Word Cloud Generator
Turn text into a word cloud where the most-used words appear largest, with stop-word filtering — and a note on why they mislead as often as they inform.
Five times the font is twenty-five times the ink
This page sets the font size to 14 + 56 × (count / max), so the commonest word is drawn at 70px and everything else scales linearly down to a 14px floor — 5× the height. You read words by their area, though, and area goes as the square of height: the same range is 25× the ink.
| Share of the top word's count | Font | Share of its ink |
|---|---|---|
| 0% | 14.0px | 4.0% |
| 6.25% | 17.5px | 6.3% |
| 10% | 19.6px | 7.8% |
| 25% | 28.0px | 16.0% |
| 50% | 42.0px | 36.0% |
| 75% | 56.0px | 64.0% |
| 100% | 70.0px | 100.0% |
A word appearing half as often as the leader gets 36% of its ink, not 50%. At the other end the floor does the reverse: a word that barely appears still gets 14px, which is 4% of the largest word's area for very nearly none of its frequency. Solving for where the two agree gives exactly 6.25% — one sixteenth. Below that share a word is drawn louder than it deserves; above it, quieter. The gap is widest around 53%, where a word gets 14 percentage points less ink than its frequency warrants. That is not a fault so much as a property. Every word cloud ever made has it, and it is why they are decorative rather than analytical — useful for noticing what is there, not for judging how much of it.
And four of the commonest words in English still get through
The stop list holds 52 words and catches 33 of the 39 commonest words in written English. Of the 6 it misses, two — I and do — are caught anyway by the rule that drops anything under 3 letters. That leaves 4: all, one, say, there. In any large body of English text those will outrank most of the words you were actually looking for, and they will be drawn large enough to dominate.
The length rule cuts the other way too, and it is not frequency-aware: nothing under 3 letters ever appears, so a document about AI, US, UK or the EU will not show those at any count at all. Three letters is the shortest word that can appear. Both are easy to work around — paste the text, look at what dominates, and if it is a word you did not mean to ask about, take it out and run it again.
How to use
- Paste the text you want visualised.
- Filter out stop words so common filler does not dominate.
- Adjust the colours, fonts and layout to suit where it will be used.
- Regenerate if the packing leaves awkward gaps.
- Download the finished image.
Frequently asked questions
What are stop words and why remove them?
Very common words such as the, of and and, which would otherwise dominate any cloud entirely while telling you nothing about the subject. Removing them is what turns a word cloud from a picture of the English language into a picture of your particular text.
Are word clouds actually good at conveying information?
They are widely criticised by people who work with data visualisation, and the criticism is fair. Area is difficult to compare accurately by eye, longer words take more space regardless of how often they appear, and the layout is arbitrary — so proximity implies relationships between words that simply do not exist.
So when is a word cloud the right choice?
When the goal is decorative or evocative rather than analytical: a conference poster, a book cover, a quick visual impression of a theme. The moment somebody might draw a conclusion from the relative sizes, a ranked bar chart conveys the same data far more precisely and far more honestly.
Should related word forms be combined?
Usually yes. Counting run, runs and running as three separate words understates how much the text discusses running. Doing it well requires careful stemming, since crude approaches merge words that ought to stay distinct, but leaving them entirely separate can be quite misleading.
Why do two clouds of the same text look different?
Because the layout placement is randomised, so each generation packs the words together differently. Nothing about the underlying frequencies changes at all — which rather underlines the point that the arrangement carries no meaning worth reading into.
Does my text get uploaded anywhere?
No. The text is analysed and the image is generated entirely in your browser, so nothing is transmitted to a server. That matters if what you are visualising is an unpublished draft, an interview transcript, or anything else confidential.
How much text do I need for a useful cloud?
Enough that frequencies mean something — a few hundred words at minimum, and ideally several thousand. Below that, one or two incidental repetitions dominate the result, and the cloud reflects noise rather than any real emphasis in the writing.
🔒 This tool runs entirely in your browser. Nothing you enter is uploaded, logged, or stored.