves — and by then you will know what you are trading.

The floors below take it three ways: what pixels are and where they live, the tools that change them, and the neighbouring questions about images you did not take yourself.

What this tower will not do

It will not tell you enlarging fixes a pixelated photograph. GIMP’s manual says no interpolation method adds information, and no page can argue with the program’s own documentation about what the program does.

It will not name a tool that recovers lost detail. Anything that appears to is generating detail rather than recovering it, and that is a different claim than the one being searched for.

And it will not judge the tools that make that claim. Testing them properly means running them against work that matters, which this site has not done, and an untested opinion is worth less than saying so. What holds instead is simple: the four places an image can be located, the statement that scaling changes the number of pixels and the memory required, the 128 MiB preference at which a confirmation is requested, the note that printing size depends on resolution and can be changed without scaling, the observation that screen size depends on resolution and zoom as well as pixel count, and the statements that enlarging adds no new detail, that no interpolation algorithm adds new information, and that a scaled layer is likely to look blurred are quoted from the official GIMP documentation, listed below.