Image to text converter (OCR)

1. Add your images. Drag pictures onto the box above, click to browse, or simply press Ctrl+V (⌘V on a Mac) to paste a screenshot straight from your clipboard. You can queue several images at once — each one is processed in turn.
2. Pick the language. English is selected by default; switch to Spanish, French, German, Italian, Portuguese, Hindi, Arabic, or Urdu if your image contains text in one of those languages. Choosing the right language matters — the recognizer uses language-specific training data, so English mode reading French text will mangle the accents.
3. Wait a few seconds. The first run downloads the OCR engine (roughly 10–15 MB), which your browser then caches, so later conversions start almost instantly. A per-image progress bar shows recognition as it happens.
4. Copy, edit, or download. Each image gets its own editable text box with a live word and character count. Fix any misread characters right there, then copy the text to your clipboard or download it as a .txt file. With multiple images, a Copy all button grabs everything at once.

That's the whole flow: no email address, no daily conversion limit, and no "premium" pages — because the work happens on your device, there's no server cost for us to gate.

OCR — optical character recognition — is the process of turning the pixels that look like letters into actual machine-readable characters you can select, edit, and search. A photo of a receipt or a screenshot of an error message is, to your computer, just a grid of colored dots; OCR is what bridges the gap between "a picture of words" and the words themselves.

Most free OCR online services work by uploading your image to their server, running recognition there, and returning the text. This tool takes a different approach: it loads Tesseract, a well-established open-source OCR engine, compiled to WebAssembly so it runs directly inside your browser tab. The trade-off is honest and simple. On the first use you download the engine and the language model — about 10–15 MB, cached by your browser afterwards — and in exchange, your images never travel across the network at all. Recognition speed depends on your device rather than someone's server queue, and the tool keeps working even if you go offline after the page loads.

That privacy property is not a small detail. The images people run through OCR are often exactly the ones they should be most careful with: screenshots of private conversations, photos of ID documents and contracts, invoices with bank details, internal dashboards. With this converter, none of that ever leaves your machine — there is no upload step to trust, log, or leak.

Being upfront about accuracy: OCR quality depends almost entirely on the input image. This tool shines on clear, straight, high-contrast printed text — screenshots, scanned documents, book pages photographed head-on, slides, and UI text. On a crisp screenshot of an article or a terminal window, expect near-perfect output. On a skewed, low-light phone photo of a crumpled receipt, expect to fix a few characters by hand — which is exactly why every result here is an editable text box rather than a locked download.

Concretely, recognition accuracy drops when an image has:

• Handwriting. Tesseract is trained mostly on printed typefaces. Neat block capitals sometimes come through; cursive rarely does. Treat handwriting as hit-or-miss, and expect to proofread.
• Low resolution. If the letters are only a few pixels tall, there isn't enough signal to recognize them. Text should be roughly 20 pixels high or more for reliable results.
• Rotation and skew. Slightly tilted text usually survives; text at a strong angle or photographed in perspective often doesn't. Straighten the photo first if you can.
• Busy backgrounds and low contrast. Text over photos, gradients, or watermarks confuses the segmentation step that decides what is a letter and what is background.
• Stylized and decorative fonts. Ornate display typefaces, extreme letter spacing, and outline-only text read far worse than ordinary body text.

A practical rule of thumb: if you have to squint to read the text in the image, the OCR engine will struggle too. When you control the source, re-capture at a higher resolution rather than upscaling a blurry image — upscaling adds pixels but not information.

The most common jobs this tool handles are format-specific versions of the same task. JPG to text usually means photos: a page from a book, a whiteboard after a meeting, a poster, a paper letter. JPEG compression slightly softens letter edges, so photos benefit the most from good lighting and a straight-on angle. PNG to text conversion usually means screen captures: PNG is lossless, so screenshots keep letters pixel-sharp, which is why screenshots are the best-case input for OCR and routinely come out nearly flawless.

Screenshot to text deserves its own mention because it's the case this tool is most optimized for in day-to-day use: someone sends you a screenshot of an address, a code snippet, an error message, or a paragraph you need to quote — and retyping it is tedious. Here you don't even need to save the file: take the screenshot, then paste it directly onto this page with Ctrl+V and copy out the text seconds later. WebP, GIF, and BMP images work the same way. The one thing this tool does not read is PDF files — for those, convert the pages to images first with the PDF to JPG tool and run OCR on the exported pages.

