Give One Exposure the HDR Look

How it works

Four operations and no model at all.

Your image is resampled to about two megapixels. Then an unsharp mask runs at a deliberately large radius — twenty five pixels, with a wide blur underneath it — which is the operation that produces the HDR look. A small global contrast lift follows, and finally a conventional one pixel sharpening pass restores the edge crispness that the large radius pass does not address.

Because there is no sampler in the chain, the output is a deterministic function of the input. Run it twice and you get the same file. Nothing is generated, so nothing in your photograph can be replaced by something that was not there.

Large radius is the whole trick

An unsharp mask increases the difference between a pixel and its surroundings. The word surroundings is doing all the work.

At a one or two pixel radius, the surroundings are the pixels immediately adjacent, so the operation increases contrast right at edges. That is ordinary sharpening and it makes an image look crisper without changing its overall structure.

At a twenty five pixel radius, the surroundings are a whole region. Now the operation increases contrast between areas rather than at edges: a building against the sky behind it, a cloud against the clear patch beside it, a shadowed wall against the lit street. The picture keeps its overall brightness, but every part of it separates from its neighbors.

That separation is what people recognize as HDR. It is not extra dynamic range — the file has exactly the range it always had. It is the existing range being redistributed so that more of it is spent on differences you can see.

What clipping means

The most common disappointment with this page has a hard physical explanation.

A sensor pixel counts photons up to a limit. Past that limit it reports the maximum and stops, and a bright sky routinely passes that limit in all three channels. Every pixel across the whole region then holds the same value.

Enhancing contrast means amplifying differences. Where there are no differences, amplification returns nothing, and a flat white sky remains a flat white sky no matter what is applied to it. This is not a limitation of the workflow, it is what the word clipped means.

You can see it before you upload. Open the picture and look at the sky: if there is any tone variation at all, this will find it and pull it out. If it is a uniform white shape, that information was never recorded, and the only fix is bracketing the shot next time.

Where it flatters and where it does not

Landscapes, architecture, interiors, machinery, and weather all improve, because their interest is in surface and structure and this pass emphasizes exactly that.

Portraits generally do not. Skin is fine-scale texture, and a process that lifts fine-scale texture lifts pores, lines, and unevenness with everything else. The result reads as harsh even when it is technically correct.

Use it accordingly, and if a landscape contains a person in the foreground, check their face before you publish.

Publication gate

This page ships once the workflow has been run against a flat overcast landscape, a backlit frame with a clipped sky, an interior with deep shadow, and a close portrait, confirming the halo along a high contrast skyline is within a tolerable width in each case.