Convert · On-device OCR
Image to Text
Pull the text out of screenshots, photos, scans and whiteboards. The OCR engine runs in your browser — nothing is uploaded, ever.
Free OCR that never sees your documents
“Image to text” tools usually work by sending your picture to a server, running OCR there, and sending the words back. That is a problem when the picture is a contract, an ID, a medical letter or a whiteboard full of unreleased plans. This tool takes the opposite approach: the recognition engine — Tesseract, the same open-source OCR used by archives and libraries worldwide, compiled to WebAssembly — downloads to your browser and reads the image on your device. Once the engine is cached, it even works offline.
That architecture is also why there are no limits here. Sites that pay for OCR servers cap you at a few pages a day or paywall batch processing. Because your own hardware does the work, you can extract text from one screenshot or three hundred scans — free either way, with a separate progress bar for every image.
Getting the best results
- Match the language. Pick the language of the text before running — each language uses a purpose-trained model (12 available, including Hindi, Japanese, Korean, Arabic and Simplified Chinese).
- Favour sharp, flat, horizontal text. Screenshots and scans are nearly perfect; for phone photos, shoot straight-on in good light.
- Crop to the text first. If a photo has lots of background, trimming it with the crop tool speeds recognition and improves accuracy.
The extracted text appears next to each image in an editable box — fix anything the OCR misread, then copy a single result, copy everything at once, or download the lot as a plain-text file.
Common questions
Is my image uploaded for OCR?
No. The engine runs entirely in your browser via WebAssembly. The first run downloads the engine and your chosen language model (a few MB, cached afterwards); your images never leave the device.
How many images can I process?
Unlimited — no daily caps, no signup. Each image shows its own progress bar.
Can it read handwriting?
Clear print works best. Neat handwriting sometimes works; cursive generally doesn't — that's a limitation of OCR models of this class, and we'd rather be honest about it.