White Balance: Deciding What White Is

Different light sources are different colours. Candlelight is orange, midday sun is near white, light in shade is blue. The eye adapts on its own — a white page under a tungsten lamp still looks white to you. The camera does not, which is what the white balance setting compensates for.

But “compensate” already assumes a position. What this chapter is really trying to install is the question you ask on location:

What colour is the light here, and do I want to keep it or remove it?

Most frustration with white balance comes from treating it as a question with a right answer. It is not one — and the sooner it stops being one, the sooner the setting becomes useful rather than anxious. Everything below is either a mechanism that explains why the automatic answer goes wrong, or a way of deciding what you want instead.

Why colour temperature is a temperature

The unit puzzles people: what does the colour of light have to do with heat?

Everything, and literally. Put a bar of iron in a furnace and it glows dull red, then orange, then yellow-white, and hotter still it goes blue-white. Any heated body emits a spectrum determined by its temperature, and the hotter it gets the shorter the wavelengths it favours.

Which gives a convenient description: at what temperature would a blackbody glow closest to the colour of this source? That number is its colour temperature.

Two things follow.

• It is an analogy, not a measurement. The sun’s surface really is near 5,800 K, which is a happy coincidence; but a 3,000 K LED bulb is cold to the touch. It merely looks like a blackbody at 3,000 K.
• The numbers run against ordinary language. We call orange warm and blue cool, yet orange is the low number and blue the high one. Beginners get stuck here not because the physics is hard but because the words are fighting.

Temperature	Source	Colour of the light
1800 K	Candle, campfire	Very orange
2700 K	Tungsten bulb, warm white LED	Orange
4000 K	Fluorescent, neutral white LED	Greenish white
5500 K	Midday sun	White
6500 K	Overcast	Slightly blue
8000 K and up	Shade, snow	Blue

The camera setting is a counter-correction: set it to 2700 K and the camera adds blue to cancel the orange. Hence the practical mnemonic — raise the number and the photograph warms up; lower it and it cools down.

A hundred kelvin is not always a hundred kelvin

Something the scale hides: equal steps in kelvin are not equal steps in appearance. Moving from 2700 K to 2800 K is a visible change; moving from 8000 K to 8100 K is invisible. The scale is stretched at the warm end and compressed at the cool end, which is why the temperature slider feels oddly insensitive on blue scenes and twitchy on tungsten ones.

Film-era practice fixed this with the mired — a reciprocal unit, one million divided by the temperature. Equal differences in mireds correspond to roughly equal visible shifts anywhere on the scale, which is why conversion filters have always been rated in mireds rather than kelvin: a filter that shifts 130 mireds does the same perceptual work whether you point it at a bulb or at the sky.

The practical residue for a digital photographer is a rule of thumb: whatever precision you need at 3000 K, you need about seven times less of it at 8000 K. If you find yourself nudging temperature in tiny increments on a blue scene, you are almost certainly chasing something that is not visible.

Why there are two axes

Beyond temperature, the camera offers a second adjustment: green↔magenta. Most makers label it tint, or G↔M. Fujifilm does not provide that axis at all — its WB SHIFT has only R (red↔cyan) and B (blue↔yellow), so reaching magenta means moving both together. Whatever the interface calls it, this is not a refinement. It is necessary.

That heated-iron path traces a curve through colour space called the blackbody locus. Sun, bulbs and flames land on it or very near, so a single temperature describes them.

Fluorescent tubes and LEDs do not glow because they are hot. A fluorescent tube is ultraviolet striking phosphor; an LED is a semiconductor plus a phosphor conversion layer. Their spectra are a few spikes and some empty stretches rather than a smooth curve, and their colour falls off the locus, usually towards green.

With only the temperature axis you can never land: warm it up and the frame goes yellow-green, cool it down and it goes blue-green. The green never leaves. The second axis exists to pull it out.

