How to Extract a URL from an Image or Screenshot
OCR can read an address that is visibly written in a picture. It cannot recover a link hidden behind the words "click here" — here is the difference, and how to get the best result when OCR is the right tool.
When OCR is the right tool
It works well for:
- Screenshots of slides, terminal windows or documentation where the address is printed.
- Photographs of printed material — a poster, a business card, a conference slide.
- Messages where someone pasted a link as an image instead of as text.
- QR-code posters that also print the address underneath.
It cannot help when the address is not visible in the picture. In that case the only routes are asking whoever sent it for the link, or opening the original page and copying it there.
How to use it
Open the image tool and drop the file in. Recognition starts immediately, in a background worker in your browser — the image is never uploaded.
What comes back is two things: an editable text box with everything the engine read, and a result table of the addresses found inside it. Editing the text updates the table.
Getting a good result
- 1
Use the biggest version you have
A full-resolution screenshot reads far better than one that has been resized or sent through a messaging app that recompressed it. If you can retake the screenshot, do — it is the single biggest factor.
- 2
Crop to the address
Less surrounding text means fewer chances for the engine to run words together or mistake layout for content.
- 3
Rotate if the photo is sideways
The Rotate button turns the image 90° and re-runs recognition on the rotated version.
- 4
Fix the characters that matter
Read the address back against the picture, paying attention to digits and to the letters
l,IandO. The one-click fixes handle the most common confusions across the whole text.
| Reads as | Often actually | Where it bites |
|---|---|---|
| `l` (lowercase L) | `1` (one) | `example1.com` vs `examplel.com` |
| `O` (capital o) | `0` (zero) | Short link codes, tracking ids |
| `rn` | `m` | `rnodern.com` vs `modern.com` — genuinely hard to see |
| `I` (capital i) | `l` (lowercase L) | Depends entirely on the typeface |
| `5` | `S` | Condensed or low-resolution text |
| `8` | `B` | Same |
The rn/m confusion deserves attention: at small sizes the two are close to identical, and it is a known technique for making one domain look like another. If an address matters, check it character by character against the image.
What the confidence figure means
The tool shows the engine's own confidence, overall and per line, and flags lines below 80%.
That number describes how certain the engine is about the characters it produced. It does not mean:
- the address exists;
- the address resolves;
- the destination is safe.
Nothing on this site checks any of those. A perfectly recognised address to a malicious site scores 99%.
Privacy
Recognition runs in your browser using a WebAssembly build of the Tesseract engine. The image is never uploaded.
The English language model — about 10 MB — is downloaded once and then cached by your browser. Depending on how the site is deployed, it comes from this site or from a public CDN; either way that request contains no part of your image. The tool tells you which is in use.
Frequently asked questions
Is my image uploaded?
No. Recognition happens in your browser, in a background worker. You can confirm it by opening your browser’s network panel while a file is processed — the only request is for the language model, and only the first time.
It read the address wrong. Is that a bug?
Usually it is the nature of OCR on small or compressed text. That is exactly why the recognised text is editable rather than final. Try a larger, sharper image, crop tighter, and correct the rest by hand.
Can it read a QR code?
No — this is text recognition, not QR decoding. Most posters print the address underneath the code, and that text reads fine. For the code itself, your phone’s camera app is the right tool.
Can it give me a URL for the image itself?
No. That is image hosting, which is a different kind of service. "Image to URL" means both things, which is why this tool says plainly which one it is.
Sources
- Tesseract.js — browser OCR engine — Tesseract.js
- Tesseract OCR — documentation on image quality — Tesseract OCR