imissfiles56 tools
  1. Home
  2. OCR tools
  3. Image to Text (OCR)

Image to Text

Pull the words out of a photo, scan, or screenshot. Recognition runs on this device.

Extract text from an image

Never uploaded

Your file is processed in this tab and never sent anywhere.

The ads come from Google. Your files stay on your device.

How to extract text from an image

  1. Add an image. A photo of a page, a scan, or a screenshot.
  2. Wait for recognition. The OCR engine loads once, then reads the image on this device.
  3. Copy or download. Take the text as plain .txt, or copy it straight to the clipboard.

About Image to Text (OCR)

Optical character recognition looks at a picture and works out which shapes are letters. That is a genuinely different job from text extraction, which reads characters already stored in a file. If you can select text in a document, you do not need OCR. If the document is a photograph of a page (a scan, a phone snapshot, a screenshot of something you cannot copy from) then recognition is the only route to the words.

This runs Tesseract compiled to WebAssembly, entirely in your browser tab. The engine and its English language data are around 6 MB and download once on first use, then stay cached. That first load is the slow part; everything afterwards is much faster. Recognition itself is CPU work on your device, so a laptop will chew through a page in a couple of seconds where an older phone takes noticeably longer.

Accuracy depends almost entirely on the input, and there is a lot you can control. A flat, well-lit, straight-on photograph of printed text at reasonable resolution will come out close to perfect. The things that wreck it are predictable: shadows across the page, the camera held at an angle so the lines converge, low resolution, and handwriting. Tesseract is trained on printed type: cursive handwriting is effectively out of reach, and even neat printing by hand is unreliable.

Doing this locally matters more than for almost any other tool here, because of what people actually run through OCR. Passports, driving licences, tax documents, medical letters, bank statements, and contracts make up an enormous share of it, precisely because those arrive as scans and photographs. Uploading that to a free web service to get the text back is a poor trade when your own laptop can do the recognition.

What affects accuracy

InputExpected resultHow to improve it
Flatbed scan of printed textNear perfectScan at 300 DPI or higher
Straight-on phone photo, good lightVery goodAvoid shadows, fill the frame
Angled photo of a pagePatchyShoot square on, or crop and straighten first
Screenshot of a webpageVery goodNothing needed
Low-resolution imagePoorRecapture larger: upscaling does not help
Neat block handwritingUnreliableExpect to correct it heavily
Cursive handwritingNot supportedTesseract is trained on printed type

Image to Text (OCR) FAQ

Is this OCR really free with no upload?

Yes. Tesseract runs as WebAssembly in your browser. The engine downloads once and then everything happens on your device: the image is never transmitted.

Why is the first run slow?

The OCR engine and English language data are about 6 MB and download on first use. They are cached afterwards, so subsequent runs start immediately.

Can it read handwriting?

Not reliably. Tesseract is trained on printed type. Neat block capitals sometimes work; cursive does not.

How do I get better accuracy?

Shoot square on rather than at an angle, avoid shadows, fill the frame with the page, and use the highest resolution available. Cropping to just the text before running also helps.

Does it work on other languages?

This build ships English only, to keep the download reasonable. Other language packs may be added later.

Why did it produce nonsense?

Usually low resolution, a strong angle, or poor contrast. Try recapturing the image flat and well lit: OCR quality is mostly an input problem.

Related tools

Related reading