Color mixer that shows all three answers

Because interpolating channel values is not what paint does. Halfway between #0000ff and #ffff00 in sRGB is #808080, chroma 0 — a perfectly neutral gray, since each channel is averaging its own opposite. Linear light gives the same neutral, brighter, at #bcbcbc, and Lab gives #ca8aaa, a dusty pink. Green comes from pigments subtracting wavelengths from reflected light, a process no interpolation between two RGB triples models; predicting it properly needs Kubelka-Munk theory and spectral reflectance data for the actual paints.

Use sRGB when the blend has to match what a browser will do, linear light when the two colors represent light rather than pigment, and Lab when a human is going to judge the result. The first is what color-mix(in srgb) and every canvas globalAlpha composite compute; the second is correct for glows, additive overlays and anything that models photons; the third is the one that keeps a mid-blend from looking darker than both of its parents.

Because sRGB numbers are gamma-encoded, so their average is not the average of the light they stand for. Red and green at 50/50 land on #808000 at L* 51.7, while the two parents average L* 70.5 — the blend is 19 points of perceived lightness darker than either of the colors that produced it. In linear light the same mix is #bcbc00 at L* 73.8, and Lab puts it at L* 70.5, exactly the average of the endpoints, because holding L* steady is what the Lab path is for.

The sRGB card is identical to color-mix(in srgb, …) and the page prints that line for you to copy. CSS also accepts other interpolation spaces, and swapping srgb for oklab in the copied line moves the browser's result close to the Lab card without matching it — Oklab and CIE Lab are different spaces with different lightness scales, and the browser computes its own. The function is available from Chrome 111, Safari 16.2 and Firefox 113.

Not in sRGB or linear light, where the fold is a weighted average and arithmetic does not care about order. In Lab it can shift slightly: each intermediate result is brought back inside the sRGB gamut before the next color folds in, so a chain that passes through an out-of-gamut midpoint loses a little chroma on the way. With ordinary screen colors the difference stays under a unit of ΔE.

Alpha interpolates alongside the channels, so mixing #ff000080 with an opaque color gives a result that is partly transparent, and the hex output grows to eight digits to say so. Note that a translucent result has no single appearance until it sits on a background — the swatch here is drawn over the page, so judge it against the surface you will actually use.

Because averaging two hues that sit far apart on the wheel pulls the result toward the neutral axis, and the further apart they are the harder it pulls. Blending #2563eb with #f59e0b at 50/50 in sRGB produces #8d817b, whose chroma is 6.0 against 80.1 and 78.8 for the two parents — technically a color, visually a warm gray. Mixing in Lab keeps more of it, at chroma 27.2, but nothing recovers saturation that the average genuinely does not contain.