What the file actually contains
A lookup table is exactly what the name says: a list of input colours and the output colour each one becomes. A 3D table samples the colour cube on a grid, commonly 33 points per axis, and everything between grid points is interpolated. There is no logic in the file and no awareness of the image. Feed it a colour, get a colour.
That mechanical simplicity is both the appeal and the limitation. The transform is perfectly repeatable across shots, across machines and across software, which is why the format became the standard way to move a look between departments. It is also completely blind to what a given shot needed, which is why the same file that transforms one clip beautifully wrecks the next.
Technical and creative uses
The two jobs are worth separating in your own head, because they have different standards of correctness.
- Technical transforms. Converting camera log to a display standard, or one display standard to another. There is a right answer, published by the camera manufacturer, and using an approximation instead is a real error.
- Camera matching. Bringing two different cameras to a common baseline so they can be graded together.
- Creative looks. Somebody's grade, packaged. No correct answer, and no guarantee it suits your material.
- On-set monitoring. Showing the crew a rough version of the intended look while shooting flat, which is what makes log footage viewable on the day.
Most frustration with these files comes from a creative look that silently includes a technical conversion, so it only works on footage from one specific camera in one specific colour space.
Why the input matters more than the file
A lookup table cannot see. It assumes footage arrives with black at a particular level, white at another, and hues distributed as the author expected. Every deviation from that assumption becomes a distortion on the way out.
The consequences are predictable. Footage exposed brighter than the LUT expects pushes highlights past the top of the table and clips them, with no recovery available afterwards. Footage in the wrong log curve produces skewed hues, usually visible first in skin. And a look built for a wide-gamut source applied to something already in a display space tends to oversaturate everything.
The remedy is the order of operations. Balance the shot to neutral first, then apply the transform, then trim exposure and saturation on top of it. Working this way means the file is doing the job it can do, and the judgement stays with you.
Using them on generated footage
Generated clips are the case where these files earn their place for a reason that has nothing to do with style. Because each generation samples independently, a sequence assembled from several clips has no shared response curve. Matching them by hand and then running one shared transform across all of them imposes a single coherent curve, which is much of what makes the sequence read as one scene.
Two cautions. Keep the strength modest, since generated images carry less latitude than camera footage and will band or posterise where real footage would hold. And do the matching before, not after: applying a strong look to clips that still disagree preserves the disagreement underneath and wastes the one advantage a fixed transform gives you.