Short answer
Two phones differ at every stage: their sensors' spectral sensitivities, their lenses and filters, the white balance each guesses, the colour matrix each applies, and above all the tone and colour tuning each manufacturer chose. Then each photo is viewed on a different screen. With so many independent choices, agreement would be the surprise. For colour you need to trust, control the light, include a neutral reference and compare on one calibrated display.
No two sensor designs have identical spectral sensitivities, and none matches human vision. Measurements across 28 cameras found a family of similar-shaped curves that nevertheless differ, including in how far each departs from the colour-matching functions. So even before any processing, two phones can record a pair of fabrics, a painted wall or a skin tone with different channel ratios. Under daylight the differences may be small; under narrow-band LED or fluorescent light, where small sensitivity differences meet spiky spectra, they grow.
Each phone estimates the illuminant for itself. Scenes that are ambiguous for auto white balance — a room with one strong wall colour, warm lamps mixed with window light, a sunset — are exactly where two algorithms diverge. One phone may neutralise the warmth; another may keep some of it on purpose. The difference is often larger than the hardware difference, and it is not an error by either phone: they are answering an ill-posed question with different assumptions.
After white balance comes the look. Tone mapping decides how bright shadows are and how much contrast midtones get; hue and saturation tuning decides how blue a sky, how green a lawn and how warm a face should be. Google has published that its HDR+ pipeline deliberately pushes skies and foliage towards more appealing hues; others make comparable choices without publishing them. Reviewers who compare phones side by side are, to a large extent, comparing these preferences.
If one phone's photos look 'more accurate', check them against the real object under the same light, not against the other phone.
The comparison usually happens on the phones' own screens, which differ in gamut, white point, brightness and automatic adjustments such as night-time warming or ambient-light matching. A photo that looks warmer on one phone may be the same file shown on a warmer display. Move both photos to one colour-managed, calibrated screen before deciding which camera is closer to reality.
Why: Different phones, white balance guesses and screens.
Fix: Agree one light, include a grey card, and view all photos on one calibrated screen.
Why: Display differences and unmanaged apps, not only the camera.
Fix: Tag images as sRGB and describe colour in words or with a physical reference as well.
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
Measured spectral sensitivities of 28 cameras, including a phone, differ from one another and in how far each deviates from the Luther condition.
FactModerate evidence
Auto white balance is ill-posed and struggles in dim or strongly coloured light, so different algorithms can reasonably produce different results for the same scene.
FactStrong evidence
At least one manufacturer's published pipeline applies deliberate aesthetic hue and saturation adjustments to skies and vegetation.
Colourwise interpretationLimited evidence
In everyday comparisons, differences in white balance choice and aesthetic tuning usually outweigh differences in sensor hardware.
Based on: Colourwise's reading of the published pipeline and white-balance sources against the relatively similar sensitivity curves measured across cameras; not a controlled comparison of phones.
Caveat: Under narrow-band lighting hardware differences can dominate.
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.