On film, a coloured filter sits in front of the lens and changes which wavelengths reach the emulsion before exposure. A digital sensor records full colour first, so the equivalent control happens after capture: the three colour channels are weighted and summed into a single grey value. A channel mixer performs exactly this sum, which is why it can imitate a yellow, orange or red filter without any glass. The imitation is close but not identical, and the difference is set by how the sensor saw colour in the first place.

What the Slider Actually Computes

Every greyscale conversion reduces three numbers to one. The default weights are not arbitrary. The luma coefficients in ITU-R Recommendation BT.709 (first issued 1990) assign 0.2126 to red, 0.7152 to green and 0.0722 to blue; the older BT.601 (first issued 1982, the standard-definition standard) used 0.299, 0.587 and 0.114. These are luma coefficients applied to gamma-encoded values, not true linear-light luminance weights, but both sum to one, which holds overall brightness steady and reflects the eye’s strong sensitivity to green and weak sensitivity to blue.

A channel mixer simply lets you choose your own weights. Take two 8-bit pixels: a patch of blue sky reading roughly R 70 / G 110 / B 200, and a red brick reading R 170 / G 60 / B 50. Under the neutral BT.709 weighting the sky converts to 0.2126·70 + 0.7152·110 + 0.0722·200 ≈ 108, and the brick to ≈ 83, so the sky comes out lighter than the brick. Now apply a red-filter mix of R 150% / G 20% / B −70% (still summing to 100%). The sky becomes 1.5·70 + 0.2·110 − 0.7·200 = −13, clipped to 0, near-black; the brick becomes 1.5·170 + 0.2·60 − 0.7·50 = 232, near-white. The dramatic blackened sky and luminous brickwork of a red-filter landscape fall straight out of the arithmetic, with no light removed at all.

The Filter Analogues, in Stops

A glass filter passes its own colour and absorbs the complement, and it always costs exposure. Kodak Wratten factors in daylight set the bill, and each has a mixer equivalent:

• Yellow 8 (K2), ~2× ≈ 1 stop — a gentle blue reduction; the standard “natural” sky filter.
• Green 11 (X1), ~4× ≈ 2 stops — green raised, blue and red trimmed; lightens foliage, darkens skin.
• Orange 21, ~2.5–4× ≈ 1.3–2 stops — red raised, blue cut harder than yellow.
• Red 25, ~8× ≈ 3 stops — heavy red weight, blue strongly negative; deeply darkened sky.
• Deep red 29, ~16× ≈ 4 stops — the most extreme weighting, sky toward black.

These factors rise further under blue-rich open shade or skylight, where there is more blue for the filter to absorb. Ansel Adams reached for a deep red Wratten 29 to render the sky near-black in Monolith, the Face of Half Dome (1927), and set out his filter practice in The Negative (1981). The mixer reproduces each look but charges no stops: you keep the shutter speed and aperture you metered for.

The Tools, and What They Mix

Photoshop’s Channel Mixer expresses the weights as percentages, with a Monochrome checkbox and a Constant offset, and nudges you to keep the total near 100%; you are mixing the three actual channels. Lightroom and Adobe Camera Raw take a different route: the black-and-white mix offers eight colour-band sliders (red, orange, yellow, green, aqua, blue, purple, magenta), not three raw channels. Those work in an HSL-style space, so an “orange” slider lifts orange hues specifically rather than the whole red channel — finer hue control, but a different mental model. Capture One, RawTherapee and Nik Silver Efex Pro sit across this spectrum. The distinction matters: three-channel mixing redistributes the sensor’s actual red, green and blue records, while band sliders remap interpreted hues after demosaicing.

Why the Match Is Imperfect

The limit on the emulation is the colour filter array over the sensor. The Bayer pattern — invented by Bryce Bayer at Eastman Kodak, US Patent 3,971,065, filed 1975 and issued 1976 — lays down 50% green, 25% red and 25% blue photosites; Bayer called the green ones “luminance-sensitive elements,” doubling them to match the eye’s peak sensitivity, the same logic that gives green the 0.7152 luma weight. Demosaicing interpolates the missing values at each site.

Those dye filters are organic pigments deposited photolithographically, with broad, overlapping passbands rather than sharp cutoffs, so the red channel responds partly to green: this is spectral crosstalk, and it worsens as pixel pitch shrinks. Silicon compounds the problem at the long end. Its ~1.1 eV bandgap leaves it sensitive out to about 1100 nm, well into the near-infrared, so colour cameras carry an IR-cut or hot-mirror filter (transition typically 650–720 nm) precisely to stop infrared reaching the channels. Without it, the red channel would be badly contaminated; with it, the residual mixing is the dye crosstalk alone. Either way, the red channel has already absorbed a blend of wavelengths that cannot be unmixed afterward. A glass red filter blocks blue before exposure; the mixer only redistributes data already recorded. Once a region clips to white or buries detail in shadow, no weighting recovers it — which is the case for working from raw, where each channel keeps its own headroom: blue tends to clip first under tungsten, red first in sunsets and skin.

The Film You’re Emulating, and the Noise You’re Not

A channel mixer is emulating a panchromatic emulsion plus a filter. Panchromatic films — FP4, HP5, Tri-X — are sensitive across the visible spectrum, which is why a red filter darkens their blue sky and lightens reds. Orthochromatic stock is a different animal: Ilford Ortho Plus 80 is effectively blind to red, rendering reds near-black, so a red filter on it is counter-productive. No two emulsions share an identical spectral curve, so “the red-filter look” is itself film-specific, and the mixer is only ever approximating one particular emulsion-plus-glass combination.

Two cautions on the digital side. The blue channel is usually the noisiest because the blue dye passes the fewest photons: lowest signal means photon shot noise, which scales as the square root of the signal, dominates, and white-balance gain under warm light amplifies it further. With modern CMOS read noise around 2 e⁻ per pixel, most daylight exposures are shot-noise limited, so pushing the blue weight hard surfaces grain you cannot meter away. And the “free, reversible” nature of the mix holds only while the colour data survives — in a raw file or a layered document. Flatten to greyscale or bake a JPEG and the conversion is as irreversible as a developed negative.