Colour temperature both ways round, with the filter shifts that only make sense in one of them — and a demonstration of why the other one cannot be added up.
Calculator
Sources:
3,200 K — warm — tungsten and warm white
312.5 mireds
a million divided by the kelvins · getting to 5,500 K needs a -131 mired cooling shift
The scale, and why it reads backwards
Ascending in kelvins is descending in mireds — and in warmth
Source
Kelvin
Mireds
Note
Candle flame
1900
526
The warmest light most people ever see, and the reason low numbers read as cosy.
Sunrise or sunset
2200
455
Low sun, reddened by the atmosphere it has travelled through.
Tungsten household lamp
2800
357
The old incandescent bulb. Genuinely a temperature — the filament really is that hot.
Warm white LED
3000
333
Chosen to imitate tungsten. Nothing inside is at 3000 K.
Studio halogen
3200
313
The photographic tungsten standard, and what tungsten-balanced film was made for.
Neutral white LED or fluorescent
4000
250
Offices and kitchens. Reads as neither warm nor cool indoors.
Noon daylight
5500
182
The photographic daylight standard. Direct sun with a clear sky.
Standard daylight D65
6500
154
The reference white for sRGB and for nearly every display you own.
Overcast sky
7000
143
Cloud scatters the warm end out, leaving a cooler light than direct sun.
Open shade under blue sky
10000
100
Lit only by the sky. The bluest ordinary light there is.
“Warm” light is a low kelvin number and “cool” is a high one, which is backwards from every other use of temperature and catches people out in lighting shops daily. Mireds put it the right way round: a higher mired figure is a warmer light. The physics is unarguable — a filament at 2800 K really is cooler than one at 3200 K, and glows redder for it — but the everyday words point the other way.
Stacking filters
Tap to add:
80A-131 miredsthe shifts simply add
Result5,510 Kfrom 3,200 K, a change of 2,310 K
This is the whole reason the mired exists. Filter strengths add — an 81A and an 81B together are simply +45 mireds — and there is no equivalent arithmetic in kelvins, because the kelvin change a filter produces depends entirely on what it is pointed at. A filter labelled “+300 K” would be meaningless; one labelled “+18 mireds” is a specification.
The same filter, pointed at different lights
One -131 mired shift, across the ladder
Starting light
Becomes
Change in kelvins
Candle flame (1900 K)
2,530 K
+630 K
Sunrise or sunset (2200 K)
3,091 K
+891 K
Tungsten household lamp (2800 K)
4,422 K
+1,622 K
Warm white LED (3000 K)
4,942 K
+1,942 K
Studio halogen (3200 K)
5,510 K
+2,310 K
Neutral white LED or fluorescent (4000 K)
8,403 K
+4,403 K
Noon daylight (5500 K)
19,678 K
+14,178 K
Standard daylight D65 (6500 K)
43,771 K
+37,271 K
Overcast sky (7000 K)
84,337 K
+77,337 K
Open shade under blue sky (10000 K)
unreachable
past infinite kelvins
One filter, one fixed mired shift, and the kelvin column is all over the place — a few hundred at the warm end and tens of thousands at the cool end, with the bluest row unreachable entirely because you would need negative mireds. Any claim that a filter or a white-balance adjustment is worth “so many kelvins” is only true for one starting point.
Why a thousand kelvins is not a fixed amount
2000 K to 3000 K167 miredscandle to warm white — enormous
6000 K to 7000 K24 miredsdaylight to overcast — barely visible
Ratio7×against a visible threshold near 5 mireds
The same thousand kelvins is seven times bigger a change at the warm end than at the cool end, because the effect follows the reciprocal. That is why lamp ranges step 2700, 3000, 4000, 5000, 6500 rather than in even thousands, and why mireds — where equal steps look roughly equal — are the unit anyone actually adjusting light reaches for.
What the number cannot tell you
Limits of a single correlated colour temperature figure
What it omits
Why it matters
Distance from the Planckian locus
Two lamps at the same CCT can sit either side of it, one greenish and one pinkish. The number is the same and the light is not.
Colour rendering
How faithfully colours appear is a separate measurement entirely. A 4000 K lamp can render skin tones well or badly with no change to its CCT.
Spectral gaps
A spectrum with a hole in it can still average to a given CCT. What is missing does not show up in the figure at all.
Whether anything is at that temperature
For an LED or a fluorescent tube, nothing is. The CCT names the nearest black body, which is a comparison rather than a property.
A tungsten filament at 2800 K genuinely is at 2800 K — the light is what a body at that temperature emits. An LED marked 3000 K is not: the figure names the nearest black body to its colour, which is a comparison rather than a property. That is what “correlated” is doing in the name, and it is why two lamps sharing a number can still visibly disagree.
What this converter covers
Kelvins and mireds, ten reference light sources, eleven standard filters that stack by addition, and the same shift shown across the whole ladder.
