AI Image Sharpen that stops before the halos.

Soft is not one problem

Three different things make a photograph look soft, and only one of them can be improved after the fact.

Missed focus means the lens resolved a plane that the subject was not standing on. If the miss is small, the subject still carries a faint record of its edges and there is something to recover. If the miss is large, the detail was never written to the sensor.

Motion blur means the subject or the camera moved while the shutter was open, so each point of the subject was recorded as a streak. The direction of the streak is often visible in the image. Long streaks have genuinely destroyed the information, and no amount of processing brings it back honestly.

Softness from processing is the recoverable one. Compression, aggressive noise reduction, resizing, and repeated exporting all reduce local contrast, and the underlying structure usually survives underneath. That is the case where sharpening does real work rather than cosmetic work.

The halo problem

Conventional sharpening finds edges and increases the contrast across them, making the light side lighter and the dark side darker. Applied gently that reads as clarity. Applied hard it produces a bright outline tracing every edge in the picture, and once you have noticed halos in an image you cannot stop seeing them.

The reason it is so common is that halos look like sharpness on a small preview. The effect that reads as crisp in a thumbnail is usually well past the point where it reads as artificial at full size, so an image edited at one scale and viewed at another goes wrong in a predictable direction.

This one is tuned to recover local contrast and stop before that outline appears. The practical result is a smaller apparent change than a slider pushed to maximum would give, and one that survives being looked at closely.

How to judge the result

View at full size, not fitted to the window. A downscaled preview hides both the improvement and the artifacts.

Check high-contrast boundaries first, because that is where halos appear: a dark branch against a bright sky, a shoulder against a pale wall, the edge of a product against its backdrop. Then check areas of fine texture such as hair or fabric, where over-sharpening turns smooth gradation into a gritty, granular surface.

If the picture has gained bite but also gained visible speckle, the sharpening has amplified noise that was already there. Run Denoise on the original and sharpen the cleaned version instead of sharpening the noisy one.

Product shots and portraits want different amounts

The right amount of sharpening depends on what the picture is of, and the two common cases pull in opposite directions.

Product photography rewards a firm hand. A buyer is trying to read a material, a seam, a finish, or printed text on a label, and crisp edges help them do that. There is no penalty for a slightly clinical look on a listing image, because nobody minds a bolt or a fabric weave being rendered too precisely.

Portraits punish the same setting. Skin is made of gradual transitions, and raising local contrast across it brings out pores, fine lines, and any unevenness that the original light was hiding. The result reads as harsh in a way that is hard to name and easy to notice. A portrait generally wants noticeably less than a product shot, applied so that eyes and hair carry the impression of sharpness while skin is left alone.

If a portrait has come back looking rougher than the original, that is the signal to start again from the unsharpened file rather than to correct it further.

When another tool fits better

Sharpening a file that is too small only makes the existing pixels more prominent. If the image is under-resolved for where it needs to go, Upscale first and judge sharpness afterwards, since the enlargement reconstructs detail that sharpening can then bring forward.

If the softness comes with grain, denoise first for the reason above. If the photograph is an old print that is soft because it has faded and lost contrast overall, that is age rather than focus, and Restore addresses the cause instead of the symptom.

Files and output

JPG, PNG, and WEBP are all accepted, and the result comes back as a PNG. Sharpening increases fine local contrast, which is exactly the kind of detail a JPG encoder discards first, so returning a compressed format would undo part of the work on the way out.

Sharpen the version closest to the original. A file that has been exported and re-saved several times has lost local contrast at each step, and recovering it once at the end is weaker than starting from the copy that never lost it.