There is no correct histogram

What the graph is, precisely

The histogram counts pixels by value. The horizontal axis runs from 0 to 255; the height at each position is how many pixels hold that value. That is all. It has no idea where in the frame those pixels are, what they are pictures of, or whether the photograph is any good.

The first thing to unlearn is the idea of a correct shape. A photograph of a black cat in a coal cellar should be piled on the left. A high-key portrait on white should be piled on the right. A foggy morning is a narrow lump in the middle and that is what fog looks like. An editor who drags every image towards a even spread is destroying the subject.

What the histogram tells you is whether data has been lost at the ends, and whether tones are missing in between.

RGB, colour, and luminosity are three different graphs

Most editors offer several histogram views and the difference matters.

• Luminosity: one graph, weighted 0.2126 / 0.7152 / 0.0722, showing perceived brightness. The right one for judging exposure.
• Colours or RGB overlay: the three channels drawn on top of each other. The right one for spotting a cast and, crucially, for spotting per-channel clipping.
• RGB in Photoshop’s menu is a combined graph that is not the same as luminosity and is the least informative of the three.

Per-channel clipping is the one that catches people. A photograph of a deep red flower can look perfectly exposed on the luminosity histogram while the red channel is jammed against 255 and every petal has lost its internal detail. Nothing on the luminosity graph reveals that. Switch to the three-channel view and it is unmissable.

This is why heavy saturation is dangerous: raising saturation pushes channels apart, and the leading one hits the wall first. Check the colour histogram after any saturation move.

The clipping warnings, and what they mean

Most editors will show clipped pixels in a flashing colour. In darktable and RawTherapee they are on by default; in Lightroom the triangles at the corners of the histogram toggle them; in Photoshop, hold Alt/Option while dragging a Levels endpoint and the preview shows only what you are about to destroy.

Read them properly:

• Specular highlights (the sun’s reflection on chrome, a bare bulb) are supposed to be pure white. There is no detail in them and never was. A clipping warning there is not a problem.
• A blown sky, a blown white shirt, blown skin are real losses. Once all three channels are at 255 there is nothing to recover.
• Clipped shadows are often fine. A few per cent of pure black in a night scene gives the picture somewhere to rest. A clipped shadow across half the frame has eaten the subject.

The histogram your camera showed you is a lie

A fact that changes how people shoot. The histogram on the back of a camera, and the blinking highlight warning, are computed from the embedded JPEG preview, not from the raw data. That preview has the camera’s contrast curve, saturation and white balance applied.

So the camera can warn you that the sky is blown when the raw file still holds a stop of it. In practice the raw file typically has between half a stop and a stop and a third more highlight headroom than the camera’s display suggests, depending on the model and the picture style.

The practical consequence is the technique called expose to the right: deliberately push exposure up until the camera warns, because the raw file will still hold it, and a brighter raw file carries more signal relative to noise in the shadows. The risk is obvious and the discipline is to test your own camera once (bracket a scene, develop the raws, find where the recovery actually stops) rather than trusting a general rule.

Setting a flat, low-contrast picture style in the camera makes the preview histogram closer to the raw and costs nothing, because the raw is unaffected by it.

The waveform, and why video people use it

A histogram discards position. A waveform keeps the horizontal position: it plots brightness vertically against the frame’s horizontal axis, so you can see that the left side of the shot is clipping while the right side is fine.

For a photograph with a bright window on one side and a dark interior on the other, this is far more useful than a histogram. darktable has a waveform and an RGB parade in its histogram module: click through the modes. Krita has one. Photoshop does not, which is a genuine gap, and the workaround is to use the eyedropper with the Info panel on specific points.

Numbers beat the graph for specific questions

The eyedropper with the Info panel open gives you exact values at a point, and for several questions that is what you want:

• Is this white shirt actually white? 240 to 250 in all three channels means printed detail survives. 255 means the texture is gone.
• Is this grey wall neutral? Equal values across R, G and B. A difference of five is a visible cast.
• Are these two product shots matched? Sample the same point on both.

Set the eyedropper’s sample size to 3 by 3 average rather than Point Sample. A single pixel in a noisy image tells you about the noise rather than the tone.

What the histogram cannot tell you

It cannot tell you where the tones are, which is why two completely different photographs can have identical histograms. It cannot tell you whether the contrast is in the right place: a punchy histogram with all the contrast in the background and none on the face is a bad photograph with a good graph.

And it cannot tell you whether the image is pleasant. Every histogram-based rule ever written (fill the range, avoid clipping, aim for a bell curve) produces competent, flat, characterless pictures when followed mechanically. The graph is a measuring instrument that catches specific failures. The decision is still yours, made by looking at the photograph.