Image to text (OCR)
Updated Nothing is uploaded
Turn any picture of text — a scanned letter, a photo of a sign, a receipt or a screenshot — into editable text you can copy. Drop or paste one or many images; OCR runs on your device and nothing is uploaded.
How to extract text from an image
- Drop images on the box, choose them, or press Ctrl+V (⌘V on Mac) to paste a screenshot.
- Select the language(s) of the text. The first time, the language model downloads (1–4 MB) and is then cached.
- Wait a few seconds per image while the text is recognized. Words the engine is unsure about are highlighted in yellow on the image and in the Review tab.
- Fix anything you need in the editable text box, then Copy text or Download .txt. With several images, Download all saves one combined file.
Getting the most accurate result
OCR accuracy depends far more on the picture than on the software. Tesseract, the open-source engine used here, reads best when letters are about 20–30 pixels tall, dark on a light background and straight.
| Problem | What to do |
|---|---|
| Small text (screenshots, crops) | Keep “Upscale small images” on — it enlarges images under 1600 px up to 3× |
| Faded, gray or yellowed paper | Keep “Boost contrast” on (stretches the darkest 1% to black and lightest 1% to white) |
| White text on a dark background | Keep “Fix dark mode” on — dark images are inverted automatically |
| Photo taken at an angle | Retake it straight-on, or crop to just the text first |
| Busy background, several columns | Crop each block of text separately for cleaner reading order |
| Wrong accents or ñ/ç/ß missing | Select the right language (you can pick more than one) |
What the confidence score means
For every word Tesseract reports a confidence from 0 to 100. The overall score is the average weighted by word length. As a rule of thumb, above 90% is a clean print scan that needs almost no correction, 70–90% usually has a few wrong characters, and below 70% means the image is blurry, tiny or stylized. Words under 70% are highlighted so you can check them against the picture.
What OCR can and can't read
- Works well: printed documents, books, invoices, receipts, slides, screenshots, signs and labels in Latin-alphabet languages.
- Hit-and-miss: decorative or very bold fonts, text over photos, low-resolution phone pictures, tables (text comes out row by row).
- Not supported: handwriting and cursive, which need a different kind of model. Math formulas come out as plain characters.
Private by design
Text recognition runs entirely in your browser using Tesseract (open-source OCR) compiled to WebAssembly. Your images are never uploaded; the only thing downloaded is the language model for the languages you pick, which your browser then keeps for next time.
Related tools
Frequently asked questions
Are my images uploaded?
No. Text recognition runs in your browser with Tesseract compiled to WebAssembly. The only download is the language model itself (1–4 MB, once per language); your images never leave your device.
Is this image to text converter free?
Yes — free and unlimited, with no sign-up or watermark. You can convert as many images as you like.
Which image formats work?
JPG, PNG, WebP, GIF, BMP, AVIF and iPhone HEIC photos. For a PDF, use an OCR PDF tool instead.
Can it read handwriting?
Not reliably. Tesseract is trained on printed text; neat block capitals sometimes work, but cursive handwriting won't.
Why is some text missing or in the wrong order?
Very small, low-contrast or decorated text can be skipped, and multi-column layouts may be read across columns. Crop to one block of text and keep the accuracy options on.
Your business here? Sponsor this tool from $9/week. See placements