The recognizer supports English, Spanish, French, German, Italian, Portuguese, Hindi, Arabic, and Urdu. Each language uses its own trained model — that's what teaches the engine that German has ß and umlauts, that Portuguese uses ã and ç, that Hindi is written in Devanagari script, and that Arabic and Urdu flow right to left with letters that change shape depending on their position in a word. The model for your chosen language downloads on first use and is cached alongside the engine.

Because the language model shapes what the engine expects to see, running French text through the English model will strip or mangle accented characters, and Arabic text through a Latin-script model produces nonsense. If your image mixes languages, pick the dominant one and expect the other to need cleanup. Switching the language re-reads every image already in the queue automatically, so it costs one click to compare results between two candidate languages — useful for text in, say, Portuguese versus Spanish where you're not sure which will read better.

OCR recovers the characters in an image, and only roughly the layout around them — so it's worth knowing what the output will look like before you rely on it. The extracted text arrives as plain text: bold, italics, font sizes, and colors are not part of what the engine reads, so a heading comes out as an ordinary line of text. Line breaks land where lines end in the image, which means a justified paragraph from a book scan arrives as many short lines rather than one flowing paragraph. That's normal OCR behavior, and it's a ten-second fix: paste the result into the text cleaner's remove-line-breaks option, or just reflow it by hand in the editable output box.

Multi-column layouts — newspapers, academic papers, some slides — are the trickier case. The engine segments the page into blocks and usually reads each column as its own block in order, but tightly spaced columns can interleave, producing text that jumps between columns mid-sentence. When a document's column structure matters, the reliable approach is to crop each column into its own image and run them separately. Tables flatten in a similar way: you'll get the cell contents as space-separated words, line by line, not a structured table — fine for grabbing a few values, not a substitute for a real spreadsheet export. In short: expect the words to be right and the visual structure to be approximate.

Yes — and unlike most converters, the privacy claim here is structural rather than a policy promise. The recognition engine executes as WebAssembly inside this browser tab. Your image is read from disk (or from your clipboard) straight into browser memory, analyzed there, and the resulting text is shown to you — at no point is the image attached to a network request. The only downloads involved go the other direction: the engine and language files coming to your browser, once, from a public CDN. There is no account, no server-side storage, and nothing for a data breach to expose, which makes this a reasonable choice for exactly the sensitive material — contracts, IDs, medical letters, financial screenshots — that you shouldn't hand to an upload-based service.

• Students and researchers extracting quotes from photographed textbook pages, lecture slides, and archival scans instead of retyping them.
• Accountants and operations teams pulling totals, dates, and reference numbers out of receipts and invoices that arrive as photos.
• Developers and support engineers converting screenshots of error messages, logs, and config files back into searchable, pasteable text.
• Translators and language learners lifting foreign-language text out of images so it can be pasted into a translator or dictionary.
• Journalists and archivists digitizing printed clippings, flyers, and documents into editable text for quoting and indexing.
• Anyone sent a screenshot of an address, a Wi-Fi password, a tracking number, or a paragraph that really should have been sent as text in the first place.

After extracting, you often want to tidy the result — the text cleaner strips stray line breaks and double spaces in one click, and the character counter checks the length against platform limits before you post it anywhere.

• Feed it the original, not a re-share. Every round of messaging-app compression softens letters. The original screenshot beats the copy forwarded through three chats.
• Crop to the text. Trimming away photo backgrounds, UI chrome, and decorative elements with the crop tool gives the engine less to misinterpret.
• Shoot straight and bright. For photos of paper, aim the camera perpendicular to the page in even light. Shadows across the text are a common cause of garbled sections.
• Prefer dark-on-light. Black text on a white background is the trained ideal. Light text on dark generally still works; text over images often doesn't.
• Proofread numbers. The classic OCR confusions — 0/O, 1/l/I, 5/S, 8/B — matter most in amounts, IBANs, and serial numbers. The editable output box exists precisely for this pass.