This also covers a teaching point that cannot be photographed well. A green cast is intrinsically hard to find an example of — nobody sets out to photograph a defect, and a decade of supermarkets, hospitals and convenience stores re-lamping to well-corrected LED has made it rarer in the wild. So it goes in a table instead:

Source	Approximate temperature	Departure from the locus	What to do on location
Tungsten	2700 K	Essentially none	Temperature axis alone is enough
Older fluorescent	3500–4500 K	Clearly green	Set temperature, then push tint towards magenta
Cheap LED	3000–6500 K	Green or magenta, varying by maker	Take a test frame and look at skin
Low-pressure sodium lamp	About 1700 K	Extreme; nearly monochromatic	Not correctable — accept it or convert
High-pressure sodium lamp	About 2000 K	Large, but the spectrum is broader	Correctable to serviceable; do not expect accuracy
Overcast daylight	6500 K	Essentially none	One axis is enough

The low-pressure row deserves a sentence. Its spectrum is essentially one narrow spike, so the information about other colours was never in the scene to begin with. No amount of white balance recovers it, because there is nothing to recover. This is the dark side of the metamerism from chapter eleven: when the three channels receive no information, every later adjustment is guesswork. Newer high-pressure sodium and LED street lighting is nowhere near as extreme and does respond to correction — so the severity of this rule changes from one junction to the next.

Two readings in one frame

It demonstrates something easy to miss: “the colour temperature of this scene” is frequently a thing that does not exist.

Why is shadow blue? Because objects in shadow are not unlit — they are merely out of the sun. What lights them is the entire blue sky, an enormous source above 8,000 K. Shadow on a clear day is blue not because it is cold but because it is lit by something blue.

Two consequences. Portraits made in open shade get their subject sitting in natively cool light while skin stays warm — the warm-cool separation of the last chapter, handed to you free. And whenever a frame contains both sunlight and shadow, no single white balance value can satisfy both. You are choosing a side.

The two ways auto white balance fails

AWB is accurate most of the time. When it does something you did not want, it is usually one of these.

First: the scene genuinely is one colour.

AWB works roughly by assuming the whole frame should average to neutral grey. That assumption holds surprisingly well in ordinary scenes and misfires regularly whenever one colour dominates the frame — sunsets, autumn foliage, blue hour, a field of grass, underwater. This is not a poorly written algorithm; the assumption itself contradicts those scenes. Recent bodies add scene recognition to compensate, and will sometimes identify a sunset and deliberately keep the orange — but that is one layer of guessing stacked on another, and you cannot predict which way it will go this time.

Second: mixed sources.

Mixed lighting has a one-line policy: serve the people first. If there is a person in frame, set white balance so that skin looks right and let the rest drift; viewers attribute the remaining cast to the room. Skin is one of the memory colours from chapter eleven and nobody forgives it being wrong. Nobody notices a blue wall.

There is no correct, only what you want

Technically correct white balance renders a neutral object as neutral. But photography is not measurement.

• Photograph a sunset and you want the orange.
• Photograph snow at dawn and the blue is what makes it feel cold.
• Photograph a candlelit dinner and the warmth is the atmosphere.

Treat white balance as a creative dial, not a calibration tool. Put differently: the question is not what white is here, but what kind of light the viewer should feel they are standing in.

A useful compromise in practice is correcting halfway. If the scene runs 800 K warm, pull back 400 K — the light keeps its character without becoming thick enough to read as a mistake. Viewers are surprisingly good at telling deliberate warmth from a forgotten setting, and the boundary sits about there.

Measuring instead of guessing

Three methods remove the judgement entirely.

Find something in frame known to be neutral. A white wall, concrete, asphalt, a white car, newspaper. Click it with the eyedropper afterwards and the temperature is settled. Free, provided the thing really is neutral — cream walls and warm-toned concrete will lie to you, and in a direction you have no way of knowing.

A grey card. One frame containing the card under the same light becomes the reference for the whole set. The most reliable option; the cost is carrying it and reshooting whenever the light changes.

The camera’s custom white balance. Photograph a white sheet and let the body compute the correction. On Fujifilm, CUSTOM 1/2/3 sits below the three AUTO entries and above COLOR TEMPERATURE — not at the end of the list, where people tend to look for it. Fastest of the three, but it bakes the result into a number, leaving only fine adjustment later unless you also recorded RAW.

Worth knowing: Fujifilm’s AUTO is really three modes, and many people have never checked which they are using.

Mode	How it decides	Suits
AUTO	A balanced compromise across scenes	The default, for anything
AUTO WHITE PRIORITY	Drives white objects harder towards pure white	Clean, neutral documentation
AUTO AMBIENCE PRIORITY	Deliberately preserves the warmth of the ambient light	Interiors, candlelight, atmosphere

The third is this entire chapter turned into a button: it does not correct warmth, it keeps it. If AWB has been washing the mood out of your interiors, switching to it usually settles the matter.

