Categorical Data: What It Is and How to Prompt It

Spotlight search

Find any model, tool or page and jump to it.

Categorical Data

  • qualitative palette
  • series colours
  • chart categories

Six unrelated hues held at one depth, for six things that have no order.

Use this palette

Opens an image model with the palette in the prompt. The same words work on a video model.

  • #1B9E77
  • #D95F02
  • #7570B3
  • #E7298A
  • #66A61E
  • #E6AB02
Categorical Data used in one room
The same room, drawn to this palette. One colour carries the frame, one takes a third, one is the accent.

What it is for

A categorical palette carries one promise: these things are different and none of them is bigger. Any drift in lightness breaks that promise, because a darker colour reads as more important before the reader has looked at the numbers.

It is also the palette most often chosen badly. A picked-by-eye set usually has one colour far brighter than the rest, and that category wins an argument it was never in.

How it works

Take a published qualitative set rather than picking six hues yourself. This one is Dark2 from ColorBrewer, drawn to hold six categories apart at one depth.

Hold the depth even. The hues are the only thing that should change across the set, so check the whole strip in grey and adjust anything that stands out.

Stop at six. Seven is where a reader starts checking the key for every mark, and past that the colour has stopped carrying meaning at all.

When to use it

Use it for a bar chart, a pie, a map of regions, a set of tags, or anything where the categories are simply different from each other.

Leave it alone when the categories have an order, a size or a middle. A ranked set needs Sequential Data instead, and an even spread of hues will hide the ranking.

Told apart from its neighbours

Sequential Data changes lightness and holds the hue. This one holds the lightness and changes the hue. Swapping them is the most common chart colour mistake there is.

Colour Blind Safe overlaps but is stricter. ColorBrewer marks its qualitative sets colour blind safe only up to three or four classes, so check a six category chart against the Okabe and Ito set as well.

Where it was used

ColorBrewer was built by Cynthia Brewer for map makers and became the default source of chart palettes across research and software. Dark2 is its six to eight class qualitative set, which is why the same six hues turn up in dashboards that have never heard of it.

  • ColorBrewer
  • Dashboards
  • Transit maps

The prompt

Run it as it stands, or drop the colour clause into a prompt you already have. It works in an image prompt and in a video prompt.

<subject>, categorical palette of six distinct hues, green, orange, violet, magenta, olive and mustard, all held at a similar depth so no one of them looks more important, flat areas of solid colour with no gradient

Run it

What goes wrong

One colour comes back much brighter and takes over the chart. Say all held at a similar depth so no one of them looks more important.

The set arrives as a ramp of one hue. Name the six hues separately and say they are distinct rather than related.

Leave out

a gradient across the set, one colour far brighter than the rest, more than six hues

Watch for

Past about six categories nobody can match a colour back to the key. Group the rest into one Other.

Questions

About six. Past that a reader cannot match a mark back to the key, so group the remaining categories into a single Other in a neutral grey.

Paint it

Pay for the one run. No subscription, and the price is on the model page before you sign in.

Use this palette

Next to it in interfaces and data