/ Image tools — text extraction

Extract Text from Images

Pull the words out of a screenshot, a scanned page or a photo of a document and get editable text back. The recognition engine runs entirely inside your browser — your image is never uploaded, never stored and never seen by a server.

  • 100% local · nothing uploaded
  • JPG, PNG, WebP, GIF, BMP
  • English recognition
  • No sign-up

1. Add an image

First run downloads a language pack The English training data is about 10 MB and is fetched once, then cached by your browser. The first extraction takes noticeably longer than every one after it.

2. Extracted text

0 characters

0 characters 0 words 0 lines
100% LocalRecognition runs in your browser
No UploadYour image never leaves the device
Free ForeverNo account, no limits, no watermark
/ 01 — How it works

Three Steps, No Upload

The whole pipeline — decoding the image, running the recognition model, assembling the text — happens on your own machine. Once the language pack is cached you can even work offline.

Upload

Drop an image onto the panel or click to browse. A preview appears immediately so you can check you picked the right file and that the text is legible before spending time on recognition.

Extract

Hit Extract Text and the OCR engine analyses the image, reporting progress as it goes. Recognition typically takes a few seconds; a dense full-page scan takes longer than a short screenshot.

Use the text

The result lands in an editable box with a live character count. Fix any misread words in place, then copy it to the clipboard or save it straight to a .txt file.

/ 02 — FAQ

Frequently Asked Questions

Which image formats are supported?

Anything your browser can decode works: JPG/JPEG, PNG, WebP, GIF and BMP. The file is read straight from disk, so there is no conversion step and no quality loss before recognition.

Format matters far less than what the image contains. A PNG screenshot of on-screen text is close to ideal because the characters are pixel-sharp. A JPEG photo of a page taken at an angle in poor light is the hard case — not because of the format, but because of blur, skew and shadows. Animated GIFs are read from their first frame only.

Why is the first extraction so slow?

Because the recognition engine has to fetch its English training data, roughly 10 MB, before it can read anything. That download happens once. Your browser caches it, so every later extraction skips straight to the recognition step and starts in a fraction of a second.

Expect somewhere around ten to thirty seconds for the very first run depending on your connection, then a few seconds per image after that. If you clear your browser cache the pack is downloaded again on the next run. The pack is also what allows the tool to keep working offline once it is cached.

How accurate is the recognition?

On clean, straight, high-contrast text — screenshots, digital documents, flat scans at 300 DPI — accuracy is typically in the high nineties, and often good enough to use with only a glance over it. On a phone photo of a curved page under a desk lamp it can drop sharply.

Four things help more than anything else: resolution (text smaller than about 20 pixels tall is guesswork), contrast (dark text on a light background), straightness (crop and rotate so lines run horizontally), and focus. Handwriting is not supported — the model is trained on printed type. Always proofread the output before relying on it; the text box is editable for exactly that reason.

Is my image uploaded anywhere?

No. The image is decoded in the page and passed to a recognition engine compiled to WebAssembly that runs inside your browser. There is no upload request, no server-side processing and no copy of your file anywhere but your own machine.

You can verify it: open developer tools, switch to the Network tab and run an extraction. The only requests are the one-time downloads of the engine and the language pack — never your image. That is what makes this safe for ID documents, invoices, medical letters and internal reports that must not be sent to a third party.