How 3D LUTs actually work

A 3D LUT (look-up table) is the simplest powerful idea in colour: instead of describing a colour transform with formulas, just write down the answer for a grid of input colours and interpolate between them.

The lattice

Imagine RGB colour as a cube: red on one axis, green on another, blue on the third. A 33×33×33 LUT places a lattice of 35,937 sample points through that cube, and for each point stores the output colour the transform should produce. That's the entire file — 35,937 lines of three numbers.

Interpolation does the rest

Your actual pixel almost never lands exactly on a lattice point, so the editor finds the eight surrounding points and blends them — trilinear interpolation (fancier engines use tetrahedral). This is why LUT size matters: with a 17³ grid the points are twice as far apart, and sharp curve changes get smoothed over or, worse, band. 33³ is the standard compromise between fidelity and file size, and what most editors expect.

The .cube format

The most portable container is a plain text file: a TITLE, a LUT_3D_SIZE, a domain declaration (usually 0–1), then the lattice values with red varying fastest. Because it's display-referred (it maps display colours to display colours), the same file works across video editors, compositors, encoders, and even streaming tools.

Two properties that separate good LUTs from bad ones

Monotonicity: along any axis, output values should never reverse direction. A non-monotonic LUT means two different exposures of the same thing can swap brightness order — visible as contouring and posterisation in gradients.
Smoothness: neighbouring lattice points should change gradually. Overfitted LUTs (built from noisy data without regularisation) produce speckled, brittle results on images they weren't built from.

A LUT is "dumb" on purpose: it can represent any global colour transform — film emulation, a camera correction, a matched reference look — because it never needs to understand the transform, only to record it.

What LUTs can't do

A LUT maps colour to colour, one pixel at a time. It cannot sharpen, denoise, vignette, or treat the sky differently from the ground. If a look depends on where something is in the frame, that part needs masks — the LUT carries everything global.

Tonemara does all of this in your browser in one click: drop a reference image and your own photo, preview the match free, and export a 33×33×33 .cube LUT plus a Lightroom-compatible .xmp preset. Your images never leave your device.