Image to Text Converter (OCR)

OCR stands for optical character recognition: software looks at the pixels of an image, finds the areas that contain writing and works out which characters they represent. The result is real text that you can search, edit, paste into a document or feed into a spreadsheet, instead of a picture of text that you would otherwise have to retype by hand.

Krawly uses Tesseract, a long-running open-source OCR engine. It was first developed at Hewlett-Packard, released as open source in 2005 and later developed further with sponsorship from Google. Here it runs as Tesseract.js, a build compiled to WebAssembly so that it can execute inside a web page without any server. Roughly speaking, the engine cleans the image up into dark text on a light background, splits the page into blocks, lines and words, and then recognizes each line with a trained model for the language you selected.

That last step is why the language choice matters. Every language has its own model with its own alphabet, accented letters and common word shapes. If you tell the engine to read English while the image is in Polish or Vietnamese, it will still try, but letters with diacritics come out wrong. Choosing the correct language, or a combination of up to three, is one of the simplest ways to get cleaner results.

OCR works best on clear, printed text with good contrast and straight lines. Screenshots are usually ideal because the text is sharp and perfectly horizontal. For scans, around 300 DPI is a good target: high enough that small letters have clean edges, without creating huge files. If you photograph a page with your phone, fill the frame with the document, hold the camera parallel to the paper and avoid shadows from your hand or the phone itself.

Contrast matters more than colour. Black text on white paper is easy; light grey text on a beige background, or white text over a busy photo, is much harder. If a result looks poor, try cropping the image to just the text area, rotating it so the lines are level, and increasing contrast in any photo editor before running OCR again. Crooked, curved or perspective-distorted lines, such as a page photographed near the book's spine, are a common cause of missing words.

Very small text is another frequent problem. When letters are only a few pixels tall, there is not enough detail to tell an e from a c or an l from a 1. Zooming in before taking a screenshot, or scanning at a higher resolution, usually helps more than any later editing. Keep in mind that enlarging an already small image cannot add detail that was never captured.

Tesseract is built for printed text. Handwriting, cursive notes, signatures and heavily stylised or decorative fonts usually produce poor or unusable results. The same goes for low-resolution, blurry or heavily compressed photos. If your image falls into one of these groups, expect to correct a lot of the output by hand, or retype the text.

The output is plain text. Columns, tables, bold and italic styling, font sizes and images are not kept. A two-column article is flattened into lines of text, and a table becomes rows of words separated by spaces, so you may need to tidy up the structure after pasting it somewhere else. This tool does not produce Word documents or spreadsheets.

Each image also gets a mean confidence percentage. It is the engine's own estimate of how sure it was about the recognized words, not a guarantee of accuracy, but it is a useful signal: a noticeably lower score usually means the image needs to be sharper, straighter or cropped tighter.

People often run OCR on sensitive material: receipts for expense reports, screenshots of chats or bank apps, invoices, contracts, ID cards and letters. Many online OCR services upload those images to a server for processing. Krawly does not. The recognition runs entirely in your browser with JavaScript and WebAssembly, and your images are never sent to Krawly's servers.

The only things your browser downloads are program files: the Tesseract engine and the language model for each language you pick, a few megabytes each. After the first run they are cached, so later runs start faster. The image and the recognized text stay in the page's memory, nothing is stored, and closing the tab discards everything.

Because the work happens on your device, speed depends on your hardware. A recent laptop handles a screenshot quickly; an older phone takes longer on large photos. The tool works in current versions of Chrome, Edge, Firefox and Safari on desktop and mobile, and it is free, with no signup, no email and no daily limit.