Why images look soft or blocky, and what actually fixes it
Soft images have three different causes, and only two of them can be repaired. Telling them apart is quick.
"Blurry" describes how an image looks, not what went wrong with it. There are three different causes hiding behind that one word, and they do not respond to the same treatment — two of them can be repaired, and one cannot. Working out which one you have takes about as long as reading this paragraph.
The three causes
| What you see at full size | Cause | Repairable |
|---|---|---|
| Square blocks, smeared halos around edges | Compression | Yes |
| Everything soft, but edges land where they should | Not enough resolution | Yes |
| Edges doubled, streaked, or soft in one direction | Focus or motion blur | No |
The test is to open the image at 100% — not fit-to-window — and look at a hard edge. Fit-to-window hides all three problems equally well, which is why images so often look fine on screen and fall apart when printed or zoomed.
Cause 1: not enough resolution
This is the most common case and the most fixable. The image is small relative to how it is being used: a 600-pixel-wide product shot on a page that displays it at 1200, an old avatar being used as a header, a crop from a larger frame.
Nothing is damaged here. There is simply not enough information, and the display is stretching what exists across more pixels than it was built for.
Ordinary resizing cannot fix this, because it has nothing to work with — it averages neighbouring pixels and produces a larger, softer version of the same image. A model trained on how detail behaves can reconstruct plausible edges instead of averaging them, which is what upscaling actually means in this context.
Cause 2: compression damage
JPEG discards information to save space, and it does so in 8-pixel blocks. When it is pushed too far, or when an image has been saved repeatedly, those blocks become visible: flat squares in gradients, and a shimmer of noise clinging to hard edges.
Screenshots pasted into chat apps, images saved from social platforms, and photos that have been through several rounds of editing all tend to arrive in this state.
This is repairable, but the order matters. Enlarging a compressed image first will enlarge the compression artefacts along with everything else, making them more obvious rather than less. Noise reduction has to happen as part of the same pass — which is why the processing here offers a noise-reduction setting alongside the scale factor rather than as a separate step afterwards.
Cause 3: focus and motion blur
This is the one that cannot be repaired, and it is worth being direct about it.
If the camera was focused on the wrong plane, or if the subject moved during the exposure, the detail was never recorded. There is nothing in the file to recover. A model can produce something sharp-looking by inventing texture that was not there, but that is a different image, not a repaired one — and on a face, it can come back as a subtly different person.
You can recognise this case by direction: motion blur streaks one way, and focus blur softens everything at one distance while leaving other distances sharp. If part of the frame is crisp and part is not, no amount of processing will bring the soft part back.
A fourth case: it was never blurry
Occasionally an image that looks soft is fine, and the display is the problem.
A screenshot taken on a high-density display and viewed at half size looks soft because it is being downscaled, not because it lost anything. An image shown in a container narrower than the file is doing the same thing in reverse. And a photograph viewed in an application that applies its own preview smoothing can look softer than the file actually is.
Before processing anything, open the file at 100% in something that does not resample it. If it is sharp there, the file is fine and the fix belongs in the page or the export size, not in an upscaler.
Fixing the two that can be fixed
- Open Pixmend and drop the image on the page.
- Choose x2 or x4. Pick x4 only if you actually need the larger size; a smaller factor keeps the result more conservative.
- Choose the photo or illustration model to match your image.
- If you saw blocks or edge noise, raise the noise-reduction setting.
- Compare before and after with the slider at full size, then download.
Processing happens inside your browser, so nothing is uploaded, and there is no limit on how many times you can try a different combination.
When it does not work
If the result still looks wrong, the cause is usually one of these:
It was focus or motion blur after all. Check whether any part of the frame is sharp. If some of it is, the soft part is unrecoverable.
The scale factor was too aggressive. x4 on an already small image asks the model to invent a lot. Try x2 and see whether it holds up better.
The wrong model. Line art through the photo model tends to soften the linework; photographs through the illustration model tend to flatten skin texture. Both are free to run, so trying each takes a minute.
The source is a screenshot of a screenshot. Each round of saving and re-encoding removes information permanently. At some point the only real fix is to find the original file.
That last one is worth taking seriously as a habit: keep the largest original you have, and do the enlarging at the end of the process rather than in the middle. Every save between the camera and the final image is a small, permanent loss.