Image to Text

Optical character recognition (OCR) turns the text inside a picture into text you can actually select, edit, search, and reuse. Instead of retyping a receipt, a slide, a screenshot, or a page from a book, you let the computer read it for you. This guide shows you how to convert an image to text properly, how to get the most accurate result, and how to fix the things OCR commonly gets wrong — all with the free tool at the top of this page, which runs entirely in your browser.

What OCR actually does

A photo or a scan is just a grid of coloured dots. To a computer, the word “Invoice” in that picture is not text — it’s a pattern of pixels. OCR is the technology that looks at those pixels, finds the shapes of letters and words, and outputs real, editable characters.

Modern OCR (including the engine this tool uses, Tesseract) is powered by a trained neural network. It works line by line: it detects where the text is, segments it into words and characters, and predicts the most likely letters for each shape using a language model. That’s why choosing the right language matters so much — the model uses the patterns of that language to resolve ambiguous letters.

How to convert an image to text (step by step)

1. Add your image. Drag a JPG, PNG, or WebP onto the tool above, click Choose image, or paste a screenshot straight from your clipboard.
2. Pick the language. Select the language of the text in the image. This is the single biggest lever for accuracy — recognising English text with the Spanish model (or vice-versa) produces noticeably more mistakes.
3. Press Extract text. A progress bar shows the engine loading and then reading the image. The first run downloads the engine and language data; after that it’s quick.
4. Read and fix the result. The recognised text appears in an editable box. Skim it and correct any slips — OCR usually nails the bulk of the text and trips on a few odd characters.
5. Copy or download. Copy the text to your clipboard with one click, download it as a .txt file, or export a searchable PDF that keeps your original image with a hidden text layer.

Get the most accurate result

OCR quality depends far more on the input image than on any setting. A few habits make a big difference:

• Use the sharpest version you have. A crisp screenshot or a 300 DPI scan reads far better than a small, blurry, or heavily compressed thumbnail. If text looks fuzzy to you, it looks fuzzy to the engine too.
• Keep the page flat and upright. Photograph documents straight-on, not at an angle, and rotate the image so the lines are horizontal. Skew and perspective are common causes of garbled output.
• Maximise contrast. Dark text on a light background is ideal. Shadows, glare, busy backgrounds, and low light all hurt. If a photo is dim, brighten it before running OCR.
• Crop to the text. If only part of the image is text, cropping away the rest gives the engine less to misread and can speed things up.
• Match the language. This is worth repeating: set the language to the one actually in the image.

Reading different kinds of images

Screenshots are the easiest case: the text is already crisp, high-contrast, and perfectly upright, so recognition is usually excellent. Scanned documents are next-best, especially at 300 DPI. Photos of documents work well when they’re flat, sharp, and evenly lit. Photos of signs, packaging, and labels are more variable — stylised fonts, curved surfaces, and reflections all add difficulty. Handwriting is the hardest: this tool targets printed text, and while tidy handwriting sometimes comes through partially, you should expect to fix it by hand.

Languages and scripts

The tool supports over 100 languages, spanning Latin, Cyrillic, Greek, Arabic, Hebrew, Devanagari, and CJK (Chinese, Japanese, Korean) scripts, among others. A few practical notes:

• One language at a time is usually best. Choosing the language that dominates the image gives the cleanest result.
• Right-to-left scripts (Arabic, Hebrew, Persian, Urdu) are recognised and shown in the correct reading direction in the editor.
• CJK languages have their own dedicated data; pick Simplified or Traditional Chinese to match the document.

Only the language you choose is downloaded, so switching languages the first time fetches that pack — quick on a normal connection.

What “searchable PDF” means

A normal PDF made from a photo is just a picture — you can’t select or search the words. A searchable PDF keeps your original image exactly as-is but adds an invisible text layer behind it, positioned over each word. The page looks identical, but now you can highlight text, copy it, and find words with Ctrl/Cmd-F. It’s the ideal format for archiving receipts, contracts, and scanned notes you’ll want to search later. Choose Download PDF after extracting to create one.

Is it private? Where does the text recognition happen?

The recognition runs in your browser, on your own device, using WebAssembly — the same technology that lets near-native code run on a web page. The first run downloads the engine and the language pack from a content-delivery network; after that, the actual reading of your image happens locally.

How any data associated with this tool is handled is described in our privacy policy. If you’re working with something highly sensitive, an offline desktop OCR application is always the most private option.

After OCR: cleaning up the text

OCR gets you 90–99% of the way there; the editable box is where you finish the job. Common things to fix:

• Look-alike characters: a recognised 0 that should be O, 1 versus l, or rn read as m.
• Line breaks: a scanned paragraph may keep the line breaks of the page. Remove them if you want flowing text.
• Stray marks: specks, stamps, and underlines can produce odd symbols — just delete them.

Once it reads correctly, copy it into your document, spreadsheet, or notes. For receipts and forms you’ll search later, the searchable PDF is often the better keep.

When to reach for OCR

Reach for image-to-text whenever the words you need are trapped in a picture: a slide from a talk, a screenshot of an error message, a printed table you’d rather not retype, a recipe in a book, a business card, a menu, or a page you want to translate. Snap it, run OCR, fix a couple of characters, and you have editable text in seconds — no typing required.