URL Extractor from Image
Someone sent a picture of a link instead of the link. Upload the screenshot and this reads the address out of it with OCR, in your browser, and lets you fix any character it misread before you copy.
- You provide
- A PNG, JPEG or WebP screenshot or photo with a web address written in it.
- You receive
- The text the OCR engine read, plus the address candidates found inside it — all editable.
- Use it when
- Use it when an address only exists as pixels: a screenshot, a slide, a photo of a poster.
Drag a file here, or
PNG, JPEG or WebP · up to 12 MB
Recognition runs in this browser tab in a background worker. Your image is never uploaded. The English OCR model is downloaded once and cached by your browser.
What this tool cannot do
- A screenshot contains pixels. If it shows the words “click here” with a link behind them, the address is not in the file at all — no tool can recover it.
- Accuracy depends on the image. Small, blurry, low-contrast or heavily compressed text will be misread, which is why the recognised text stays editable.
- The confidence figure comes from the OCR engine and describes its certainty about the characters. It says nothing about whether the address exists, works, or is safe.
- English only. Files up to 12 MB and about 40 megapixels.
- This does not host images or create an image address. “Image to URL” sometimes means image hosting — that is a different kind of service.
The limitation worth understanding first
A raster image — a PNG, a JPEG, a screenshot — is a grid of coloured pixels. It contains no hidden data about what was underneath them.
So if the picture shows this:
Read the full report here
…the word "here" is just letters. The address it pointed to when it was a real web page is not in the image. OCR can read the word "here" perfectly and still give you nothing, because there is nothing to give.
This tool works when the address is written out where you can see it: https://example.com/report. That is the common case in screenshots of slides, terminal windows, printed material, and messages where someone pasted a link as a picture.
If you are stuck with a "click here" screenshot, the only routes are asking the sender for the link, or opening the original page and copying it there.
Everything happens in your browser
Recognition runs in a Web Worker in this tab, using a WebAssembly build of the Tesseract engine, both served from this site. Your image is never uploaded. You can watch the network panel while it runs and see for yourself.
The one thing that may cross the network is the English language model, about 10 MB, which your browser then caches. When the site owner has self-hosted it, even that is served from here. Either way it is a model file — your image is not part of that request.
Fixing what OCR gets wrong
OCR is very good and not perfect, and addresses are exactly where imperfection hurts: rn reads as m, l reads as 1, 0 reads as O. Get one character wrong in a domain and you have a different site.
So the recognised text is presented as an editable box, not a result. Correct anything that looks wrong and the address list updates. There is also a row of one-click fixes for the most common confusions, which apply to every occurrence in the text. None of them are applied automatically — quietly rewriting an address you are about to visit would be worse than showing you the wrong one.
Lines the engine was unsure about are flagged with their confidence, so you know where to look first.
Getting the best result
- 1
Use the largest version you have
A full-resolution screenshot reads far better than one that has been resized down, or sent through a messaging app that recompressed it. If you can retake it, do.
- 2
Crop to the address
Less surrounding text means fewer chances for the engine to run words together. Crop before uploading if you can.
- 3
Rotate if needed
A photo taken sideways will read poorly. The Rotate button turns the image and re-runs recognition.
- 4
Check the characters, then copy
Read the address back against the picture — especially digits and the letters l, I, O. Then copy it.
Frequently asked questions
Is my image uploaded to your server?
No. Recognition runs in your browser, in a background worker. The image never leaves the tab, and we never receive it. The only network request involved is for the English OCR model, which contains no part of your image and is cached by your browser after the first use.
It read the wrong characters. Is that a bug?
Usually not — it is the nature of OCR on small or compressed text. That is exactly why the recognised text is editable rather than final, and why per-line confidence is shown. Use a larger, sharper image if you can, and correct the rest by hand.
The screenshot shows a link but the tool found nothing.
Check whether the address is actually visible in the picture. If the image shows linked words such as “click here” or a button, the address is not in the file and cannot be recovered from it.
Can it give me a URL for the image itself?
No — that would be image hosting, which is a different service entirely. This reads text out of a picture you already have.
What does the confidence percentage mean?
It is the OCR engine’s own estimate of how sure it is about the characters it produced. It is not a check that the address exists, resolves, or is safe. Nothing on this site checks that.
Related tools
- You provide:
- A local PDF file.
- You receive:
- Clickable link targets with page numbers, separated from internal jumps, plus URL-shaped text if you want it.
- Use it when:
- Use it for reports, papers, brochures and slide exports where the references are hidden behind clickable text.
- You provide:
- Pasted text, or a local TXT, CSV, JSON or Markdown file.
- You receive:
- Every web address found, with its line number, in the order it appeared.
- Use it when:
- Use it when the links are sitting inside prose, an email, a chat log, a notes file or a data export.
- You provide:
- A list of URLs, one per line.
- You receive:
- The cleaned list, with a before/after view of every address that changed.
- Use it when:
- Use it before sharing links, publishing a reference list, or importing URLs that would otherwise carry campaign parameters.
- You provide:
- A public page address, or pasted HTML.
- You receive:
- Each image address with the attribute it came from, its width descriptor where srcset provides one, and its alt text.
- Use it when:
- Use it to inventory the images on a page, check which sizes are being offered, or collect asset URLs before a migration.
Guides for this tool
- How to Extract Links from a PDFClickable link annotations, addresses printed in the text, and internal jumps are three different things inside a PDF. Here is how to get each one — and why a scanned PDF gives you nothing.
- How to Extract a URL from an Image or ScreenshotOCR 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.