Kelvins to mireds and back, and the shift between any two sources
Standard photographic conversion and light-balancing filters, stacked
The same filter applied to candlelight, tungsten, daylight and open shade
Why a thousand kelvins is seven times bigger a change at 2000 than at 6000
What a single correlated colour temperature figure leaves out
Shifts that add Kelvins do not Ten reference sources K ↔ mireds
Free, no signup — exact by definition, not an estimate.
Updated 8 September 2026
At a glance
Formula shown
mireds = 1 000 000 ÷ kelvins · a filter is a fixed mired shift, and shifts add
Scenario support
3200 K is 312.5 mireds · an 80A is −131 mireds, which takes 3200 K to 5500 K
Educational estimate
Planning support from the values you enter — not professional advice.
A million over the kelvins
The conversion itself is one line: mireds = 1,000,000 ÷ kelvins, and the same line backwards. The name is short for micro-reciprocal-degree, which is exactly what it says it is.
What makes it worth a page is why anyone would want the reciprocal. Colour temperature is one of those quantities where the number people use and the number that behaves are not the same number.
Familiar sources in both units
Source
Kelvin
Mireds
Candle flame
1900
526
Tungsten lamp
2800
357
Studio halogen
3200
313
Noon daylight
5500
182
Display white, D65
6500
154
Open shade
10000
100
Look at the spacing. In kelvins the warm sources are crowded together and the cool ones spread out; in mireds it is the other way round, and the mired spacing is much closer to how different the lights actually look.
That is the entire design goal, and everything else on this page follows from it.
Why filters are labelled in mireds
A colour-correction filter does a fixed thing to the light passing through it. The question is what unit describes that fixed thing, and kelvins cannot.
Take an 80A, the standard blue conversion filter. It is −131 mireds. Here is what that single filter does depending on what it is pointed at:
One filter, four starting points
Starting light
Becomes
Change
Tungsten, 2800 K
4422 K
+1622 K
Halogen, 3200 K
5510 K
+2310 K
Daylight, 5500 K
19678 K
+14178 K
Open shade, 10000 K
unreachable
past infinity
Same filter, and the kelvin column runs from sixteen hundred to fourteen thousand and then off the end of the scale entirely — because taking 10,000 K down by 131 mireds would need a negative mired value, and there is no such light.
So a filter labelled “+300 K” would be meaningless, while one labelled “+18 mireds” is a specification that holds wherever you point it. And the payoff is that shifts add: an 81A and an 81B stacked are simply +45 mireds, with no arithmetic in between.
The same logic runs through white balance in a camera and colour temperature adjustment in editing software: the underlying control is a reciprocal one, even where the slider is labelled in kelvins.
Equal steps are not equal
The second consequence of the reciprocal is that a thousand kelvins is not a fixed amount of anything visible.
From 2000 K to 3000 K is 167 mireds — the difference between candlelight and a warm white bulb, which nobody could miss. From 6000 K to 7000 K is 24 mireds, and side by side most people would struggle to say which was which.
The ratio is exactly seven. Same thousand kelvins, seven times the visible change, purely because of where on the scale it sits.
This is why lamp ranges step 2700, 3000, 3500, 4000, 5000, 6500 rather than in even thousands — those are roughly even steps in mireds, which is to say roughly even steps in appearance. The spacing looks arbitrary in kelvins and is not.
It also sets a useful sense of scale for the filter numbers. A shift of about five mireds is around the threshold of being noticeable side by side, so the smallest standard warming filter at +9 is a deliberate, visible nudge, and a +131 conversion filter is an entirely different light.
Warm is the low number
A smaller problem, and one that catches people in lighting shops rather than in studios: the vocabulary runs backwards.
“Warm white” is 2700 K. “Cool white” is 5000 K or more. The light everyone calls warm has the lower temperature, which is the opposite of how the word behaves everywhere else.
The physics is not confused, only the language. A filament at 2800 K really is cooler than one at 3200 K, and a cooler body glows redder — think of an electric hob going from dull red to orange as it heats. Red genuinely is the cool end.
“Warm” and “cool” are describing how the light feels, borrowed from the association of red with fire and blue with ice, and those associations point the other way from the thermometer. Both are internally consistent; they simply disagree.
Mireds sidestep it neatly, since a higher mired figure is a warmer light. It is a small thing, but it is one more reason the people who work with this daily use the reciprocal and everyone else does not.
Not a temperature at all
One last distinction, and it is the one that matters most for modern lighting.
A tungsten filament at 2800 K genuinely is at 2800 K. The light it emits is what a body at that temperature emits, and the colour temperature is a physical description of the source.
An LED marked 3000 K is not at 3000 K, and nothing inside it is. The figure is a correlated colour temperature: the temperature of the black body whose colour most closely resembles the LED’s. It is a comparison, not a property, and the word doing that work is the one everyone drops.
