Soft is not the same as blurred
A soft photo has the detail present but low in contrast. The edges are there in the data; they are just not distinct enough for your eye to register them as crisp.
A blurred photo has had its detail averaged across neighbouring pixels. The information is not dim, it is gone — spread out and combined with its surroundings in a way that cannot be separated back out.
Both look unsatisfying on screen, which is why people reasonably assume one tool fixes both. Sharpening addresses the first and does nothing useful for the second.
What sharpening actually does
An unsharp mask finds places where brightness changes quickly, then exaggerates the change — darkening the dark side of the transition and brightening the light side. Local contrast goes up and your visual system reads that as crispness.
Nothing is added. Existing structure is emphasised, which is why it works so well on a photo that is nearly right and not at all on one that is genuinely blurred. If an edge was never recorded, there is no contrast to exaggerate.
The name is a historical oddity from darkroom practice, where a blurred negative was used as a mask to increase apparent sharpness. The digital version does the same thing arithmetically.
What responds well
These all have detail present but low in contrast, which is exactly what sharpening is for.
- Slightly missed focus, where the plane of sharpness fell just in front of or behind the subject
- Diffraction softening from shooting at a very small aperture
- Shooting through glass or a slightly dirty lens
- Softness introduced by an aggressive downscale
- The general flatness of an image straight out of a camera that applies no in-body sharpening
What does not respond
Motion blur. The sensor integrated light across a path while the shutter was open, so each output pixel is an average of many scene points. Averaging is not reversible without knowing exactly what the path was.
Significant defocus. Beyond a small amount, each point in the scene has been spread into a disc across many pixels and the discs overlap. Reconstructing the original requires deconvolution with a precisely known blur kernel and, even in laboratory conditions, amplifies noise dramatically.
Compression softness. A heavily compressed thumbnail lost its detail to the encoder rather than the lens. Sharpening it emphasises the compression artefacts instead, producing crisp blocks.
Upscaled images. Interpolated pixels are guesses. Sharpening a guess produces a confident-looking guess.
Why the films are lying
The enhance trope — zoom into a reflection, recover a licence plate — is not how any of this works. The pixels do not contain more information than they contain.
What modern systems can do is different and worth understanding. A generative model trained on millions of faces can produce a plausible sharp face consistent with a blurry input. It looks impressive because it is producing a real face; it is not producing your face.
For anything evidentiary, that distinction is decisive. An invented plate number that looks convincing is worse than no plate number, because it is confidently wrong.
Halos are the signal to stop
Push sharpening too far and bright outlines appear along high-contrast boundaries — a white fringe against a dark roofline, a glow around text.
That is the algorithm exaggerating an edge past plausibility, and once visible, the image reads as processed rather than sharp. It is a specific look that people recognise as over-sharpened even if they cannot name it.
Sharpening also amplifies noise, because noise is high-frequency detail and the filter cannot distinguish it from real texture. On a grainy photo, denoise first and sharpen second, or you end up with very crisp grain.
How to judge the result
Look at full size, not at a thumbnail. Sharpening artefacts are invisible when scaled down and obvious at 100 percent, and it is easy to over-apply while judging a small preview.
Then look at the size it will actually be displayed. A web image in a 600-pixel column tolerates more aggressive sharpening than a print, because nobody will see it large.
If you cannot find a setting that helps without introducing halos, the source is blurred rather than soft. That is a shooting problem and it is worth accepting quickly rather than spending an hour on it.
Avoiding it next time
Most phone blur is motion, not focus, and most of it comes from tapping the shutter. Use the volume button or a timer, and brace against something solid.
In low light the camera lengthens the exposure, which is when hand shake becomes visible. More light is the actual fix; a surface to rest on is the practical one.
Tap to focus on the subject rather than trusting the camera to choose. Autofocus picks the wrong plane constantly, especially when something is in the foreground.
Take three frames instead of one. Storage is free and one of them is usually noticeably sharper than the others.
Frequently asked questions
Can you really unblur a photo?
You can sharpen a soft photo by boosting edge contrast. You cannot reverse motion blur or significant defocus, because that information was averaged away rather than dimmed.
What is the difference between soft and blurred?
Soft means the detail is present but low in contrast, which sharpening fixes. Blurred means it was averaged across pixels and is gone.
Why does my sharpened photo look worse?
Either the strength is too high, producing bright halos along edges, or the image is noisy and you have sharpened the noise along with the detail.
Should I denoise before or after sharpening?
Before. Sharpening cannot tell noise from texture, so sharpening first gives you very crisp grain.
Can AI recover a blurry licence plate?
A generative model can produce a plausible sharp plate, but it is invented rather than recovered. For anything evidentiary that is worse than nothing.
Why are my phone photos always slightly blurry?
Usually motion from tapping the shutter, often combined with a long exposure in low light. Use the volume button or a timer and brace against something.