Build a colour transform once and carry it everywhere. A .cube file describes what a grade does rather than what it did to one photo, so the same file works on a video timeline, a batch of stills, or a live stream.
What a LUT actually stores
A 3D LUT is a cube of input colours with an output colour recorded at each grid point. A 33-point cube holds 33³, or 35,937, of them. Software finds where a pixel’s RGB value sits inside that cube and interpolates between the eight nearest entries.
That structure explains both the strengths and the limits. Any per-pixel colour operation can be baked into it, because the table only has to answer “this colour in, that colour out”. Anything that depends on where a pixel sits in the frame cannot be: vignettes, blurs, grain, and sharpening have no place in a LUT, and no LUT format can carry them.
The preview shows the table, not the sliders
The image on this page is graded by applying the exported table, not by running the slider math directly. The distinction matters at small cube sizes. Set the size to 17 and push contrast hard, and the banding you see on a smooth sky is the banding the file will produce in Resolve. Raising the size to 33 or 65 removes it here and there in the same way.
The preview uses the same tetrahedral interpolation as Apply LUT to Image, so a file exported here and loaded back there renders identically.
Order of operations
The controls run in a fixed order, which is why two grades with the same slider values in a different tool can look different:
- Exposure and white balance, in linear light. One stop is a doubling of the linear value, so applying exposure to the encoded sRGB number instead would give the wrong answer.
- Lift, gamma, and gain, on the encoded signal. Lift moves the black point and leaves white alone, gamma bends the midtones with both ends pinned, gain scales everything so highlights travel furthest.
- Contrast, pivoting around middle grey.
- Saturation, against Rec. 709 luminance.
- Shadow and highlight tint, weighted by luminance with a soft crossover.
Each tint has its own grey removed before it is applied, so tinting shadows blue shifts their colour without also darkening them.
Choosing a size
17 produces a file of a few hundred kilobytes and is fine for gentle looks. 33 is the working default across grading software. 65 is worth its roughly 7 MB only when the transform turns sharply somewhere in the cube, such as a hard channel clip or an extreme split tone.
Larger is not automatically better. A big cube storing a smooth transform is a large file describing something a small one already handled.
Limits
A LUT clamps to the 0 to 1 domain written in its header, so anything the grade pushes past white or below black is lost at export rather than in the host application. Grades built to be recovered later belong in the host’s own tools, not in a baked table.
The output is also display-referred. Applying it to log or raw footage that has not been converted to a display space first will not give the look shown here, because the input colours land somewhere else entirely in the cube.