Skip to content
Screenshot2Text

Screenshot2Text

Photo to Text

A phone camera is often the fastest way to grab words you cannot type: a whiteboard after a meeting, a posted notice, a slide on someone else's laptop, a menu, or a page from a book you do not want to rekey. Photo to text, sometimes called picture to text, is OCR aimed at those camera shots rather than a clean desktop screenshot.

Screenshot2Text accepts a photo in this browser tab and returns selectable text. Save the picture, drop it on the zone below, wait for recognition, then copy. We do not upload the photo for OCR. The first run may briefly fetch small recognition files to your device (cached after); those requests do not include the picture.

Camera images are harder than screenshots. Lighting falls across the page, the phone tilts, and JPEG compression softens thin strokes. Hold the camera parallel to the surface, fill the frame with the paragraph you need, and watch for glare on glossy paper or a glass whiteboard. Printed posters, typed letters, and sharp photos of a screen convert more reliably than scribbled notes. Messy handwriting is frequently incomplete; we do not promise a usable transcript from a notebook page.

A picture of a picture still works when the letters are large and in focus. If you can take a direct screenshot instead of photographing a monitor, do that. You avoid moire and extra noise. Drop one photo or several; a PDF works the same way. Use this page or open the main free tool on the homepage to start a picture-to-text conversion.

Or open the main free tool

OCR runs in the browser. We don't upload your file for OCR.

Drop a screenshot, image, or PDF here

PNG, JPG, WEBP, GIF, BMP, or PDF · several files at once · tap to browse

Or paste a screenshot with Ctrl+V or ⌘V

Nothing to copy yet

A real run: a receipt photo

Here is one real example. The picture below is num01-receipt from our OCR accuracy test, saved as a simulated phone photo: the screenshot was tilted, softened, lit unevenly, given some noise, and saved as a JPEG. It is not a real camera photo. It is 417 by 464 pixels and 44 KB. We dropped it on screenshot2text.com on October 8, 2026 and copied the output as it came back.

Slightly tilted, soft photo of a Corner Market receipt listing milk, bananas, sourdough bread, eggs, and coffee beans, with subtotal $26.63, tax $1.60, total $28.23, and a VISA card ending 4821
Input: num01-receipt as a simulated phone photo, from our accuracy set.
Raw OCR output
CORNER MARKET
Order #20483 09/21/2026 4:17 PH
whole milk 1 gal sa.20 [7
Bananas 2.31 1b $1.36 i
Sourdough bread $5.99 a
Eggs large 12 ct $3.49
Coffee beans 12 oz $11.50 i
Subtotal $26.63 3
Tax 6% $1.60
Total $28.23 JB
VISA 4821

The totals came through: subtotal $26.63, tax $1.60, and total $28.23 are all right, and so are four of the five item prices. The milk price, $4.29, came back as "sa.20 [7". Stray marks such as "i", "a", "3", and "JB" were added after several lines. PM became PH, lb became 1b, Whole lost its capital W, and the masked card digits **** were dropped.

That is 25 wrong characters out of 218, a character error rate of 11.47%, the same score this image got in the run 3 CSV. The same receipt as a clean PNG scored 3.21%. With a receipt, check every price against the photo before you use it.

How phone photos did in our test

In run 3 of our OCR accuracy test, on September 25, 2026, we read the same 22 screenshots twice: once as clean PNG files and once as simulated phone photos of the screen. The photos are simulated, not taken with a real camera, so treat this as a direction, not a promise for your phone.

As PNG, the 22 screenshots had a mean character error rate (CER) of 0.65% and 12 of them read with no errors. As simulated phone photos, the mean CER was 9.74% and 3 read with no errors. Word recall, the share of words that came back right, dropped from 96.82% to 85.13%. The figures come from the run 3 CSV.

Mean character error rate by kind of screenshot, PNG versus simulated phone photo, run 3, September 25, 2026
Kind of screenshotPNGSimulated phone photo
Paragraph text (5)0.09%0.87%
Receipts, invoices, and price lists (4)0.92%4.68%
Code and terminal (4)2.03%10.69%
App and settings screens (5)0.41%12.87%
Chat conversations (4)0.00%21.03%
All 22 screenshots0.65%9.74%

Plain paragraphs held up well as photos. Chat bubbles and app screens, with small type and lots of short labels, lost the most. Three habits close most of that gap.

If the text is on a screen you can reach, take a real screenshot instead of a photo. On our set that is the difference between 0.65% and 9.74%.

Get close enough that the letters fill the frame, and hold the phone parallel to the page or screen. Small, tilted type is what the photo versions of the chat and app screens struggled with.

Keep it sharp and evenly lit. Tap to focus, hold still, and move so no glare or shadow falls across the words. Then check the output against the photo before you paste it.

What it can't read well

On October 8, 2026 we took one clean paragraph screenshot from our test set, para01-article, which reads with no errors, and changed it one way at a time. Each version went through the live site once. All scores and images are in the limits check CSV. It is one image per row, so treat these as examples, not averages.

Character error rate for one paragraph screenshot after a single change, read once on October 8, 2026
Version of the screenshotCharacter error rate
Clean screenshot, no change0.00%
Blurred, radius 2 pixels47.81%
Blurred, radius 3 pixels86.22%
Heavy tilt, about 40 degrees38.41%
Shrunk to 30%, capital letters about 3.6 pixels tall44.47%

Blur did the most damage. At radius 2 nearly half the characters came back wrong, and at radius 3 not one word of the original survived. A steep angle broke most words too. The mild tilt in our simulated phone photos was much easier: those 22 photos averaged 9.74% in run 3.

Tiny text fails once the letters get only a few pixels tall. At 30% size most words were lost, while 40% still came back with 4 wrong characters.

Handwriting is not in our test set, so we have no number for it. A neat handwriting-style font read with 0.84% errors, but real handwriting changes from letter to letter. Expect gaps and check every line.

What helps most: hold still and tap to focus, shoot straight on, and get close enough that the letters fill the frame.

FAQ

Can I turn a phone photo into text?
Yes. Save the picture to your device, drop the file here, and copy the result once recognition finishes. The same reader accepts screenshots and PDFs if you have those instead. Recognition runs in this tab with open-source OCR, so the photo is not sent to our servers for the read.
Why does my picture-to-text result look messy?
Tilt, shadow, glare, and heavy compression all confuse OCR. Retake the photo straight-on, fill the frame with one block of type, and avoid photographing a dim screen. Hold the camera parallel to the page when you can. Printed text still reads far better than handwriting, and a photo of a monitor is usually worse than a real screenshot of that same screen.
Do you keep my photo?
No. Recognition runs in the browser. We do not upload the photo for OCR and we do not store it or the extracted text on a server. The first run may fetch small recognition files to your device; those requests do not include the picture. The privacy page lists everything the site does collect.
When should I use photo to text instead of another page?
Use this page for camera shots of paper, whiteboards, menus, posted notices, and slides on someone else's laptop. A clean desktop screenshot belongs on PNG to text or the homepage. A multi-page scanned PDF belongs on PDF to text. Casual snapshots and document scans share this reader; the entrance page just matches how you searched.
Is photo to text free?
Yes, and it stays free. There is no account and no daily limit on how many photos you read. Ads on the page cover hosting. Drop one photo or several in the same batch.