Looks and rebuilds: two meanings of filter

People say filter for two different jobs, and both are useful. A look keeps the photograph and changes its colour; a rebuild remakes it out of something else. This lab does both and labels them — here is the test that tells them apart, and fifteen cases to try it on.

The test, in one question

On this site the first kind is called a look (and the colour finishes, like black and white or sepia, are looks too); the second is a rebuild. Below, “adjustment” means a look.

Ask whether the operation changed what the picture is made of, or only what colour its parts are.

A photograph arrives as a grid of pixels, each one standing for a small patch of the world. An adjustment leaves that arrangement exactly as it found it. Every pixel still corresponds to the same patch, the grid is the same size and shape, and the operation can be described completely as a rule for turning one colour into another. Warmer. Darker. Flatter. More contrast in the midtones. The photograph is still a photograph of the same thing at the same resolution; it is wearing different clothes.

A rebuild throws that arrangement away. It measures regions of the picture, decides what each region is worth, and draws something else in its place — a dot whose area stands for a brightness, a block of one flat colour, a character with roughly the right weight of ink, a palette entry plus a pattern that makes up the difference. Afterwards the picture is made of dots, or blocks, or letters, and the pixels that used to be there are not recoverable in any meaningful sense.

That is the whole distinction, and it survives every awkward case below. One is a rule about colour; the other is a change of material.

Fifteen cases, and where each one lands

The test is only worth having if it gives an answer where intuition does not. These are the ones people argue about.

Applying the question to fifteen common operations
Operation	Did the material change?	Side
Warming the white balance	No — a rule about colour	Adjustment
Lifting the shadows	No — a rule about tone	Adjustment
Converting to greyscale	No — three numbers become one	Adjustment
Sepia toning	No — one colour rule per pixel	Adjustment
Adding film grain	No — the picture is intact underneath	Adjustment
A strong vignette	No — a position-dependent tone rule	Adjustment
Blurring the whole frame	No — pixels are still pixels	Adjustment
Posterising to five levels	Yes — tone becomes a step count	Effect
Duotone	Borderline — a colour rule with no way back	Adjustment, just
Dithering to one bit	Yes — the picture becomes a pattern	Effect
Halftone screening	Yes — the picture becomes marks	Effect
Pixelating into blocks	Yes — the unit is now a block	Effect
ASCII conversion	Yes — the unit is now a character	Effect
Mosaic and tile rendering	Yes — the unit is now a tile	Effect
Edge detection	Yes — the output is a different quantity	Effect

The two rows that are genuinely difficult

Posterising looks like an adjustment. It is a per-pixel rule, it touches no neighbours, and you can write it on one line. But what comes out is not a description of the same tonal range at lower fidelity — it is a picture made of a countable number of flat regions, with hard boundaries the photograph never had. The unit stopped being a continuous tone and became a step. That is a change of material, arrived at by a per-pixel route.

Duotone goes the other way. It discards more information than posterising does — all of the hue, permanently — and it still lands on the adjustment side, because what comes out is a continuously-toned photograph of the same scene at the same resolution. Nothing was measured and replaced. The test is about structure, not about how much was thrown away, and this is the case that proves it.

This site builds duotone anyway, and that is an honest inconsistency rather than a hidden one. It sits here because it is the natural first step before a two-colour dither or a two-colour screen, and because sending someone to another page for the setup step of the thing they are already doing would be tidy and useless.

Why the vocabulary drifted

The word arrived from photography, where a filter was a piece of glass in front of the lens — and a piece of glass can only change which light gets through. Every optical filter is on the adjustment side of this line by construction, because glass cannot rearrange an image into dots. So for a century the word meant precisely one of the two things, and there was no ambiguity to have.

Software borrowed the word for the whole menu, because the menu was called Filters and everything on it had to be a filter. Halftone screening had existed as a printing process since the 1880s and had never been called a filter by anyone; it went on the menu next to the warming filter and inherited the name. The ambiguity is an accident of user-interface labelling that then had forty years to set.

The practical residue is that half the people searching for one of these two things are using a word that mostly returns the other. Naming the operation — halftone, dither, posterise, ASCII — is the only reliable way through, which is why this site names them on every page, and why its strip keeps the rebuilds under a tab of their own instead of mixing them in with the colour looks.

What to do with the answer

If the test says adjustment, you want controls that are about colour and tone: curves, temperature, contrast, a ramp between two inks. Reach for a tool whose entire interface is about deciding what colour something should be, and expect it to give you no say whatever over structure.

If the test says effect, you want controls that are about units: how big a cell is, how many colours are in the palette, what shape the mark is, how many columns the grid gets. That is what this lab’s Rebuild tab puts in front of you. Its tone controls act on the finished render, after the marks are made, and never before — adjusting the tone of a picture that is about to become four hundred dots is a decision that gets averaged away in the first pass.

And if what you actually want is both — a picture toned one way and then rebuilt — do them in that order and keep the intermediate file: save the colour look, open that file again, and pick the rebuild. Rebuilds are the ones you cannot walk back.

Questions the distinction raises

Is black-and-white conversion an effect or an adjustment?

An adjustment, by this test. Converting to greyscale replaces three numbers per pixel with one, but the pixel grid is untouched, every pixel still stands for the same patch of the world, and the operation is described entirely by how much weight each channel gets. Nothing has been rebuilt. It sits at the far edge of the adjustment side — it discards more than most — but it is on that side.

What about adding grain or noise?

Adjustment, and it is the case that makes people argue. Grain adds something that was not there, which feels like a rebuild, but look at what it does to the structure: every pixel still corresponds to the same patch, the grid is unchanged, and the picture underneath is intact beneath the noise. Nothing was measured and replaced. A dither, which also produces a scatter of dots, is on the other side — because there the dots ARE the picture rather than an addition to it.

Does the distinction matter, or is it a way of organising a website?

It decides which tool will be able to help you. An adjustment tool gives you exact control over colour and tone and no control at all over structure; an effect tool gives you cell sizes, palettes and grids and very little say over the tonal result. Asking either one to do the other's job produces a fight rather than a picture. The distinction is not a taxonomy for its own sake — it is a way of telling which set of controls you actually need in front of you.

Can an operation be both?

A pipeline can contain both, and most interesting results do. A duotone is an adjustment-shaped operation — it maps tone onto colour — and dithering the result is a rebuild. Running one after the other is normal. What cannot happen is a single operation sitting on both sides at once: either the units the picture is made of changed, or they did not.

Back to the lab

Rendered on your computer, not on a server.