How to convert an image to text
Drop an image onto the page, choose a file, or — for the fastest path — just take a screenshot and paste it in with Cmd/Ctrl+V. Click "Extract text" and the engine reads the image and returns the text it finds in an editable box a few seconds later, depending on the image size and how much text is in it.
From there you can fix anything it got wrong, copy the whole thing to your clipboard, download it as a .txt file, or save it straight into an OmniCanvas note. Nothing about the flow requires you to sign up first — the free path is genuinely complete on its own.
What this is actually good at — and what it isn't
Be honest with yourself about the input and you'll get good results. Clean, high-contrast printed text — a screenshot of an article, a scanned book page, a typed document, a clear photo of a sign or menu — comes out close to perfect. That's the sweet spot OCR engines are built for.
It gets meaningfully worse on handwriting, low-resolution or blurry photos, low-contrast text (light gray on white, text over a busy photo), skewed or rotated images, and dense multi-column layouts where the reading order isn't obvious. The engine will still produce an answer in those cases, but treat it as a rough first draft you proofread, not a finished transcript. This tool doesn't fake confidence it doesn't have — it just gives you the raw result and lets you fix what's wrong.
Why nothing is uploaded
Most free OCR tools work by sending your image to a server, running recognition there, and sending text back. That's a reasonable architecture, but it means a third party's server briefly holds whatever was in the image — which is a real concern if you're scanning IDs, financial statements, medical paperwork, or anything with a name and number on it you wouldn't want sitting in someone else's logs.
This tool takes a different path: the OCR engine itself — an open-source project called Tesseract, compiled to WebAssembly — downloads once into your browser and does the actual reading locally. Your image is decoded, processed, and discarded entirely on your machine. There's no network request carrying the picture anywhere, so it works exactly the same on a private document as it does on a public one.
Screenshot to text, specifically
A lot of what people actually want isn't a scanned document at all — it's a screenshot. A tweet you can't select the text of, an error message in a video call, a quote in an image someone sent you, a paragraph in a PDF viewer that blocks copy-paste. Paste support exists specifically for this case: screenshot it, Cmd/Ctrl+V, and the text is out a few seconds later without ever saving a file to disk.
It works the same way on a phone photo of a printed page, a whiteboard after a meeting, a business card, or a street sign — anything where the text exists as pixels rather than selectable characters.
Getting better results
A few things reliably improve accuracy: crop the image tight to just the text you care about rather than a whole cluttered screen, use the highest resolution version you have rather than a compressed thumbnail, and straighten the image first if the text is at an angle. Good lighting and a flat page (no curl or shadow) make a large difference on photographed documents.
If the result still has errors after that, it's very likely the image itself is a hard case for OCR — not a bug in the tool. That's exactly why the extracted text lands in an editable box instead of a locked result: fix the handful of words it missed and move on.