Color science foundations

ExpertDuration ~35 min video + readingTools DaVinci Resolve (free)

Why this lesson

Everything you have done so far — normalize, balance, match, look — has worked because a color-managed pipeline was quietly doing the science underneath. Level 3 is where you learn to reason from the science, so that when a pipeline misbehaves you can debug it in the right language instead of guessing. This first lesson builds the mental picture the whole level stands on: how color is plotted, how a gamut and a white point live on that plot, the handful of color models you will keep meeting, and the single most important distinction in modern color — display-referred vs scene-referred. No video on the open internet teaches the CIE diagram properly, so the explainer carries it and points you at the two best free readings.

Watch

Watch for: Kelly builds intuition in Fusion rather than lecturing. Watch the 3D RGB cube demo: a uniform cube 'doesn't represent anything' on its own — it's useful only as a reference you compare a manipulation against, and when he swaps the uniform cube for the actual pixels of an image you see a color model made concrete. This is the accessible on-ramp; the CIE diagram and the referred-domains distinction below are ours to carry, because he's teasing a paid masterclass and never gets to them.

The explainer

Start with the map. To reason about color you need a way to draw it. The CIE chromaticity diagram is that map: a 2D plot where every visible color’s hue and purity land at a point, with brightness factored out. Its outer boundary — the horseshoe curve — is the spectral locus: every pure, single-wavelength color the eye can see, violet at one tip, sweeping through blue-green and green, round to red at the other. The straight line closing the bottom is the line of purples (magentas, which have no single wavelength). Everything the human eye can perceive sits inside that shape; nothing sits outside it. The original 1931 version uses x, y coordinates; the 1976 revision uses u’, v’, spaced to be more perceptually uniform, so equal distances on the plot look like roughly equal color differences. When FilmLight or a colorist shows you a “horseshoe,” that’s this.

A gamut is a triangle on that map. Any RGB display has three primaries — a red, a green, a blue — and each is a point on the diagram. Join them and you get a triangle: the display’s gamut. Every color inside the triangle is reproducible by mixing those primaries; every color outside is not. Rec.709 is a smallish triangle, DCI-P3 a bigger one, Rec.2020 bigger still — but even Rec.2020 doesn’t fill the whole horseshoe, which is why “wider gamut” is always relative. This one picture explains gamut mapping (Lesson 3.2): it’s the problem of what to do with a color that falls outside the triangle you have to deliver into.

The white point is a point inside the triangle. A gamut says nothing about which color counts as neutral — that’s the white point, and the standard one is D65, roughly 6500 K daylight, at x≈0.3127, y≈0.3290. Rec.709, sRGB and Rec.2020 all share D65, which is why neutral grey stays neutral when you convert between them. Digital cinema uses a different, greener white (roughly DCI white), one reason cinema masters can feel subtly different.

Color models are just coordinate systems. RGB (the cube in the video) is device-referred — great for driving a screen, terrible for reasoning about appearance, because equal steps in RGB don’t look equal. So the science uses others. CIE XYZ is the device-independent master space every other space converts through. CIELAB and IPT are perceptual, opponent models: they split color into a lightness axis and two color axes (roughly red–green and yellow–blue), matching how the eye actually encodes color after the cones. IPT (and its cousin ICtCp) has more stable hue lines, which is why HDR standards lean on it. You don’t compute these by hand — you just need to know that “a perceptual space” means one where a straight push holds its hue instead of drifting.

Now the big one: display- vs scene-referred. In a display-referred encoding, a code value means an amount of light out of the display — 100% white is “as bright as this monitor’s white,” and the numbers are bounded 0–100%. Rec.709 is display-referred; it bakes the whole camera-to-eye rendering into the signal. In a scene-referred encoding, a value is proportional to the light that was actually in the scene — linear, unbounded, with no built-in idea of “white.” A scene-referred value of 18% is middle grey as the camera saw it; a specular highlight might be 800%. A scene-referred image is not yet a picture — it needs a rendering transform (the DRT of Lesson 3.2) to squeeze that open-ended range down to what a display can show. This is why ACES protects highlights as you push (you’re grading the scene, not the screen) and why “grade scene-referred, deliver display-referred” is the shape of every modern pipeline.

