RGB, sRGB, Adobe RGB, ProPhoto, Display P3, and Rec.2020

An RGB triplet is meaningful only inside a color space, transfer function, white point, and encoding depth. This lab lets you push one color through those layers at once: decode to linear, convert through XYZ, optionally adapt white, quantize the result, and inspect clipping or banding risk before the image ever reaches a screen.

Three things are routinely conflated under the label "RGB." All of them matter.

The triple (255, 0, 0) is meaningless without all three layers specified. It could mean "the most saturated red sRGB can produce" (one chromaticity), or "the most saturated red Display P3 can produce" (a substantially redder chromaticity), or even an out-of-range value in a 16-bit half-float linear image. A well-managed file tags its color space; an unmanaged file is silently assumed to be sRGB.

Almost every digital image you touch lives in one of these spaces. They are listed roughly by gamut size, smallest first.

sRGB was defined by HP and Microsoft in 1996 and standardized as IEC 61966-2-1 in 1999. The goal was simple: a color space whose primaries matched typical CRT monitors and whose transfer function matched the rough 2.2-gamma response of those CRTs. With sRGB in place, untagged image files could be assumed to mean something specific, and the web could render color consistently.

Adobe RGB (1998) widens the green primary substantially while keeping sRGB's red and blue. The wider green captures saturated cyans and greens that occur in nature (foliage, water, cyan inks) and cannot be encoded in sRGB. For photographers whose output is print, where CMYK gamut overlaps these green-cyans, Adobe RGB preserves information that sRGB would clip.

Adobe RGB and sRGB share the D65 white point and use a similar gamma (~2.2). Conversion between them is a simple matrix transform. But the larger gamut spreads the same 8-bit code values over a wider color volume, so 8-bit Adobe RGB risks more visible banding than 8-bit sRGB. Most Adobe RGB workflows use 16-bit files.

ProPhoto RGB (also called ROMM RGB) was designed by Kodak for archival photographic imaging. It uses imaginary primaries that lie outside the visible spectral locus - the same trick CIE used for XYZ. Its gamut covers about 91% of human chromaticities, including colors no device can reproduce but that may need to be preserved through the editing pipeline.

The catch: such a large gamut spreads the same numerical range over a much wider color volume. 8-bit ProPhoto bands aggressively, so the format effectively requires 16-bit encoding. ProPhoto also uses D50 (not D65) and gamma 1.8 (not 2.2), marking it as a print-oriented intermediate space rather than a display space.

Display P3 took the DCI-P3 cinema primaries (wider than sRGB, especially in reds and greens) and recombined them with the D65 white point and sRGB-style transfer function. The result is a consumer-friendly wide-gamut space that maps cleanly onto modern display hardware. Apple pushed it heavily from 2015 onwards; Android, Chrome, Firefox, and Safari now all support it.

CSS Color Module Level 4 made Display P3 a first-class web color space: color(display-p3 1 0 0) renders the wide-gamut red on capable displays and falls back gracefully on sRGB displays. For most photo content shipping to consumers in 2026, Display P3 is the working space of choice.

Rec.709 (ITU-R BT.709, 1990) is the HDTV equivalent of sRGB. Its primaries are essentially identical to sRGB - both inherit the same CRT-era phosphors - but its transfer function (BT.1886) and YCbCr encoding differ. Practical color difference between sRGB and Rec.709 is small.

Rec.2020 (ITU-R BT.2020, 2012) is its UHD/HDR replacement, with primaries placed on the spectral locus - the widest broadcast space possible. Few consumer displays achieve full Rec.2020 coverage, so HDR mastering typically happens inside a P3-D65 sub-volume within a Rec.2020 container. The container's BT.2100 HDR companion spec adds PQ (perceptual quantizer) and HLG (hybrid log-gamma) transfer functions designed for high dynamic range encoding.

Every RGB space has two forms: a linear form where pixel values are proportional to scene or display luminance, and an encoded form where values have been bent through a transfer function. Files almost always store encoded RGB - it makes 8-bit encoding visually adequate by allocating more code values to dark tones, where the eye is most sensitive.

Linear RGB is what you need for any physically correct operation: alpha compositing, blurring, color mixing, lighting calculations, gamma-correct rendering. Convert to linear, compute, then convert back to encoded for display. Skipping linearization is the single most common source of visible color bugs in graphics code.

Three white points dominate the working spaces. D65 is used by sRGB, Adobe RGB, Display P3, Rec.709, Rec.2020 - essentially everything aimed at screens. D50 is used by ProPhoto, ICC PCS, and almost all print workflows. DCI (slightly green-shifted) is used by theatrical cinema. Mixing white points across a workflow without explicit chromatic adaptation produces silent color shifts.

A color space and a bit depth must be chosen together. The same 8-bit channel allocates 256 levels regardless of the underlying space - so wider gamuts spread the same code values across a larger color volume. The result: 8-bit ProPhoto bands aggressively where 8-bit sRGB looks smooth. Higher gamuts demand higher bit depth.

A correct RGB → RGB conversion has four steps:

Skipping the linearization step (Step 1) and applying matrix math directly to gamma-encoded values is the most common mistake. It looks plausible but produces muddied saturated colors and shifted whites. The chromatic adaptation step is the second most common omission.

The space you pick depends on what you're delivering, not what you're capturing. Capture in the widest practical space, edit in something even wider, and convert down for delivery.

Six questions on RGB spaces, primaries, and transfer functions. Wrong answers come with brief explanations.