How to extract text from an image: screenshots, phone photos, and scanned PDFs

Every time someone searches for "extract text from image," they're in one of three situations. They have a screenshot with text they can't select. They have a phone photo of a document. Or they have a scanned PDF that won't let them search or copy a single word. The right approach is slightly different for each, so let's take them one at a time.

Situation 1: a screenshot

An error message. A code snippet. A quote from a page that disabled copying. A chart with a label you need for your notes. Screenshots are the most common thing people convert, and also the easiest.

Here's the fastest path with the converter on this site:

• Take the screenshot (or copy it to your clipboard).
• Open the converter and press Ctrl+V anywhere on the page. Pasting works — no file dialog needed.
• Hit convert. The text comes back in a few seconds, ready to copy.

One tip that saves pain later: save screenshots as PNG or WEBP, not JPG. Screenshots are flat colors on solid backgrounds, and lossless formats keep the letter edges razor sharp. It matters more than people expect. I wrote about the JPG angle in the JPG to text guide.

Situation 2: a phone photo of a document

Receipts are the classic case — you need the total for an expense report and the paper is already in your wallet. Bank statements, business cards, whiteboards, textbook pages: same idea.

The photo quality decides everything. Before you snap, do these three things and the extracted text will be dramatically cleaner:

• Hold the phone parallel to the paper. Straight on, not from a corner. Perspective bends the letters and OCR hates that.
• Light the paper, not the camera. Put the light source over the document. Shoot with the window behind you, not behind the paper.
• Fill the frame. Move in until the page fills the viewfinder. Extra desk around the edges is wasted pixels.

If the photo contains sensitive details — a card number, an address, a tax ID — use the mask button on the file before converting. Paint over anything you don't want seen; only the masked version is uploaded. This gets asked about a lot, so there's a dedicated FAQ answer about it.

Situation 3: a scanned PDF

A PDF that's really just a stack of scanned pages. You know the type: Ctrl+F finds nothing, selecting text selects nothing, and every word has to be retyped by hand. Old manuals, archive documents, contracts that were scanned and emailed around.

The converter handles PDFs the same way it handles images, except the work happens page by page:

• Drop the PDF into the converter.
• Each page is read separately and the results come back in order.
• Multi-page PDFs are no problem, as long as the file stays under 20 MB.

After conversion the text is plain and searchable — paste it into a document, save the .txt, or index it. That's the whole point of the exercise.

What to do with the result

Every converted file gets its own result box on the page. From there you have three options:

• Copy — straight to your clipboard, ready for notes, email, or a search bar.
• Download .txt — a plain text file named after your original, useful for archiving or feeding into other tools.
• Edit inline — the result box is a real textarea, so you can fix the odd misread character right there before you copy.

Where it gets less reliable

Worth saying plainly: some inputs convert worse than others, and no setting on the page will fix that. Heavy handwriting. Text at a steep angle. Glare on glossy paper. A screenshot saved at thumbnail size. These fail or come back messy, and the fixes are about the image, not the tool. The accuracy article covers the full list of what helps and what's a waste of time.

The workflow that actually works

For the majority of people, the fastest routine is: screenshot or photo → paste or drop → convert → copy. If you're doing this more than a couple of times a week, that sequence saves real hours.

Ready when you are. The converter is on the homepage, handles 100+ languages and 30+ formats including PDF, and deletes files from the server as soon as each task finishes.