Six common symptoms and their causes

1. Interiors come back orange and faces look ill → daylight balance used under tungsten or warm LED, or AWB dragged off by a large warm area → set 2800–3200 K manually; if window light is also present, favour skin and correct halfway.
2. Skin looks right but a green layer sits over everything → the source is off the blackbody locus and the temperature axis cannot reach it → move towards magenta: one to three steps on the tint axis if your body has one, or on Fujifilm R and B together, one to three steps positive on each.
3. Snow photographs grey and blue → two problems stacked: metering pulled down by a bright field (exposure) and shadow lit by blue sky (white balance) → apply +1 to +2 EV first, then decide how much blue to keep.
4. The sunset is far paler than it looked → AWB treated the dominant colour as a cast → switch to fixed Daylight; in RAW you can also pull it back afterwards.
5. Consecutive frames of one situation jump in colour → AWB re-decides every frame and the dominant colour’s share changes as you recompose → lock a fixed value for the whole set, which matters especially for work meant to be shown as a series (chapter sixteen).
6. However you grade it, one area stays wrong → mixed sources needing different values in different regions → correct one region properly and treat the other locally; or accept that it cannot be saved and convert, which makes the problem vanish (chapter ten).

White balance used to be something you bought

That white balance is a menu item is a very recent development. In the film era it was decided when you bought the film.

A colour emulsion had to be balanced for one temperature, and essentially two existed: daylight film at about 5500 K and tungsten film at about 3200 K. Buy wrong and the whole roll was wrong.

Converting between them meant putting glass in front of the lens. Kodak’s Wratten numbers are still in use: 80A is blue, letting daylight film work under tungsten; 85B is amber, letting tungsten film work outdoors. These filters cost light — about two stops for the 80A, about two-thirds of a stop for the 85B — so correcting white balance was paid for in shutter speed.

Two things follow from that history. The number 5500 K is an artefact of that period — an average of temperate midday sun, not a physical constant; the little sun icon in your menu inherits a Kodak specification. And the colour of your curtains and lampshades counts, which is why film-era studios cared about wall colour: bounced light arrives carrying the colour of whatever bounced it. Digital did not repeal this, it only made small amounts of it repairable — and not when it lands on a face.

Understanding this explains the shape of the presets: there are six or seven of them rather than a continuum because they are fossils of film and filter combinations.

Where RAW and JPEG differ most

White balance is one of the few settings a RAW file leaves entirely open. RAW records the sensor’s values and white balance is a tag; changing it later costs nothing.

But working the Fujifilm straight-out-of-camera way — the reason many people buy Fujifilm — makes it a decision at capture, interacting with the film simulation. The next part covers that in detail.

Practical suggestion: record RAW + JPEG together. The JPEG gives immediacy, the RAW keeps a door open. Both the X-T5 and the X-E4 will do it.

One exception is worth knowing: extreme sources — sodium, underwater, single-colour LED — are not rescued by RAW either, for the reason in the table above. RAW protects your right to decide later; it does not manufacture information that was never recorded.

White balance shift

Beyond temperature, Fujifilm offers a two-axis shift: red↔cyan (R) and blue↔yellow (B), ±9 steps each. These axes are Fujifilm’s own — Canon, Nikon and Sony use amber↔blue and green↔magenta instead, so shift values found online cannot be copied across brands. This is for fine styling, not for correction.

• A touch towards cyan (R −1 or −2) → cleaner skin with less red
• A touch towards blue (B +1) → cooler shadows, pairs well with Classic Chrome
• Towards yellow (B −2) → a little more daylight warmth

It lives under IMAGE QUALITY SETTING → WHITE BALANCE; choose a mode, then press MENU/OK to reach the shift grid. The shift is remembered per white balance mode — what you set for AUTO does not apply to DAYLIGHT, which regularly convinces people the setting did not take. Chapter twenty-one puts it to work.

A table for the field

What you see	How to set it	Why
Sunset, autumn colour, blue hour	Fixed Daylight, never AUTO	AWB reads the dominant colour as a cast
A portrait in open shade	Shade, or about 7000 K manually	The source is sky, not sun; some blue left in is an asset
Mixed indoor light	Favour skin, correct halfway	Skin is a memory colour; walls are not
Fluorescent or cheap LED	Set temperature, then push towards magenta (on Fujifilm, R and B together)	The source is off the blackbody locus
Low-pressure sodium, underwater	Stop fighting; consider black and white	The information was never recorded
A set to be published together	Lock one value for the whole set	AWB jumps frame to frame and the set reads as noise