Why a word cloud still matters
A word cloud is the fastest way to answer one specific question: what is this piece of writing actually about? Pasted into a slide, glued to a research debrief, or pinned at the top of a customer-feedback report, a good cloud collapses ten pages of prose into a picture you can read in two seconds. The biggest words are the topics. The medium ones are the supporting ideas. The small ones are the texture. That is the whole pitch.
Word clouds are not a serious analytics tool. They will not tell you sentiment, they will not surface causality, and they will lose hard to a structured topic model on every measurable axis. What they do well is communicate gist to a non-technical audience faster than any other format, and they do it without you having to write a single sentence of explanation. That is why they keep showing up in product reviews, classroom posters, marketing decks, and earnings-call recaps thirty years after they were invented.
This tool runs entirely in your browser. The text you paste never leaves your tab. There is no upload, no signup, no account, no rate limit, and no tracker on the page beyond a single privacy-respecting analytics pixel. You can disconnect from the internet, paste a confidential transcript, and the cloud will still render.
How the counts are computed
A word cloud is only as honest as its tokenizer. Here is exactly what this one does.
Tokenization. Every run of letters, digits, and apostrophes is one word. The regex is Unicode-aware, so Devanagari, Cyrillic, Arabic, and Han ideographs all parse correctly. "Don't" is one word, "self-driving" is two words because the hyphen is not a letter, and "U.S.A." is three words. This matches what almost every other word cloud generator does.
Lowercasing. By default every token is lowercased before counting, so "Apple" and "apple" become the same word. Flip the Case sensitive toggle to keep them separate. We default to lowercase because most pasted text is prose, not a list of brand names, and a cloud where "The" appears next to "the" feels broken.
Stopword filtering. A stopword is a high-frequency function word that carries grammar rather than meaning: "the", "and", "of", "is" in English; "el", "la", "de" in Spanish; "का", "के", "है" in Hindi. Stopwords would dominate any cloud they were left in, so every word cloud generator removes them. This tool ships five curated lists (English, Hindi, Spanish, French, German) plus a None mode for the rare case when you actually want filler words in.
Custom stopwords. Beyond the language list, the chip input lets you add your own words to ignore. This is useful when one specific term (the brand name, the speaker's name, the survey topic) would otherwise dwarf everything else. Type a word and press Enter or comma; tap the × to remove one; press Backspace on an empty input to nudge off the last chip.
Min word length. Slide this up to drop short words. Setting it to 4 kills "she", "him", "but", "yet" without you having to add them to the stopword list. Setting it to 1 lets everything through, even single-letter words and digits. The default is 3, which is the sweet spot for English prose.
Bigrams. A bigram is a two-word phrase. Toggle Include phrases and the tool counts every adjacent pair of kept words alongside the unigrams. "Customer support" appears next to "customer" and "support" separately. Phrases that only show up once are dropped so the cloud doesn't get cluttered with every adjacent pair. This is the right setting for survey analysis, marketing copy, and any text where compound nouns carry meaning ("machine learning", "battery life", "social media").
Max words. The visible cloud is capped at this number. Sorting happens by raw count first, alphabetic order as the tiebreak, then the top N are rendered. Lower the cap if the cloud feels crowded. Raise it if you want a long tail.
How the sizes are calculated
The most-frequent word gets the largest font (around 92 pixels in the default layout). Every other word scales relative to it on a square-root curve: a word that appears 100 times is roughly 10 times larger than a word that appears once, not 100 times larger.
This matters more than it sounds. Linear scaling makes the densest word so big that the next word looks miniature, which collapses the cloud into a single-word brag. Square-root scaling preserves the rank ordering ("most frequent" still wins) while keeping the rest of the cloud legible. Most professional word cloud libraries use the same trick.
What the colors mean
Nothing semantic. Colors are assigned in frequency rank order from the chosen palette and cycle when the palette runs out. The Saffron palette mirrors the ToolBook brand accent and reads well on a warm-paper background. Ocean and Forest are calm and good for print. Sunset and Spectrum are loud and good for slides. Mono Ink uses the warm-neutral ink ramp and is the safe pick when the cloud will sit inside a layout that already has its own accent color.
If you want to encode meaning in color (positive vs negative sentiment, category buckets, etc.) export the SVG and recolor it in Figma. That is well outside what a free, in-browser tool should try to do.
Rotation modes, explained
Rotation changes how the cloud feels but not what it says.
None. Every word is horizontal. The cloud reads like a paragraph. Use this when the cloud will sit next to body copy and you want a quiet, readable visual.
Mixed. Most words horizontal, a few at ±30 degrees. The default. Looks lively without becoming a puzzle. Picks the rotation deterministically by word rank so re-renders are stable.
Vertical. Alternates 0 and 90 degrees. Packs more words into less space at the cost of forcing the reader to tilt their head. Use it when space is tight (a sidebar widget, a square Instagram post).
Exports and what to use each for
PNG. Raster image at roughly 2400×1500 pixels. Drop it into a slide, an email, a Twitter post, a Notion page, anywhere that wants a picture. PNG preserves the rendered look exactly but cannot be edited.
SVG. Vector. Scales to any size without pixelation. Open it in Figma or Illustrator to recolor, resize, add a mask shape, or composite it into a poster. SVG is also the right format for print, since it stays sharp at any DPI. Open the SVG in a browser to preview it before downloading the final asset.
JSON. The raw frequency data. Each entry has the word, the count, the density percentage, and a flag for whether it is a bigram. Drop the file into a spreadsheet to pivot the data, or into another analysis tool to do something the cloud can't (sentiment, clustering, change-over-time charts).
Copy SVG. Same SVG, just on your clipboard. Useful for pasting straight into a design tool that supports SVG paste-in (most do).
When not to use a word cloud
We should be honest. Word clouds are bad for:
- Comparing two datasets. Two clouds side by side look impressive but make it almost impossible to tell whether the changed words are actually changing the meaning, or just changing position because the layout engine reshuffled. A bar chart of "top 20 words by frequency" beats a pair of clouds every time.
- Showing change over time. Same problem. Use a line chart or a small multiples view.
- Numeric data. A cloud of numbers tells you nothing a histogram couldn't tell you ten times better.
- Long, structured documents. A research paper, a legal contract, a financial report: each section says something different and a single cloud averages them out. Run the cloud on individual sections instead.
Use a cloud when the goal is to communicate the topic of a single body of text to a non-technical audience as fast as possible. That is the one thing it does better than anything else.
A note on privacy
Word clouds get applied to sensitive material more often than people think: customer survey verbatims, internal user research, NDA-protected interview transcripts, draft press releases, exit-interview notes. Most online generators upload your text to a server for processing, which is a real privacy problem and sometimes a contractual one. This tool does not. Tokenization, frequency counting, layout, and rendering all happen in your browser using JavaScript. The URL stores your visual settings (palette, language, max words, etc.) so a shared link reproduces the picture, but the text and your custom stopwords stay local. Nothing about your input is sent anywhere.
That is the version we wished existed when we kept reaching for these tools and finding none we'd happily paste a contract into.