Short answer
White balance applies one correction to the whole image, but a room lit by a window and a warm lamp has two illuminants. Balance for the window and the lamp-lit areas turn orange; balance for the lamp and the daylit areas turn blue. Your eye adapts locally and barely notices. The fix is to reduce the problem at the light — switch sources off, match them, or filter one to the other — or to correct different areas separately afterwards.
White balance is a set of channel gains applied to every pixel. It can neutralise one illuminant exactly. In a room with daylight from a window at around 5500–6500 K and warm lamps at 2700 K, surfaces lit mostly by one or the other need different corrections, and blends of both need something in between. Any single setting leaves some areas cast. Faces lit on one side by a window and on the other by a lamp show it most: one cheek neutral, the other orange.
Human colour constancy is partly local. The visual system judges a surface against its surroundings, so a white wall in lamplight and a white sheet by the window both read as white even though the light reaching the eye from each is very different. A photograph flattens the scene into one picture viewed later under a third light, and the local adaptation that happened in the room does not happen when you look at the print or screen. The camera did record what was there; the eye was never seeing it that way.
The most reliable approach is to make the light consistent before taking the picture. Turn off the lamps and use the window, or close the curtains and use the lamps. Replace mismatched bulbs with ones of the same colour temperature and a good colour rendering index. Film and photography crews put colour-correction gels over lights or windows to convert one source to the colour of the other — warming gels on daylight or cooling gels on tungsten — and then set one white balance for the result. Tunable LED panels do the same electronically.
Green casts often come from mixing fluorescent or cheap LED sources with others; gels and white balance tint controls include green–magenta correction for this reason.
When the light could not be controlled, correct locally. Raw editors allow masked adjustments: set white balance for the dominant light globally, then paint a mask over the areas lit by the other source and shift their temperature and tint separately. Some phone pipelines attempt a local correction automatically, with mixed success. Local correction works best on raw files; on JPEGs the baked-in balance limits how far colours can be pulled without artefacts.
Why: Window light on one side, tungsten lamp on the other.
Fix: Turn off the lamp or turn the subject to face the window; otherwise mask-correct each side.
Why: Fluorescent or low-quality LED tubes mixed with other light.
Fix: Correct the tint for the dominant source; use higher-CRI lamps for future shoots.
Why: Dusk daylight and warm lamps together; one balance cannot suit both.
Fix: Shoot earlier or later when one source dominates, or mask-correct the window.
Each statement is labelled by kind — established fact, a standard’s requirement, observed market data, a convention, or Colourwise’s own interpretation or analysis — with the strength of the evidence behind it.
FactStrong evidence
Camera white balance is a global linear scaling of the colour channels, so it can neutralise only one illuminant colour at a time.
FactModerate evidence
The perceived colour of photographs taken under one light and viewed under another looks tinted because the colour constancy that operated in the scene does not transfer to the viewing situation.
Colourwise interpretationModerate evidence
Matching light sources before capture is more reliable than correcting mixed lighting afterwards.
Based on: Follows from white balance being a single global correction; local correction in editing is an approximation limited by how light sources blend across surfaces.
Reviewed 29 September 2026. Colourwise summarises its sources in its own words and does not reproduce standards text or proprietary colour data. Spotted an error? Tell us.