Free OCR that runs on your device, not on a server
A photo of a page is not a document — you can't search it, quote from it, or paste it into an email. Image to text conversion (OCR, or optical character recognition) turns that picture of words back into real, editable text. This page does it free, with no account and no uploads: the recognition engine runs 100% in your browser tab, on your phone or computer.
The privacy angle is the whole point. Most online OCR tools work by uploading your image to a server, reading it there, and sending text back. That's fine for a meme; it's a bad idea for a medical report, a bank statement, a legal notice, or a photo of someone's ID. Here the engine comes to your image instead: a one-time download of the OCR software from the library's CDN, cached by your browser, and from then on your photos never move.
It handles the everyday cases well: receipts for expense reports, whiteboard notes after a meeting, a page of a book you want to quote, a screenshot of an error message you need to Google, a restaurant menu in another language. Upload, read, copy — that's the workflow.
How to convert an image to text
- Pick the recognition language. English + Hindi is the default and covers most mixed documents. Match the language to the photo — it noticeably improves accuracy, especially for Hindi (Devanagari) text.
- Choose your image. Tap the big button to open your file picker, or drag and drop the photo onto the upload area. JPG, PNG, WebP, GIF and BMP work; iPhone HEIC photos are converted automatically in your browser first.
- Let the OCR engine load (first time only). The recognition software downloads once from the library's CDN with a progress bar. After that it's cached — your next conversion starts instantly.
- Review the extracted text. The text appears in a preview box with word count, character count, and the engine's confidence score. Skim it — especially names, numbers, and addresses — before you rely on it.
- Copy or download. Copy the text to your clipboard or grab it as a clean .txt file. Either way, the picture of words is now real text you can search, edit, and reuse.
Where image-to-text OCR actually saves time
Receipts and bills. Snap a receipt and convert it instead of retyping line items into your expense tracker. Dates, totals, and merchant names copy straight out.
Whiteboards and meeting notes. That wall of ideas at the end of a brainstorming session? One photo becomes meeting notes you can actually search next month.
Book pages and quotes. Want to quote a paragraph from a physical book in an essay or a post? Photograph the page and copy the text — faster than typing, and no transcription typos.
Screenshots of errors and chats. Screenshots are OCR's best friend: perfectly sharp text with zero camera blur. An error dialog becomes searchable text you can paste into a search engine in seconds.
Signs, menus and labels while traveling. Photograph a sign or a menu and get the words as text you can translate, save, or share — especially useful combined with the Hindi + English mode.
Tips for accurate OCR
- Light the page evenly. One soft, even light source beats a bright lamp casting a shadow across half the page. Daylight from a window is ideal — avoid flash, which creates hotspots.
- Hold the camera straight above the page. Tilted, keystoned text is harder to read than straight-on text. Line the page edges up with your screen edges before you shoot.
- Fill the frame with the text. Text that occupies most of the photo has more pixels per letter. Crop out the desk, the coffee cup, and the empty margins — or just move closer.
- Pick the right language. A Hindi document run through the English-only mode will come back mangled. The English + Hindi mode handles mixed documents well.
- Clean the lens. It sounds silly, but a smudged phone lens is the #1 cause of "why is my OCR bad" — one wipe on your shirt fixes a shocking number of blurry results.
- Screenshots beat photos. If the text is already on your screen, screenshot it instead of photographing the screen. Zero blur, zero glare, near-perfect accuracy.
Copy text from any screenshot in seconds
Screenshots are OCR's sweet spot: crisp edges, uniform fonts, no camera distortion. Error messages, UI labels, chat bubbles, code snippets, meme captions — the text you can see but can't select. Upload the screenshot, pick the language (or leave it on English + Hindi auto), run OCR, and copy.
This beats retyping in the cases where retyping hurts most: long error codes you must transcribe character-perfect, addresses from a map screenshot, or text trapped in an app that blocks selection. The result lands in an editable box — fix any hiccups, then copy or download as a .txt.
What OCR gets wrong (and how to fix it)
Most OCR errors come from the image, not the engine — which means most are fixable. Tilted photos: straighten the shot before uploading; even a few degrees of rotation costs accuracy. Low contrast or dim light: retake in bright, even light — glare and shadows break letter shapes. Tiny text: crop tightly around the text so it fills the frame; OCR reads pixels, and more pixels per letter means fewer mistakes. Busy backgrounds (posters, banners): crop out the graphics and keep only the text block.
Then proofread the confidence zones: names, numbers, and addresses deserve a second look against the photo. And know the hard limits: cursive handwriting stays unreliable, decorative script fonts are designed for looks not legibility, and vertical or heavily rotated text confuses the engine. Printed, straight, well-lit text — that's where OCR is genuinely excellent.
Honest limits: what OCR still gets wrong
No OCR is magic, and this one is no exception. Handwriting is the big one: neat, well-spaced block letters are sometimes readable, but cursive and cramped handwriting usually come out garbled — that's a limit of OCR technology everywhere, not just here. Low light and blur are the next killers: a dim, shaky photo of a page will confuse any engine. Shadows, glare, and reflections (looking at you, laminated menus) break letter shapes. Decorative and script fonts — wedding invitations, logos, stylized headings — are designed to look pretty, not to be machine-readable. Very tiny text and rotated or vertical text also struggle.
The results show a confidence score: high 90s means the engine is sure; low 80s and below means proofread names, numbers, and addresses against the photo. And if your source is a scanned PDF rather than a photo, the OCR PDF tool handles whole documents page by page — privately, on your device, with no upload and no account.