What happens when you press Read
Opening a picture decodes it in the browser, the same way every tool on this site does. Pressing Read loads the text reader — Tesseract’s LSTM engine, compiled to WebAssembly, with its English language file — from this site, once, and runs it in a background worker so the page stays usable. The picture is copied onto a canvas at a size the reader likes (small pictures are enlarged, very large ones are reduced to 4,000 pixels on the long edge) and read there.
What comes back is a set of lines, each with a box and a confidence. The boxes are drawn over the picture, the words the reader was least sure of are outlined, and the text goes into an editor you can correct before you copy it or download it as plain text or as JSON with every box.
Getting a better read
The reader is good at print and only as good as the picture. Straight on beats an angle; close beats far; even light beats a shadow across the page. Pick “One block of text” for a single paragraph or a caption, and “Scattered text” for a sign, a label or a screenshot with text in small separate pieces. If the result runs lines together oddly, the other layout setting usually fixes it.
What it will not do
- It does not read handwriting. The page says so when the confidence across the whole picture is low.
- It does not keep the layout. Tables, columns and fonts are not rebuilt; you get the words in reading order.
- It offers English and Hindi only. Other languages are not loaded, and the page does not claim them.
Questions people ask
Is my picture uploaded?
No. The text reader (Tesseract, compiled to WebAssembly) is downloaded from this site the first time you press Read and runs in your browser tab. The picture is handed to it in memory. There is no server here that accepts a file, and the network requests the page makes are plain downloads of the reader's own files.
Does it work with JPG, PNG, screenshots and phone photos?
Yes. JPG, PNG, WebP, GIF, BMP, TIFF, AVIF and, in most browsers, HEIC all open. Screenshots read best with the layout set to “Scattered text” when the text is in small separate pieces; a scanned page or a notice reads best on “Page with columns”. A phone photo reads best when it is taken straight on, close, in even light.
Does it read Hindi?
Yes, printed Hindi in Devanagari. On Automatic, the page measures the rows of type first; Devanagari words hang from a continuous headline, and when most rows look like that it asks once before loading the Hindi reader (about 1.4 MB). Lines that mix Hindi and English are read with both languages. You can also pick Hindi directly.
How accurate is it?
On clean printed text it is usually very close; on photos at an angle, low light, decorative fonts or very small type it makes mistakes. Every word comes with the reader's own confidence, and words under 60% are outlined on the picture and listed so you can check them. This site has not yet published a measured accuracy figure for its test pictures, so it does not quote one.
Can it read handwriting?
No. The reader is built for print. On handwriting it produces a low confidence across the page, and the page says so rather than presenting the guess as text.
How is this related to alt text?
When a picture is mostly text — a notice, a quote card, a screenshot of a message — the text is the picture's alternative, and WCAG asks for it. When the read text is short enough, the page shows it graded as alt text, ready to copy. When it is long, the advice is a one-line alt plus the text on the page.
Where to take this next
- Alt textWrite the alt attribute yourself, with the picture open and the findings live.
- Alt text checkerPaste alt text you already have and read the findings. No picture required.
- Long descriptionThe paragraph a chart or a diagram needs when one attribute cannot hold it.
- WCAG 1.1.1 and GIGW, in plain termsWhat the success criterion asks for and what an Indian government audit checks.
