Image to Text: Free OCR
This tool reads the text out of an image: a screenshot, a scanned page, a photo of a document, a receipt. Paste or drop the image and the text comes back ready to copy, with the words the engine was unsure about marked so you know what to double-check. It runs entirely in your browser, which matters more here than anywhere else on this site: the things people OCR, contracts, invoices, ID documents, are exactly the things that should never be uploaded to a stranger's server.
Everything happens in this tab. Your files are never uploaded, because there is nowhere here to upload them to.
How to use it
Drop or paste your image
Ctrl+V works directly: screenshot something, come here, paste. For photos of paper, the straighter and closer the shot, the better the reading.
The engine reads it on your device
The first run loads the OCR engine and your language's data into the browser, a few megabytes, once. Then it finds the text, straightens slightly rotated pages by itself, and reads.
Copy the text, check the marked words
The result appears with a copy button and a download option. Words the engine doubted get a dotted underline. Check those, trust the rest, and you are done proofreading in a tenth of the time.
How this runs without a server
Every other tool in this category works the same way: you upload your image, a GPU in a data centre somewhere runs the model, and the result comes back. That architecture is why they all have credits, watermarks and sign-up walls. Someone has to pay for that GPU, and it is metered by the second.
This page does it differently. When you drop in an image, your browser loads the AI model itself and runs it locally. If your browser supports WebGPU, the model runs on your own graphics hardware, which is typically five to sixty times faster than the CPU fallback. If it does not, we fall back to WebAssembly, which is slower but works everywhere.
Two things follow from that, and they are the entire reason this site exists. First, your photo never goes anywhere: it is read into a canvas in your tab and stays in your device's memory until you close it. There is no upload endpoint on this domain to send it to. Second, our cost per image is zero, so the free tier is not a loss-leader with a trapdoor in it. It is just what the thing costs to run, which is nothing.
The model file is cached by your browser after the first download. That is the one wait you have to sit through, and it only happens once per browser. After that the tool works with your network disconnected, which is a decent way to prove to yourself that nothing is being sent.
Checking that for yourself
You do not have to take our word for it, and you should not. Open your browser's developer tools, go to the Network tab, and use the tool. You will see the model coming down. You will not see your image going up, because it never does. Then turn your wifi off: once the model is loaded, the tool keeps working with no network at all, which is not something we could fake. We walk you through both checks here.
Your photo stays on your device.
Not "deleted after 24 hours". Not "encrypted in transit". Never sent. Here is how we prove it.
Frequently asked questions
Is my document uploaded to read it?
No, and for this tool that is the entire point. OCR sites normally receive your document, process it, and promise to delete it. This page has no server to receive it: the OCR engine is downloaded to your browser and reads the image in your tab. A contract, a passport scan, a medical letter, none of it leaves your machine.
How accurate is it?
On clean printed text, screenshots and good scans: very, typically above 95%. On phone photos it depends on the photo: sharp, straight-on and well-lit reads well, while blur, strong angles and shadows cost accuracy quickly. It does not read handwriting, and we would rather tell you that here than let you find out after pasting.
Which languages does it support?
The picker covers seventeen languages including English, Spanish, French, German, Portuguese, Russian, Arabic, Hindi, Japanese, Korean and Chinese. Each language's data loads once, a few megabytes, and is cached in your browser afterwards. It reads one language at a time, so pick the one the document is mostly written in.
What do the dotted underlines mean?
The engine reports how confident it is about every word, and we surface the doubtful ones instead of hiding them. OCR is never 100%, whatever any tool claims. Marking the uncertain words turns proofreading from re-reading everything into checking a handful of flagged spots.
Can it read text from a PDF?
Not directly yet, it takes images. The workaround that works today: screenshot the PDF page and paste it here. A proper PDF-to-text flow is on our list.
Why is the first run slower than the rest?
The engine and language data have to reach your browser before anything can be read there, which is a few megabytes, once. Your browser keeps them, so from the second use the tool starts immediately, and it keeps working offline.