the light lab

① The Mona Lisa under a sodium lamp

Every color in the painting gets a plausible reflectance spectrum (Jakob–Hanika, Newton-solved by reflectanceOf), then is re-integrated under the illuminant you choose — tungsten, fluorescent, candlelight, or the near-monochromatic sodium doublet, under which color appearance collapses to a single hue. "Eyes adjusted" applies Bradford chromatic adaptation to the new white.

Honesty note: these are metamer spectra — smooth reflectances that integrate to exactly the painting's sRGB colors under D65, not Leonardo's actual pigments. Under smooth illuminants the difference is small; under spiky ones (F11) it is a live demonstration of metamerism. Under sodium it vanishes almost entirely: with one wavelength to reflect, the spectrum's shape stops mattering. Exposure is normalized — what changes is chromatic content.

② Thirty meters down

Sunlight filtered through pure water using Pope & Fry's integrating-cavity absorption measurements (WATER_ABSORPTION, 1/m) and Beer–Lambert attenuation (attenuate). Red light is effectively gone by 10 m — watch the reds die first, then orange, then yellow.

Model: downwelling D65 attenuated through depth d of pure water (Pope & Fry 1997). Real seawater adds scattering and dissolved matter; pure-water absorption is the physical floor of color loss.

③ Nightfall — and noon on Pluto

Below ~5 cd/m² your cones begin to hand over to rods, whose spectral sensitivity V′(λ) peaks blue-shifted at 507 nm and carries no color. This dial applies the CIE 191:2010 mesopic system (mesopic): reds darken and die first (the Purkinje shift), blues hold on longest, and at starlight the painting is rod-vision gray. NASA's "Pluto time": noon on Pluto is only about as dim as Earth a few minutes after sunset — still nearly full color.

Per palette color: photopic luminance from the CMF ȳ, scotopic from CIE 1951 V′(λ) (K′m = 1700.06 lm/W), blend weight m from the CIE 191 implicit equation. The rod image's grays are ordered by scotopic luminance — blue fabric outshines red lips at night, exactly as Purkinje observed in 1819. Exposure-adapted view.

④ Watercolor, outdoors, in weather

The canvas stores pigment concentrations, not colors. Strokes mix by Kubelka–Munk (K/S = (1−R)²/2R, Duncan's law) — blue over yellow makes green, like real paint and unlike RGB. Then the whole painting is re-integrated under the weather you pick: the light changes, the paint doesn't.

Paint spectra are Jakob–Hanika reflectances of eight classic watercolor hues — plausible metamers with real K-M mixing, honestly labeled (measured pigment K/S curves would slot straight in). Wash behavior — deposit saturating on wet paper, pigment migrating to stroke edges, paper granulation — is stylized presentation; the color math underneath is untouched. Painter's tip the physics reproduces: cadmium red + yellow makes the vivid orange; alizarin (a bluish red) mixes it duller — and thin washes glow where heavy ones go dark. The "blue-sky shade" weather is the Hosek–Wilkie zenith spectrum from section ⑤.

⑤ The sky, solved

The Hosek–Wilkie spectral sky-dome model (whitepoint/sky) — the SIGGRAPH 2012 fit to brute-force atmospheric path tracing, ported exactly, coefficients verbatim. A fisheye view of the whole sky: drag the sun down and watch the dome warm; raise the turbidity and watch the blue wash out.

⑥ Sunlight on other worlds

The same sun, filtered through different air. Direct sunlight as a 5772 K Planckian (planckianSPD — Planck's law, no fit) through a parameterized Beer–Lambert atmosphere: Rayleigh ∝ λ⁻⁴ plus a blue-absorbing dust term.

Honesty note: this is parameterized extinction, not radiative transfer — it shows why Mars light is butterscotch and Venus light is dim orange, with real spectra and real integration, but the τ/δ values are illustrative. The Earth dome in section ⑤ is the rigorously fitted model; an equivalent fit for other atmospheres would slot into the same API. Pluto needs no atmosphere model at all — see section ③.