Make a scanned PDF searchable — OCR on your own device
A scan is a photograph of words. OCR reads the picture and lays an invisible, selectable text layer over it, so you can search, select and copy. It happens here, on your device — the scan is never uploaded.
The costs, first: the OCR engine is 1.5 MB and your language model loads with it — English 2.0 MB, Simplified Chinese 1.7 MB. Measured, not estimated. Both load from this site, only on this page, only when you run it.
All the tools
What OCR actually does
It does not turn pixels into text. It lays an invisible, selectable text layer over the picture — the page looks identical afterwards, and the recognized text now exists twice: once in the pixels, once in the layer. (This is also why redacting an OCR'd page always destroys pixels: deleting only the layer would leave the secret readable in the image.) Your pages are not re-encoded; the pixels are untouched.
Printed text, stated expectations
The recognizer is built for printed text. Handwriting is outside what it was trained on, so don't expect much — but it's your device and it costs you a moment, so you're free to try. Either way the confidence score is reported afterwards. On our printed test page it reads near-perfectly (confidence 93/100); on a real handwritten page it produced nonsense and scored 41/100, and the tool said so instead of handing over the nonsense as if it were a result.
Confidence is a signal, not a guarantee
A recognizer can be confidently wrong — an 8 read as a B can score well. Skim the recognized text against the image before you rely on it, especially for numbers. The tool puts the score in front of you precisely so you know how much skimming is warranted.
Why OCR lives on its own page
The engine plus a language model is a 3.2–3.5 MB download (1.5 MB engine plus your language model) — roughly ten times the weight of any other tool here. It loads on this page only, so nobody who came to compress a file pays a byte for it.
Questions people actually ask
- How do I make a scanned PDF searchable?
- Drop the scan here, pick your language, and run it. The tool renders each scanned page, recognizes the printed text, and writes an invisible text layer over the image — the output looks identical but can be searched, selected, and copied.
- Does it work on handwriting?
- Usually not well, and the tool will not pretend otherwise. It is trained on printed text; on our real handwritten test it scored 41/100 and produced nonsense. You are free to try — it costs a moment on your own device — and the confidence score will tell you honestly how it went.
- Which languages are supported?
- English (2.0 MB model) and Simplified Chinese (1.7 MB model) at launch. One model loads at a time, from this site — never from a third-party CDN.
- Is my scan uploaded for recognition?
- No. The OCR engine runs in your browser as WebAssembly. The scan, the recognition, and the output all stay on your device.
- Will OCR change how my pages look?
- No. The text layer is invisible and the page pixels are untouched — the file just gains searchable text underneath the picture.
We invite you to double-check our work
Open your browser’s developer tools, switch to the Network tab, and run any tool on this page and nothing leaves your device.
The whole source is on GitHub. Read it.
Found a bug, or a claim that doesn’t hold up? Tell us: pdf@geodeforge.com. An example file turns a guess into a quick fix — and if yours is too sensitive to share, we understand; a description still helps.
That’s the difference between a privacy policy and a fact.
This slot pays for the page
One static, labeled sponsor link — an image we host and a plain link. No ad network, no tracking script, no third-party request. That is the entire business model, and it is why nothing above has a price.
Sponsor this slot →More from GeodeForge
GeodeForge is a one-person studio that crafts open-source tools that do what they say, and let you check — in your browser.