Image to Text (OCR)
Pull the text out of a screenshot or picture in around thirty languages. Runs entirely in your browser, so the image never leaves your device.
What this reads well, and what it does not
Screenshots, exported documents and printed text photographed square-on come out close to exact. Photographs taken at an angle, low contrast images and text over busy backgrounds get noticeably worse. Handwriting does not work at all, and multi-column layouts and tables lose their structure. Every result carries a confidence score so you can tell which case you landed in.
Drop an image here
Or press Ctrl+V / Cmd+V to paste a screenshot. Nothing is uploaded.
Max 1 files, up to 50MB each
FAQs
How accurate is this OCR?
On clean digital text it is very accurate, typically somewhere between 95 and 99 percent of characters correct. That covers screenshots, exported documents, slides and anything captured straight from a screen. Accuracy drops sharply on photographs taken at an angle, low contrast images, and text sitting over a busy background. Every result comes with a confidence score so you know which situation you are in rather than having to guess.
Can it read handwriting?
No, and it is better to say so plainly than to waste your time. The engine here is trained on printed type, and handwriting is a fundamentally different problem. You will get a page of nonsense. Handwriting recognition needs a different class of model than anything that runs inside a browser tab.
What works best?
Screenshots are ideal, because the text was rendered rather than photographed. Beyond that: get the text straight rather than skewed, fill the frame with it, use good contrast, and capture at the highest resolution you can. A photo of a page taken square-on under decent light does well. The same page at an angle in dim light does not.
Does it keep tables and columns?
Not reliably. The engine reads in lines, so a two-column layout tends to come out with the columns interleaved, and table structure is flattened into plain text. For anything where the layout carries meaning, expect to reassemble it yourself. Single-column text is where this tool is strongest.
Are my images uploaded anywhere?
No. The recognition runs in your browser using WebAssembly, and the image never leaves your device. The engine and the language model are downloaded to your browser to do the work, rather than your image being sent away to be processed. That matters for anything containing personal details, invoices or identity documents.
Why is the first run slow?
The engine and the language data are fetched the first time you use it, which is a few megabytes. That download is cached afterwards, so subsequent runs in the same browser start almost immediately. Switching to a new language downloads that language once.
Which languages are supported?
Around thirty, including English, Spanish, French, German, Portuguese, Russian, Arabic, Hindi, Japanese, Korean and both Chinese scripts. Pick the language before you run it, because recognising Spanish text with the English model mangles every accented character.
Can I extract text from a PDF?
Not directly here. Convert the PDF pages to images first with our PDF to Image tool, then run each page through this one. If the PDF already contains selectable text, you do not need OCR at all, you can copy it straight out of the document.
What can I do if the result is poor?
Recapture rather than reprocess, because no amount of retrying changes what the engine can see. Crop tightly to just the text, straighten the image, and increase the contrast. If you photographed a screen, take a screenshot instead. If the source is handwriting or a heavily stylised font, this is not the right tool.