Two colors. More than one number.

CIELAB derives coordinates from XYZ relative to a chosen reference white. L* represents model lightness; a* runs between green and red regions and b* between blue and yellow. These are not three light sources like RGB components, nor is L* the video luma component Y′.

The space makes approximate color differences more convenient, but is not perfectly perceptually uniform. Equal distances in different Lab regions need not appear equal. Do not confuse CIELAB with Hunter Lab, Oklab or an unspecified “Lab” setting: similar names do not imply identical formulas.

Normalizing XYZ by a reference white is part of the Lab calculation. Two triplets without conditions cannot safely be compared just because both say “Lab”. Align the observer, white and method used to obtain XYZ. Material comparisons depend on illumination and measurement geometry; displays depend on the signal path and viewing conditions.

Both lab samples use D50 and a consistent 2° colorimetric context, with XYZ normalized to Yn = 1. This does not switch the monitor to D50. Previewing separately adapts D50 → D65 and converts to sRGB. The numeric pair used for differences remains unchanged.

First form X/Xn, Y/Yn and Z/Zn. Each passes through a piecewise function: cube root above a threshold and a linear segment near zero. L* comes from transformed relative Y, while a* and b* come from differences of transformed components. Exact rational constants ε = 216/24389 and κ = 24389/27 avoid unnecessary rounding.

L* = 100 represents the reference white in the model’s normalization, not 100 cd/m² on every device. L* = 50 corresponds to relative Y near 0.184 rather than half the linear light. This nonlinearity is neither sRGB encoding nor Resolve’s Gamma control. The near-black linear branch defines behaviour without using a cube root everywhere.

C* = √(a*² + b*²) measures distance from the neutral axis on the a*b* plane. h° comes from atan2(b*, a*) with correct quadrant handling. L* is unchanged. Switching Lab → LCh does not move the color into a different space or add gamut.

At zero chroma, hue direction is undefined: the lab says so rather than assigning grey a red hue. Angles wrap, so 359° and 1° are not hugely separated. An angular step nevertheless depends on chroma and is not a ready-made ΔE. Nor should C* be labelled HSL saturation percentage without explanation.

ΔE*ab, also called ΔE76, is the square root of squared ΔL*, Δa* and Δb* summed together. Keep signed component differences separately: the total distance does not say whether a sample became lighter, greener or yellower. Different deviations can share the same distance.

Its simplicity makes it a useful first numerical reference, but does not remove CIELAB’s nonuniformity. An xy distance is not ΔE76: coordinates and geometry differ. Substituting raw RGB or XYZ into the formula and calling the result ΔE76 is equally misleading. Establish a valid common Lab first.

CIE94 adds weighting for lightness, chroma and hue differences. Common graphic-arts and textile settings use different parameters; the lab makes that choice explicit. The samples do not change — the evaluation model does.

In this formulation, weights depend on the first, reference sample’s chroma. Swapping A and B can therefore change ΔE94. State that convention in the report instead of silently averaging it away. ΔE76 and CIEDE2000 with the same parameters are symmetric, but CIE94 should not be required to behave that way. A tolerance agreed for one formula and conditions does not transfer merely because both names start with ΔE.

CIEDE2000 keeps CIELAB coordinates but evaluates their difference with corrections for lightness, chroma, hue and their interaction. It is not a new RGB space or an automatic image-correction method. The lab uses kL = kC = kH = 1 and states that alongside the result.

Correct hue branches around 0/360°, zero-chroma cases and the signed cross term are particularly important. Simple examples may miss implementation defects. The numerical module is therefore checked against the 34 published Sharma, Wu and Dalal pairs in both directions. That verifies computation on a reference set, not applicability to every viewing condition or the entire HDR range.

Read the value with its formula, task, patch size, surround, measurement conditions and acceptance criterion. Detecting a difference between adjacent patches and rejecting it in a finished production are different decisions. An arbitrary threshold should not simultaneously become a visibility limit, monitor tolerance and client requirement.

ΔE is not a percentage and is not automatically bounded by 0…100. Zero colorimetric difference under specified conditions does not prove identical spectra or identical material behaviour under every illuminant. A reliable conclusion needs a defined task, not merely more decimal places.

A file eyedropper reports image values in an application context, not light emitted by your monitor. An instrument links a presented patch to the device’s actual output. Material comparisons need aligned illuminant, observer and geometry; displays additionally need stable state and a defined signal path.

Do not mechanically combine coordinates from different reports without checking conditions. Repeat measurements, inspect variability and ensure displayed precision is not hiding instability. Changing an application profile and physically correcting a device are different actions. This page does not access your display or colorimeter or issue measurement certificates.

A color chart or known patch can serve as a reference when capture and comparison conditions are defined. Matching several patches does not guarantee camera agreement across every scene. Select comparable regions: highlights, noise, exposure changes or mixed lighting alter the data.

Scopes show signal distributions; ΔE compares specified samples. Do not average a whole frame to one color and call that grading accuracy. Minimizing ΔE between intentionally different shots is not an editing objective. For a technical transform, define independent reference samples and keep creative decisions separate.

Differences are computed from the original Lab pair. Display uses a separate path: Lab/D50 → XYZ → D65 adaptation → sRGB. If resulting RGB falls outside 0…1, only the preview is clipped and a warning appears. The numeric comparison is not replaced by the difference between clipped patches.

CIELAB and CIEDE2000 cannot be applied indiscriminately to arbitrary absolute HDR values without checking normalization and applicability. This is neither a 1,000-nit simulation, a gamut measurement nor a promise of identical browser appearance. Similar preview patches may conceal differing out-of-gamut coordinates; viewing conditions can also produce the opposite impression.

Start with the goal and sample list. Preserve original coordinates, reference white, observer, comparison direction, formula and parameters for every pair. For a set, report sample count, mean, median and maximum instead of only the most flattering summary. Keep the worst sample available for inspection.

Agree acceptance criteria before measurement, including their source and permitted variation. Do not choose a formula afterwards to obtain a smaller number. If Lab comes from a file, call it a calculated comparison; if from an instrument, state the instrument and conditions. Preserve software version and reference pairs for repeatability. A clear verification boundary is more useful than an unsupported claim of perfection.