The consequence is practical. Two lamps at the same CCT can sit on opposite sides of the Planckian locus — one visibly green, one visibly pink — and the number will not distinguish them. Colour rendering is a separate axis again: a 4000 K lamp can flatter skin tones or ruin them without the CCT moving at all.
Which is why lighting standards specify a chromaticity region around each nominal CCT rather than a point, and why two lamps sold as “4000 K” from different makers can be visibly different on the same ceiling. If matching matters, the CCT is where the specification starts and not where it ends.
Related calculators
Other colour, light and imaging tools:
ColourHex, RGB and HSL in both directions, with the WCAG contrast ratio on white and black — because that is usually the real question.
Exposure StopTrade stops between aperture, shutter and ISO with the exposure held — and see why the printed f-numbers are not the exact series.
IlluminanceLux, foot-candles, lumens, candela and nits — asking for the area, the beam angle or the surface, because three of those pairs have no answer without one.
TemperatureCelsius, Fahrenheit, kelvin and Rankine — and temperature CHANGES, which convert differently from readings.
Frequency to WavelengthFrequency, wavelength, photon energy and wavenumber — in vacuum and in the medium, including coax velocity factors.
Pixel DensityPPI, dot pitch and physical size from any resolution — plus pixels per degree, which is what decides whether you can see the pixels.
The first defines what a correlated colour temperature is and what it deliberately leaves out; the second is the measurement basis behind it; and the third is where the familiar 2700, 3000, 4000 and 6500 bins come from, along with the tolerance that explains why two lamps at one nominal figure need not match.
International Commission on Illumination (CIE)CIE · verified 2026-09-08 · The definition of correlated colour temperature as the temperature of the Planckian radiator whose perceived colour most closely resembles that of the source, and the chromaticity framework in which distance from the Planckian locus is a separate quantity from the temperature itself
Sensor Science — optical radiation measurementNational Institute of Standards and Technology · verified 2026-09-08 · The measurement basis for spectral radiance and chromaticity from which a correlated colour temperature is derived, and the point that two sources agreeing on that single figure can differ substantially in spectrum
ANSI and NEMA lighting standardsNational Electrical Manufacturers Association · verified 2026-09-08 · The nominal correlated colour temperature bins used for solid-state lighting products — 2700, 3000, 3500, 4000, 5000 and 6500 K — and the chromaticity tolerance around each, which is why two lamps sold at the same nominal figure need not match
Conversion note
The reference colour temperatures here are representative figures for common light sources, not measurements of any particular lamp or any particular sky. Daylight in particular varies continuously with the sun's elevation, cloud, atmospheric conditions and what is reflecting into the scene, and a single figure for it is a convention rather than an observation. Filter mired shifts are nominal values for the standard photographic filters and vary between manufacturers and batches; a critical application should be metered rather than calculated. Correlated colour temperature says nothing about colour rendering, about spectral completeness, or about how far a source sits from the Planckian locus, so two sources agreeing on the figures here can look visibly different — for colour-critical work, measure the source with a meter that reports tint and rendering as well as CCT. Nothing here addresses display calibration, which involves a white point, a transfer function and a gamut, and cannot be reduced to a colour temperature.
Published the Kelvin to Mired Converter: colour temperature in both directions, ten reference light sources, eleven standard photographic filters that stack by addition, and the shift needed between any two sources.
Demonstrates rather than asserts why the reciprocal unit exists. One 80A filter is a fixed -131 mireds; it moves tungsten at 2800 K by +1622 K, daylight at 5500 K by +14178 K, and cannot move open shade at 10000 K at all because that would need negative mireds.
Stacks filters by simple addition, which is the property the mired was invented to have -- an 81A and an 81B together are exactly +45 mireds, and there is no equivalent arithmetic in kelvins.
Records that equal kelvin steps are not equal visible steps: 2000 K to 3000 K is 167 mireds and 6000 K to 7000 K is 24, a ratio of exactly seven, which is why lamp ranges step 2700, 3000, 4000, 5000 and 6500 rather than in even thousands.
Explains the inverted vocabulary -- warm light is the LOW kelvin number -- and that the physics is not confused, only the words, since a cooler body genuinely glows redder.
Separates a correlated colour temperature from a temperature. A tungsten filament at 2800 K really is at 2800 K; an LED marked 3000 K is not, and the figure names the nearest black body rather than a property of the source.
Lists what a single CCT figure cannot carry: distance from the Planckian locus, colour rendering, and spectral gaps -- which is why two lamps sold at one nominal figure can look visibly different.
Verified by 93 automated cases, asserting that stacking equals summing, that one fixed mired shift produces kelvin changes more than five times apart across the ladder and one unreachable case, that a shift past zero mireds returns null rather than a large number, and that kelvins ascend while mireds descend on every adjacent pair.
Add this calculator to your site
Responsive embed — and private: nothing your visitors type leaves their browser.