Do it

1. Open Resolve’s CIE scope (Color page → Scopes → CIE). Put a colorful shot up and watch the trace sit inside the horseshoe. Overlay the Rec.709 and P3 gamut triangles (scope settings) and find pixels that fall outside Rec.709 but inside P3.
2. Push saturation hard on a saturated red and watch the trace crawl toward the edge of the triangle — that’s a gamut excursion, live.
3. Switch a clip’s color science between display-referred (plain Rec.709) and a scene-referred managed setup (DaVinci Wide Gamut / Intermediate). Push exposure +2 stops in each. Note how the display-referred version clips and the scene-referred version rolls off — the referred domain, felt in your hands.
4. Find D65 on the CIE scope (drop the white point) and confirm neutral greys land on it.

Level 3 workbook — every Do it exercise, 3.1–3.10, plus the capstone (printable)level-3-workbook.pdf764 KBOriginal course material — free to use

Terms introduced

Terms introduced in this lesson

• CIE chromaticity (xy / u′v′)
• Spectral locus
• D65 white point
• Color model (Lab / IPT)
• Display-referred

Each term links to the glossary, where it is defined by the way you just used it.

Check

Check yourself

1. What does the horseshoe-shaped outer curve of the CIE chromaticity diagram represent?

   The range of colors film can printThe boundary of sRGBThe gamut of a Rec.709 displayThe spectral locus — every pure single-wavelength color the eye can see, from violet through green to red

2. A gamut is drawn on the chromaticity diagram as…

   A single pointThe whole horseshoeA circle centred on the white pointA triangle whose corners are the three display primaries — everything inside it is reproducible, everything outside is not

3. What is D65?

   A camera log formatA specific white point — roughly 6500 K daylight, at chromaticity x≈0.3127, y≈0.3290 — that Rec.709, sRGB and Rec.2020 all shareA codecA brand of reference monitor

4. What is the core difference between display-referred and scene-referred encoding?

   In display-referred, a code value means "this much light out of the display"; in scene-referred, a value is proportional to the light that was in the scene, unbounded, and needs a rendering transform to become a pictureScene-referred only works in HDRThey are two words for the same thingDisplay-referred is always 8-bit; scene-referred is always 16-bit

Check answers

You can move on when you can… read a chromaticity diagram out loud — point to the spectral locus, a gamut triangle, and D65 — name what CIELAB or IPT buys you over RGB, and explain in one clean sentence the difference between a display-referred and a scene-referred image.

Mark this lesson complete

Go deeper

FilmLight — sRGB… We Need To Talk (first half): the single best free demonstration that a color standard has two halves that don’t agree — sRGB’s display is defined as a pure 2.2 power, but its encoding is a piecewise compound curve, and they mismatch in the shadows. This is display-referred color science shown from first principles, and it sets up Lesson 3.2.

Segment: 0:00–24:59 — the sRGB EOTF-vs-encoding mismatch, from first principleswatch full video

Watch for: Daniele reads the actual IEC standard: section 4 defines a pure 2.2-power display, section 5 defines a compound encoding curve with a linear toe — and the standard itself admits the mismatch is deliberate. Watch him plot the encode/decode error and show it's real and it lives in the shadows.

FilmLight Colour Masterclasses — theory stretch: Daniele Siragusano on why “there is no such thing as no look,” the physical construction of a print-emulation curve from densitometer readings, the brilliance boundary between surface and glowing colors, and the opponent-color rationale behind an IPT-like space. Dense but foundational.

Segment: 8:22–39:10 — the history of the look + color-science theorywatch full video

Watch for: The move from video-look to film-look to digital, then the deeper science: skin-tone homogeneity, the brilliance boundary, S-curves as natural-system behavior, and why an opponent color space is the right home for a look.

Read

The CIE diagram and the referred domains are best learned in prose. These are the free, authoritative sources:

• Charles Poynton — Color FAQ & Gamma FAQ (poynton.ca): the canonical plain-language treatment of chromaticity, white points, gamma and the difference between light and code value. Read the Color FAQ first.
• Chris Brejon — CG Cinematography, ch.1 “Color Management” (chrisbrejon.com): the best free long-form explainer of scene-referred vs display-referred, written for people who actually push pixels.
• ACESCentral (acescentral.com): community docs and webinars on the scene-referred model in practice.

The paid path: rigorous CIE-and-appearance-model teaching is the moat of Poynton’s fxphd course and Cullen Kelly’s own creative color-science masterclass (the one he teases in the video). When you’re earning and want the doctorate-adjacent depth, those are where it lives — this lesson is deliberately the free floor under them.

Next up: 3.2 · Tone & gamut mapping, DRTs — what actually turns a scene-referred image into a picture, and how to debug it.