Extract Text from an Image
Turn a screenshot, scan, or photo of text into text you can copy and edit.
Using toolbook means you agree to our Terms and Privacy Policy.
or drag them here
A screenshot, scan, or photo · PNG, JPG, WebP, or HEIC
Your files stay on your device
How to use it
Drop in a screenshot, scan, or photo and pick the language it’s written in. You get the text back, ready to copy and edit. Your images stay on your device.
Need the words from a screenshot, a scanned page, a receipt, or a photo of a sign? Drop the image onto the box above, choose the language it’s written in, and press Extract.
Pick the right language — it really matters. Reading French text as English won’t give an error; it gives believable-looking wrong words. (It doesn’t translate, either: the text comes back in the language it’s written in.)
Choose “Document” for pages and “Scattered text” for things like receipts, signs, and labels.
The first time, a reader is downloaded (about 4–7 MB). After that it’s saved, so next time starts instantly. Every result shows how confident it is — give it a quick proofread, because mistakes look like normal words.
Which languages are supported
Thirteen languages, chosen so that as many different alphabets as possible are covered. Each one is downloaded only when you pick it, and only once.
| Language | Download size |
|---|---|
| English | 1.9 MB |
| Spanish | 1.1 MB |
| French | 595 KB |
| German | 834 KB |
| Portuguese | 986 KB |
| Italian | 1.2 MB |
| Dutch | 2.9 MB |
| Russian | 1.5 MB |
| Chinese (Simplified) | 1.6 MB |
| Japanese | 1.5 MB |
| Korean | 1.1 MB |
| Arabic | 709 KB |
| Hindi | 901 KB |
Tips for better results
Crop to just the text first. It’s the single most helpful thing you can do.
Straighten the image if it’s tilted — a few degrees is fine, more than that hurts.
Use a screenshot instead of a photo of a screen whenever you can. Screenshots read almost perfectly.
Got nothing back, but the image clearly has text? Switch the layout to “Scattered text”.
FAQ
Does the language setting translate the text?
No. It tells the reader which language is in your image. Set it to match.
Why is there a download the first time?
The reader needs to get onto your device to work there. It’s saved, so it only happens once per language.
Can it read handwriting?
Usually not. It’s built for printed text.
Can it keep columns and tables?
Columns come out in the right order, but you get plain text, not a table. For tables in a PDF, use Extract Text from PDF.
How many images can I do at once?
As many as you like — no page limit or daily cap.
Why is some of the text wrong?
Check the language setting first. Otherwise, try a clearer, straighter, or cropped image.
How it worksFor the curious
This runs Tesseract, the open-source OCR engine, compiled to WebAssembly. Its recogniser is an LSTM neural network trained on text images — not a large language model, and no API is called. The engine (about 3.7 MB) and a language model (0.6–3 MB) are downloaded once from this site before your image is read, so they never count against the network figure, and the browser caches them for every tab.
Large photos are scaled to about 2600 pixels on the long edge before recognition, which improves accuracy as well as speed because the model was trained near that scale.
The language model supplies vocabulary and letter frequencies to resolve ambiguous glyphs, which is why a mismatch produces confident wrong words. “Document” layout detects columns and reads them in order; “Scattered text” makes no structural assumption. Output is plain text — for tables inside PDFs, the PDF text extractor exports CSV.
Every result carries a confidence figure; below about 60% on many words usually means a blurred, skewed, or low-contrast source. Tesseract is trained on printed text, so handwriting produces confident nonsense rather than an obvious failure.