Transfer Function Explorer

Three different curves get conflated as "gamma." The OETF encodes scene light into signal at the camera; the EOTF decodes signal into display light; their end-to-end combination, the OOTF, is deliberately not unity. SDR rendering applies roughly a 1.2 system gamma to compensate for dim-surround viewing. Everything plotted here is an EOTF: signal in, cd/m² out.

Video encoding was built around the CRT's native response, close to L = Lw·V^2.4, a happy accident, since it also approximates perceptual uniformity. Pure power curves have infinite slope at zero, so sRGB's EOTF splices in a linear toe below V ≈ 0.04. The common claim "sRGB is gamma 2.2" is an approximation: the piecewise curve tracks darker through the shadows than pure 2.2. Drag the γ slider against the sRGB curve to see where they diverge.

The reference EOTF for HD. Not simply "gamma 2.4": L = a·(V + b)^2.4, where a and b are solved from the display's measured white Lw and black Lb. A nonzero black level lifts the entire shadow region; as Lb → 0 (OLED), b → 0 and the curve converges to pure 2.4. Drag the BLACK LEVEL slider and watch the shadows detach from the power curve: that is why the same grade reads differently on an LCD and an OLED, and why calibration measures both ends of the curve.

The HDR EOTF, and it is absolute: a code value means one exact luminance regardless of the display. 50% signal ≈ 92 cd/m², 75% ≈ 1,000 cd/m². Notice the virtual-display sliders never move it. The curve was fit to the Barten contrast-sensitivity model, so 12 bits span 0.0001 to 10,000 cd/m² with quantization held below the visibility threshold. When a display's peak is below the mastering peak, tone mapping decides what happens to the top of the curve: that negotiation is what HDR10 static metadata and Dolby Vision dynamic metadata exist for.

ARIB STD-B67, adopted into BT.2100 for broadcast. Scene-referred: the signal encodes relative scene light (square-root below 50%, logarithmic above) and the display applies a system gamma that depends on its own peak: γ = 1.2 + 0.42·log₁₀(Lw/1000). Drag the PEAK slider: HLG rescales with it while PQ stays fixed. That is the fundamental split. HLG adapts to the display and the room, PQ dictates to them. No metadata required, and it degrades gracefully on SDR screens, which is why live broadcast chose it.

Contrast ratio is Lw/Lb, and the black term dominates perception: halving the black level does more for perceived contrast than doubling the peak. The classic "brightness" and "contrast" controls moved black offset and signal gain, reshaping exactly the curves plotted above. The virtual-display sliders reproduce that: peak sets where curves top out, black level lifts their floor and re-solves BT.1886. Whether a real panel actually tracks its target curve is a measurement question: probe-verified grayscale tracking, before and after. Book a calibration →