There are four kinds of blur and only some of them can be improved. Identifying which one you have takes thirty seconds and saves an hour of pointless sharpening.
Whether a blurry photo can be fixed depends on the type of blur. Softness from diffraction or a soft lens, and mild camera shake in one direction, can be improved by denoising and then applying deconvolution sharpening. Badly missed focus and strong motion blur cannot be recovered, because the detail was never recorded; AI deblur will make those frames look sharp by inventing detail, which on a face can change who the person appears to be.
"Can you fix this, it's a bit blurry?" is one of the most common requests a retoucher gets, and the answer depends on which of four different things happened. They look alike in a thumbnail and are completely different problems at 100%.
Identify the blur first
Open the file at 100% and look at a high-contrast edge.
Missed focus. Something in the frame is sharp, just not the thing you wanted. The blur is even in every direction and increases with distance from the plane of focus. Partly recoverable when the miss is small.
Camera shake. The whole frame is blurred in one consistent direction. Point lights become short streaks, all pointing the same way. Partly recoverable, and the streaks tell you the direction and length.
Subject motion. The background is sharp and the subject, or part of it, is not. Barely recoverable, but often not a problem: blur on a moving subject frequently reads as intentional.
Diffraction or a soft lens. Everything is slightly soft, with no direction, usually from a very small aperture or a lens used wide open. Recoverable to a degree, because the detail is present but low in contrast.
What can be improved
Low micro-contrast. Diffraction and lens softness leave detail in the file but flatten it. Deconvolution sharpening, which models the blur and attempts to reverse it rather than simply raising edge contrast, produces a visible improvement.
Mild, uniform camera shake. If the streak is short and consistent, a deblur tool that is given its direction and length can recover a meaningful amount.
Softness from heavy noise reduction. If an export was over-denoised, the detail may still be in the raw. Reprocess from the original with gentler settings instead of sharpening the damaged JPEG.
What cannot be recovered
Badly missed focus. If the eyes were well behind the plane of focus, information about the eyes was never recorded. No tool can retrieve it. AI tools will produce something, and that something is a plausible pair of eyes rather than your subject's.
Severe motion blur. For the same reason.
Blur in a heavily compressed file. Compression has already discarded the fine detail a reconstruction would need.
The order that gets the most from a soft file
- Start from the raw if there is one. A JPEG has already been sharpened and compressed, and both work against you.
- Denoise before sharpening. Sharpening noise makes more noise. This order is not optional.
- Use deconvolution rather than unsharp mask where your software offers it.
- Sharpen selectively. Mask to the areas that matter — eyes, edges, fabric texture — and leave smooth areas alone. Global sharpening improves eyes and worsens skin at the same time.
- Stop earlier than feels right. Over-sharpening looks fine while you are doing it and obvious an hour later.
- Check at 100% and at final output size. Halos invisible at 50% are unmissable in print.
What AI deblur adds
Machine-learning sharpening and deblur tools are clearly better than the sliders that came before, particularly on mild shake and lens softness.
They also invent. On a badly out-of-focus face, AI deblur produces a sharp face, and it may be a face that is confidently, subtly not the person. On a portrait that is worse than delivering the soft version, because the soft version is at least a photograph of the right person.
Test a tool on a file where you know the answer: take a sharp frame, blur it deliberately, and try to recover it. You will see what the tool reconstructs and what it makes up.
When to accept softness
Some frames are soft and still the right picture. A slightly soft candid with the right expression beats a technically perfect frame with the wrong one, and clients choose the expression more often than photographers expect.
Two things help: deliver the frame smaller, because a soft image at 1200px can be a perfectly good image at 1200px, and consider black and white, which forgives softness more readily than colour.
Preventing it next time
A faster shutter speed, more light, single-point focus on the nearer eye, back-button focus, and burst mode on anything moving. None of it helps today's file, but it means fewer of these conversations.
Sending a soft frame to a retoucher
If the soft frame is part of a set you are outsourcing, a few lines in the brief save a round of revisions.
Send the raw file rather than an exported JPEG, and say which part of the frame has to read as sharp, usually the nearer eye.
Say whether any reconstruction is acceptable. Some clients would rather have a slightly soft photograph of the real person than a crisp one with invented detail, and that decision belongs to you and your client, not the retoucher.
Everything else the frame needs is ordinary finishing — skin cleanup, exposure and contrast, colour correction — which is what Standard Retouching covers at $4 per image. For formats and file transfer, see How to Send Photos Safely to a Professional Retoucher.
Common questions
Can AI fix an out-of-focus photograph? It can produce a sharp-looking photograph. Whether it is still your photograph depends on how far out of focus the frame was: when focus was badly missed, the detail was never recorded and what comes back is a reconstruction.
What is the difference between deconvolution and unsharp mask? Deconvolution models the blur and tries to reverse it; unsharp mask raises contrast along edges. On true softness, deconvolution recovers real detail while unsharp mask gives an impression of it.
Should I sharpen before or after noise reduction? After. Sharpening noise creates more noise.
Why does AI deblur look convincing on landscapes but not on portraits? Invented detail is hard to spot in foliage, rock or water and easy to spot on a face. Check faces, text and anything recognisable at 100%, however good the rest of the frame looks.
How do I tell which kind of blur I have? Look at a high-contrast edge at 100%. Blur even in every direction is focus; one consistent direction is shake; a sharp background with a soft subject is motion; slight softness everywhere with no direction is the lens or diffraction.
Related reading: How to Upscale a Photo Without Making It Look Fake and RAW vs JPEG for Professional Retouching.
BeEdits Editorial Team
Post-production studio
A photography post-production company: high-end retouching, creative editing and scalable production support for photographers, studios and brands.