For anyone with more text than anyone will read
Text to Visual Summary — Any Document, One Image
Text to visual summary is the whole job: you hand over something too long to be read by the people who need it, and one image comes back carrying the argument — the claim, what holds it up, and how the parts connect. This page is the general case; the four after it are the same machine pointed at a specific reader.
Before / after
The same content, twice
What you have
A document written to be filed, not read
Anatomy of a decision document
Most internal documents fail the same way: they record what was decided and lose everything that made the decision legible six months later. A document that survives contact with a new team has five distinct parts…
The decision itself belongs at the top, in one sentence, in the past tense. Anyone who reads no further should still be able to repeat it correctly.
Under it sits the reasoning… The third part is the options that were rejected… Fourth is reversibility… Last is ownership and date.
What comes back
One image carrying the whole structure
- The five parts stacked in the order the document argues for
- Each layer labelled with what it is for, not just what it is called
- The relationship between layers drawn, not left to a reader's inference
- Legible at the size it will actually be seen — a slide, a thread, a wall
The document is finished and nobody will read it
The writing is not the problem. The problem is what happens after: the review goes into a channel and gets three emoji, the paper you spent a week on is skimmed for its abstract, the handover doc is opened once by the person who wrote it. Length is a filter, and it filters out most of your readers before they reach the part that mattered.
The usual fix is to write a shorter version, which quietly costs more than the original. A summary is not a trim — it is a second act of judgement about which two or three things survive and what they hang off. Then you still have to make it look like something anyone would stop for, in a design tool that knows nothing about your subject.
So most documents get one shot at being read in the form they were written, and the version that would have travelled — one image, ten seconds, the shape of the argument visible at a glance — never exists.
What a visual summary is, and what it is not
A visual summary keeps the relationships. That is the whole distinction. A bulleted recap tells you there were five things; a visual summary tells you that two of them are consequences of the third, which is the part you would have had to read the document to learn. Structure is the information that summarising normally destroys, and it is the information a picture is unusually good at holding.
It is not an illustration. An illustration decorates a subject — a lightbulb next to a paragraph about ideas — and adds nothing you could not have got from the words. It is also not a chart: a chart plots numbers you already have, and nothing here connects to a spreadsheet. What comes back is closer to a diagram of your argument, drawn from the argument rather than chosen from a menu.
Plenty of things will turn text into a visual; far fewer produce one that can be read cold. The practical test is whether someone who never opens the document ends up knowing what it said. If the image only makes sense once you have read the source, it is decoration. If it stands on its own, it is a summary that happens to be visual.
What you can put in
Paste the text, drop the file, or hand over a URL. PDFs, Word documents, Markdown and plain text are parsed in your browser, and a PDF can be worked page by page, so one chapter out of a long book is a page selection rather than a copy-paste-and-clean exercise. The ceiling is 400,000 characters, which is longer than most books and far longer than anything you should put on one image.
What makes a source good is not its subject but its ratio. If the input is meaningfully longer than the output, generating from the text is doing the expensive work for you: reading it, deciding what is load-bearing, and throwing the rest away. If the input is a headline and three adjectives, there is nothing to compress and a design editor will be faster — worth saying plainly, because it decides whether this page is relevant to you at all.
A few properties of the source change the result more than any setting does. Prose beats slide fragments, because an argument written out has connective tissue a bullet list has already removed. Numbers are read exactly as written, so a figure that is wrong in your draft will be wrong and prominent in the picture. And a document that argues one thing produces a far better summary than one that surveys six — if yours surveys six, six images will serve you better than one crowded diagram.
- Reports, reviews and internal memos
- Papers, preprints and textbook chapters
- Articles and blog posts, pasted or by URL
- Meeting notes, transcripts and your own reading notes
- Documentation, policies and process write-ups
Which shape your text wants
Every route from document to visual summary starts the same way — the whole thing gets read, and most of it gets thrown away — and then they diverge. One text can become several different images, and they are not interchangeable: the same source can produce something that works in a feed and something that works in a board pack, and swapping them fails badly in both directions. The variable that actually decides it is where the output has to live, not what the document is about.
So the four pages below are this one, narrowed. Each starts from the same text to visual summary machinery and fixes the decisions that job has already made for you — the structure, the proportions, the amount of text a panel can carry — so you begin with them applied rather than guessed at.
- Read at thumb speed in a feed → a set of carousel cards
- Distilled out of a PDF, paper or book you have to get through → a knowledge card
- Built from notes, a lecture or a transcript rather than a finished document → a knowledge card from scratch
- Handed to a busy person in a deck or an email → a one-pager
- Placed above the article to make somebody open it → a banner, which is the one job here that is not a summary at all
How to do it in about a minute
- 01
Give it the text
Paste it, drop the PDF or Word file, or hand over a URL. Longer is fine — the work is deciding what survives, and there is more to decide when there is more to read.
- 02
Say where it will be seen
The one decision worth making yourself is the shape: upright for notes and phones, wide for slides and documents, square for feeds. Everything else has a sensible default.
- 03
Generate, then check it against the source
About a minute, and back comes a full-resolution PNG with no watermark. Read it once beside the document — this is a generated summary, and anything carrying numbers deserves thirty seconds of verification.
Questions
Questions about text to visual summary
Up to 400,000 characters. Length is not the constraint worth worrying about — focus is. A single argument makes a far better image than a survey of six topics, so if your document does several things, generate one image per thing.
The whole thing. It works out which claim is load-bearing and which material is example, caveat or citation, then picks a structure that matches what it found rather than pouring the text into a fixed template.
Yes — the image comes out in the language you write in. The interface is English, Japanese and Spanish, but the source text is what decides the output, so a document in any of those languages produces a card in that language.
No. What you get is a finished PNG, not an editable canvas, so there is no clicking into a heading to change a word. When something is wrong the loop is to fix the source text or switch the structure and generate again — fast when the problem is editorial, irritating when it is cosmetic.
It works only from the text you give it and never invents a source, but it is a generative model producing a summary: emphasis can land somewhere the document did not put it, and a figure can come through subtly wrong. Check anything that will leave your own notes against the original before it circulates.
Turn a document into one image
Paste your own text and the settings from this page are already applied. A free credit covers the first one.