Two light sources, one global slider

Two axes, not one

White balance has two controls because light varies in two directions.

Temperature runs blue to yellow, measured in Kelvin, and the scale is counter-intuitive: a low number like 2700 K is warm orange candlelight, a high number like 7000 K is cool blue shade. The number describes the light the picture was shot under, so setting the slider to 2700 K tells the software “this was lit by something orange” and it compensates by adding blue.

Tint runs green to magenta, and it exists because real light sources are not on the blackbody curve. Fluorescent tubes and cheap LED panels are green; some discharge lamps are magenta. A photograph can have a perfect temperature and still be sickly green, and no amount of the temperature slider fixes it, because green is at right angles to the blue-yellow axis.

Most people only ever touch temperature, and the green cast under office lighting is a tint problem. Learning to reach for the second slider fixes half the difficult casts you will meet.

The reference that makes it objective

Put something known-neutral in the frame. A grey card costs a few pounds; the white side of a colour checker works; a sheet of plain white printer paper is not neutral but is close enough for most jobs; the back of an unprinted receipt is not.

Shoot one frame of the card at the start of each lighting setup. In the developer, click the white-balance eyedropper on it, then copy that setting to every other frame from the same setup. The whole shoot is now neutral in ten seconds.

Two details that matter. Sample mid-grey, not bright white: a white highlight is often clipped, and a clipped channel gives a wrong answer with no warning. And set the sample size to a 3×3 or 5×5 average, because one pixel of a noisy card tells you about the noise.

Without a card, look for something in the frame you know is neutral: a concrete wall, a white shirt in shade, newsprint, a stainless steel surface. Sample it and check whether the result looks plausible. A car’s silver paint and a whitewashed wall are usually close enough.

Mixed light: why one slider cannot win

The common hard case. A room with a window on one side and tungsten bulbs overhead. The subject’s left is lit at 6500 K, their right at 2900 K.

Set the white balance for the window and the tungsten side goes deep orange. Set it for the tungsten and the window side goes deep blue. Split the difference and both sides are wrong, which is what most photographs of Indian weddings, restaurant interiors and hotel lobbies look like.

A global control cannot fix two different casts because there is only one of it. The fix is local:

1. Set the white balance for the dominant light: whichever covers more of the subject, usually the window.
2. Add a Photo Filter or Colour Balance adjustment layer set to cool the remaining orange, or a Curves layer with the blue channel lifted in the midtones.
3. Mask it to the tungsten-lit region, with a large soft brush. The boundary between the two lights in the real scene is gradual, so the mask should be too.
4. Check skin across the transition. Skin is where a mismatch shows.

This takes five minutes and it is the single most valuable local correction in interior work.

The better fix is at capture, and it is cheap: gels. A sheet of CTO (colour temperature orange) over a flash makes the flash match tungsten; CTB does the reverse. A full sheet costs a few pounds and lasts years. Or switch off the tungsten lamps and use the window.

Auto white balance, and why it fails predictably

Automatic white balance works by assuming the scene averages to neutral (the “grey world” assumption) or by looking for the brightest pixel and calling it white.

Both assumptions break on the same kinds of scene, and knowing which tells you when to override:

• A frame dominated by one colour. A field of yellow mustard, a red brick wall, a green cricket pitch. The algorithm sees an average that is not neutral, concludes the light was coloured, and removes the colour of the subject. Your yellow field comes out grey.
• A frame with no neutral in it at all. Nothing to anchor to.
• Sunsets. The whole point of the photograph is that the light is orange, and auto white balance dutifully removes it.

That last one is the useful general principle: white balance is not always meant to be corrected. Firelight should be orange. A blue hour photograph should be blue. “Correct” means “as if under neutral light”, which is a technical statement and not always the right one. Set it neutrally first so you know where neutral is, then move deliberately away from it.

Raw, JPEG, and the cost of getting it wrong

In a raw file, white balance is a pair of multipliers not yet applied. Change it and nothing is lost.

In a JPEG it has been baked in, and correcting it means stretching one channel relative to the others inside 256 levels. Correcting a heavy tungsten cast in a JPEG means multiplying the blue channel by something like 2.5, and a blue channel stretched that far bands, goes noisy, and in the darkest areas has nothing left to stretch. The picture can be made neutral, and it will look worse than one developed from raw.

This is the clearest single argument for shooting raw in any situation with unusual lighting, which is most indoor work, most evening work, and every venue with coloured LEDs.

Free tools: darktable and RawTherapee both have temperature and tint with a spot eyedropper and camera presets. GIMP has Colours > Colour Temperature, which takes an original and an intended temperature, plus Colours > Levels with the grey eyedropper. Photopea has a grey point in Curves and Levels. On a phone, Lightroom mobile’s free tier and Snapseed both expose temperature and tint.

One caution on all of them: correcting white balance changes the apparent exposure, because you are scaling channels. Set white balance first, then exposure, then the tone curve. Doing it in the other order means redoing the exposure.