Word Cloud Generator

About the Word Cloud Generator

Paste text, open a file (text, Word, PDF and more) or type a word, count list and get a word cloud in seconds: the more often a word appears, the bigger it is drawn. Stop words (“the”, “and”, “of” …) are removed in 58 languages — the language is detected from your text — and you can count two-word phrases, set a minimum frequency, leave words out, or count different spellings as one.

Choose the shape the words fill (a rectangle, circle, heart, star … or the silhouette of your own picture), the font, the direction of the words and the colours, then download a PNG, an SVG (real text or shapes) and the word-frequency table as CSV. Everything is done in your browser: your text and pictures are never uploaded.

Common uses

• Showing the main themes of survey answers, reviews or feedback at a glance.
• Illustrating a speech, essay, article or lyrics on a slide, poster or social post (PNG with a transparent background).
• Studying a text: which words a chapter, a speech or a set of interviews uses most.
• Visualising a keyword list sized by search volume or clicks, from a word, count export.
• Classroom activities and vocabulary games: students’ words become a shared picture.

How the words are counted

Text is cut into words with your browser’s Unicode word-break rules (UAX #29, through Intl.Segmenter), so Hindi, Arabic and Cyrillic words stay whole and Chinese, Japanese and Thai are split with a dictionary. Hyphenated words (“well-known”) and letters joined by a full stop (“example.com”, “node.js”) stay together, abbreviations such as “U.S.A.” and “Ph.D.” are counted without their full stops, “John’s” counts as “John”, and French or Italian elisions (“l’homme”) lose their article.

Words that differ only by capitals are counted together and shown in the spelling you use most. Numbers, web addresses and e-mail addresses are ignored unless you change that, words shorter than the minimum length are skipped (a single Chinese character is a word and is kept), and so is any string of more than 120 characters without a space, such as a code or a hash: it is not a word, and a picture could not hold it.

The share in the table is the count divided by all counted words. In a word list it is the count divided by the sum of all counts.

Stop words and languages

The stop-word lists come from stopwords-iso (MIT licence) and cover 58 languages. With Detect the language automatically, the language whose list covers the most of your words is used. It needs at least two different stop words of that language that make up at least 4% of your words, so a short list such as “cat dog bird” is never taken for Romanian or Danish and keeps all its words; a second language can be added for mixed text.

The English list is trimmed before use: 75 everyday nouns that are often the subject of a text (“system”, “work”, “world”, “year” …), about 70 words of opinion, size, request and reply that reviews and survey answers are made of (“good”, “great”, “best”, “new”, “help”, “thanks”, “yes”, “no” …) and its two-letter entries that are not real function words (country codes such as “ai”, “tv”, “uk”, “id”) are kept as words. To drop any of them, type it under Words to leave out. Lists only remove whole words: they do not reduce inflected forms, so “dog” and “dogs” are two words unless you merge them with Count different words as one.

How the words are placed

The layout follows the method Jonathan Feinberg described for Wordle, which d3-cloud also implements: each word is drawn into a bitmap of the pixels its letters cover; the biggest words go first, each tried along a spiral that starts near the middle until it meets no word already placed. MySmartCoPilot’s version starts from a board whose blocked pixels are the outside of the chosen shape, which is how a circle, a heart or your own silhouette are filled, and it uses a seeded random generator, so the same layout number and settings give the same picture again on the same device and browser (another system draws letters a little differently, which can move a word).

Sizes follow the count by the square root (balanced), directly, or by the logarithm. With Automatic word sizes (the default) the cloud keeps the ratio you set under Biggest to smallest and makes the whole set as large as it can be while every word still fits, so the picture is full. With Fixed sizes, the largest and smallest word are a share of the picture’s height, and Shrink words until every one fits reduces them together only as far as needed; the note under the picture says by how much.

Words are never drawn smaller than 9 pixels of the picture. If even that leaves no room for all of them, the least frequent words are left out and marked in the table.

Colours and contrast

Every colour of the built-in palettes has at least 4.5:1 contrast on its background (the WCAG level for normal text), with one variant for light and one for dark backgrounds. The colour-blind-safe palette uses the hues of the Okabe–Ito palette, darkened on light backgrounds for the same reason. Your own colours are checked against the background and the page warns about those that are hard to read.

Word list format

One word or phrase per line with its count: apple, 12, apple;12, a tab, or apple 12. Two columns copied from Excel or Google Sheets arrive separated by tabs and work as they are. A line without a count counts once. Put a word that contains a comma in quotes: "hello, world", 7. Use a comma for words that end in a number: windows 11, 5 (without it, “windows 11” would mean “windows” counted 11 times). Counts may be whole numbers, decimals (3.5), numbers with thousands separators (1,234 in a tab- or semicolon-separated list) or percentages (25%). A first row such as word,count or term,frequency is recognised and skipped. Lines that cannot be read are listed by line number and the rest is used.

The files you download

PNG is drawn at 1× to 4× the preview size; the size is lowered if your browser could not draw it (pictures are limited to 16 million pixels). A transparent background gives a transparent PNG. SVG keeps the words as real text with the font included in the file, or — for vector editors and cutting machines — turns the letters into shapes. CSV has the rank, word, count, share, type and whether the word is in the cloud; cells that a spreadsheet would run as a formula (starting with =, +, - or @) get a leading apostrophe.

Limitations

• The stop-word lists remove whole words only. Languages that attach endings to words (Korean, Turkish, Finnish …) keep many words that a stemmer would remove.
• There is no stop-word list for languages such as Tamil, Telugu, Kannada, Malayalam, Punjabi, Nepali, Sinhala, Burmese, Khmer, Amharic, Georgian or Kazakh. Their words are counted all the same; common words stay unless you type them under Words to leave out.
• Chinese, Japanese and Thai are split with the dictionary built into your browser; other browsers can split a sentence slightly differently.
• Packing is heuristic: the arrangement is good but not optimal, and another layout number gives another arrangement. More than about 300 words make each word small.
• The built-in fonts cover Latin, Vietnamese, Cyrillic, Greek and (Noto Sans and Poppins) Devanagari letters. Other scripts are drawn in your device’s font, which the PNG uses as well, or in a font file you open yourself. SVG files that must look the same everywhere should use Letters as shapes under Downloads, which needs a built-in font or your own font file.
• Text is read up to 1,000,000 characters. Text, CSV and HTML files may be up to 5 MB, documents (Word, PDF, PowerPoint, OpenDocument, EPUB) up to 60 MB. A PDF that is only scanned pictures has no text to count, and a password-protected PDF must be unlocked first. Emoji and symbols are not counted as words.
• A picture works best as a clean silhouette (dark on white, or on a transparent background). Photos with a busy background give ragged shapes.

Privacy

Everything happens in your browser. What you enter or open here is not uploaded or stored by MySmartCoPilot. The first time you use it, the stop-word lists (about 70 KB compressed) and the font you choose are downloaded from MySmartCoPilot. Your text, word lists and pictures never leave your device.