Android coming soon

The Android app launches in the coming weeks. Drop us a line at [email protected] to be notified at launch.

← Back to blog
Sep 2, 2026 Turn a photo into text: the method that works cosmonote.ai

You photographed a page. A leaflet, a contract, a page from a book, the whiteboard at the end of the class. The photo sits in your camera roll and does nothing for you. You cannot search it, copy from it, or paste it into an email. It sleeps among four thousand other images.

Turning a photo into text is the step missing between the two. Once the text is out, the page becomes useful again: you can read it back, quote it, translate it, summarise it, or simply find a figure in it three months later.

What usually goes wrong

Most attempts fail before the text recognition even starts, because of the photo itself. A page shot at an angle gives you lines in a trapeze that nothing straightens out. A shadow cast by your own hand eats half the paragraph. Too wide a frame drowns the page in the pattern of the table.

Three habits fix that. Lay the sheet flat, get above it rather than beside it, and look for light coming from the side rather than from behind you. This is not art photography, it is just enough for every line to stay a line.

A scanning app does the rest. It finds the edges of the page, straightens the perspective, crops what sticks out and puts the white back where it belongs. That step is the difference between clean text and a mush of characters.

The handwriting question

For a long time handwriting was the wall. Software read print correctly and produced nonsense the moment a human hand was involved. That is no longer true. Legible handwriting, even hurried, even with crossings-out and arrows in the margin, comes through well today.

Two limits remain. Very cursive, tightly packed writing needs a sharp photo, otherwise words run into each other. And highly specialised symbols, maths formulas or chemical notation, sometimes come out approximate. For a lecture, a meeting scribbled by hand, a recipe notebook, the result is thoroughly usable.

A good tool warns you when a page looks doubtful instead of confidently handing you wrong text. You retake the shot on the spot, not three weeks later when you discover the problem.

Text alone is not enough

Here is the classic mistake: stopping at raw text. You get a block of words with no headings, no lists, no hierarchy, and you end up with something as unreadable as the photo you started from.

What you actually want is a note. The heading stays a heading, the list stays a list, the table keeps its columns. And on top of that, a summary telling you in three lines what the document holds, with the important points lifted out of the body.

That is the idea behind scanning a document with Cosmonote: the page becomes a structured note, with the full text kept underneath, page by page. You keep the photo and gain everything else.

After extraction, ask questions

The real gain comes once the document has become a note. You can talk to it. Ask for the end date on a lease, the dosage on a leaflet, the list of amounts in a quote. On a scanned lecture, ask for flashcards or a quiz to test yourself.

This is where plain text recognition and a notes app part ways. The first hands you characters. The second hands you a document you can work with, sitting among your other notes, searchable months later.

How long it takes

One page, three seconds. A ten-page document, under a minute if you keep shooting without stopping in between. That is far quicker than retyping, and above all you make no typos on the numbers, which matters when the number is an amount or a dose.

Compare it with the alternative: retyping a dense page takes ten to twenty minutes and the result holds errors you will not spot. Across a semester of lectures, the gap is measured in days.

What you can do next

Once your pages are notes, they behave like the rest of your content. You share them as PDF, send them by email, pull a summary for someone who has no time to read it all. You can translate them too, which helps with an official document received in a language you barely read.

And because they are indexed, you find them again by searching a word you half remember. Which is exactly what a photo in your camera roll will never let you do.