How to create a LUT from a reference image

You've found an image whose colours you love — a still, a photo, a frame from your own footage — and you want that look on your material as a reusable LUT. Here is what's actually involved, and the trade-offs of each route.

What "matching a reference" means

A look lives in three measurable places: the tone curve (how shadows, midtones and highlights are distributed), the white balance and colour cast (does neutral grey lean warm, cool, green?), and hue and saturation shifts (are greens pushed toward teal? are highlights golden but shadows blue?). Matching a reference means measuring those properties in the reference and building a transform that imposes them on your image.

Route 1: grade by eye, then export

The traditional way: put the reference beside your footage in a colour tool, adjust wheels and curves until they match, then export the correction as a LUT. It gives full artistic control and it's the right approach when the reference is only inspiration. The costs: it needs skill, it takes time per look, and "matched by eye" drifts with your monitor and your patience.

Route 2: extract a look from the reference alone

Several tools build a LUT from just the reference image — sampling its palette and creating a transform that pulls any input toward those colours. This is quick, but the tool has never seen your image, so it cannot know how far (or in which direction) your material needs to move. Results vary from decent to muddy, and there is no way even in principle for the tool to measure its own accuracy.

Route 3: two-image statistical matching

The approach Tonemara takes: analyse both images and fit the transform between their colour distributions — tone curves per channel in a perceptual colour space, plus the chroma rotation and the luminance-dependent colour shifts that make up split-toned looks. Because both sides are measured, the fit is specific to your image, and its accuracy can be measured (and reported to you) as a colour-difference number.

What to check in any generated LUT

Banding: apply the LUT to a smooth gradient (sky, mist). Poorly built LUTs introduce visible steps. Look for monotonic, smoothed grids.
Clipping: highlights and shadows should compress gracefully, not slam to pure white/black.
Size: 33×33×33 is the sweet spot most editors expect; smaller grids (17³) interpolate more coarsely.
Format: the .cube text format with sRGB in/out is the most widely supported for display-referred work.

Applying a LUT built from sRGB stills onto log footage will look wrong — convert to a display-referred space first, or apply the LUT after your normalisation node.

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.