Skip to content

Extract Text From an Image (OCR): Screenshots, Photos, Scans

The Quick Answer

To extract text from an image, open Image to Text (OCR), add the picture (choose a file, drop it on the page, or paste a screenshot with Ctrl+V or ⌘V), click Extract text, check the words the engine was unsure about, and copy the result or download it as a .txt file.

The recognition runs in your browser with Tesseract, an open-source OCR engine, so the image is never uploaded. The tool reads English, German, French and Spanish; choose the language of your text in the Text language menu. Screenshots and clean printed pages usually come out close to perfect; phone photos of paper need a little more care, which is what most of this guide is about.

Step by Step: From Picture to Editable Text

  1. Add one or more images. PNG, JPG, WebP, GIF and BMP work in every browser. You can add several images at once and extract them together with Extract all.
  2. Look at the preview. The preview shows the image as it will be read. Rotate it with the arrow buttons if it is sideways. Phone photos are turned upright automatically using the orientation the camera stored in the file.
  3. Pick a layout. Automatic suits pages and documents. Choose Single column, Single block, Single line or Sparse text when the picture is one of those (the table below helps).
  4. Clean up only if needed. Grayscale, Contrast, Invert and a black-and-white Threshold are there for faded, colored or dark pictures. Leave them off for ordinary screenshots.
  5. Extract. The first run downloads the OCR engine (about 4 MB) and the model for the language you picked (English about 5 MB; German, French and Spanish 1–3.5 MB) from writepadpro.com. Your browser caches them, so later runs start much faster.
  6. Review and export. Read the confidence score, click each uncertain word to jump to it, fix what is wrong, then copy the text or download it. With several images, “Copy text of all images” and “Download all as one .txt” combine them, with a header line naming each file.

What Happens to Your Image

Knowing what the tool does with a picture makes it easier to choose the right settings.

1. Resizing to a readable size

Tesseract struggles with tiny letters and gains nothing from huge ones. The tool enlarges small images (long side under 1,000 pixels) toward 1,800 pixels, up to three times, and reduces very large ones to at most 4,000 pixels on the long side and 12 megapixels. These are the actual results of that rule for some common sizes:

ImageSize in pixelsWhat the tool reads
Small cropped screenshot800 × 600Enlarged 2.25× to 1800 × 1350
One line cut from a screenshot600 × 200Enlarged 3× to 1800 × 600
Full-HD screenshot1920 × 1080Unchanged
Tall phone screenshot1170 × 2532Unchanged
12-megapixel phone photo4032 × 3024Reduced slightly to 4000 × 3000
24-megapixel camera photo6000 × 4000Reduced to 4000 × 2667

Enlarging cannot add detail that was never captured. A blurry 300-pixel crop becomes a bigger blurry crop, so take a sharper picture when you can.

2. Optional clean-up

Grayscale removes color. Contrast spreads light and dark apart. Invert swaps them, which matters because Tesseract expects dark text on a light background. Threshold turns every pixel pure black or white: “Auto (Otsu)” picks the brightness level that best separates text from background, and “Manual” lets you set it yourself.

3. Recognition

The prepared image goes to Tesseract (through tesseract.js) running as WebAssembly in a background worker, using its LSTM neural-network recognizer and the model for the language you chose (English, German, French or Spanish). The layout setting tells it what kind of page to expect.

Choose the Right Settings for Your Picture

PictureLayoutClean-up
Screenshot of an article or documentAutomatic or Single columnNone; keep “Enlarge small images” on
Screenshot of an app, form or chatSparse textNone
Dark-mode screenshot (light text on dark)Same as aboveInvert
Photo of a book pageSingle columnGrayscale and a little contrast
Page with shadows or yellowed paperAutomaticThreshold: Auto, then Manual if letters break up or fill in
ReceiptSingle columnGrayscale and contrast; Threshold if the print is faint
Serial number, code or one titleSingle lineCrop tightly first
Sign, label or diagramSparse textContrast if the background is busy

Switch the preview to “As it will be read” after each change. If you can barely read the processed preview yourself, the engine will struggle too.

How to Take a Photo That OCR Can Read

  • Shoot straight on. Hold the phone parallel to the page. Tilted text lines are one of the main causes of poor results; Tesseract’s documentation recommends rotating pages so text lines are horizontal.
  • Use even light. Daylight from a window beats a ceiling lamp that throws your phone’s shadow across the page. Turn off the flash if it creates a glare spot.
  • Fill the frame with text. Get close enough that the text covers most of the photo, then crop away the desk, fingers and neighboring page. Stray marks at the edges can be read as characters.
  • Keep letters big enough. The tool page suggests text at least 20 pixels tall. For flatbed scans, Tesseract’s guide on improving output quality recommends scanning at no less than 300 dpi (checked on 2026-10-14).
  • Prefer a screenshot over a photo of a screen. A photo of a monitor adds glare, moiré patterns and perspective. A real screenshot is pixel-perfect.

Reading and Fixing the Result

The confidence score

The score is Tesseract’s average certainty across all recognized words, from 0 to 100. As the tool page explains, above about 85 the text is usually close to perfect, between 65 and 85 you should expect some mistakes, and below that it is worth improving the image. A high average can still hide one wrong digit, so always check names, numbers and amounts against the picture.

Uncertain words

Words below 70 percent confidence are listed under the result (up to 40 distinct words). Click one and the tool selects it in the text so you can correct it. Typical OCR confusions are 0 and O, 1, l and I, “rn” read as “m,” and a period read as a comma.

Joining broken lines

OCR keeps every printed line break, which is awkward when you paste the text into a document. “Join broken lines” merges the lines inside each paragraph and keeps blank lines between paragraphs. Here is what it actually does with a short note:

Recognized text:

The meeting is moved to Thurs-
day at 3 pm in room B12.
Please bring the well-
known agenda.

Thanks,
Sam

After “Join broken lines”:

The meeting is moved to Thurs-day at 3 pm in room B12. Please bring the well-known agenda.

Thanks, Sam

Two things to notice. The hyphen at a line end is kept and the gap closed, because the tool cannot tell a split word (“Thurs-day”) from a real compound (“well-known”), so delete the leftover hyphens yourself. And a sign-off on separate lines is joined too, so use the button on body text and fix short lines by hand. Press Ctrl+Z to undo a join.

For more control over line breaks and spacing, paste the text into Remove Line Breaks, which can also rejoin words hyphenated across lines, or Remove Extra Spaces; our guide to removing line breaks covers the options. If you need a word count for the extracted text, the Word Counter gives it instantly.

Four Common Jobs, Done Well

Copying a quotation from a textbook photo

Crop to the paragraph you need, use Single block, and compare the result word by word with the page before you paste it into an essay. A quotation has to be exact, and OCR errors in quotations are easy to miss. Note the page number while the book is in front of you.

Turning receipts into a list of expenses

Photograph each receipt flat on a dark table, add them all, and use Extract all. Check every amount against the paper: a 5 read as an S or a missing decimal point changes the total. “Download all as one .txt” gives you one file to work from.

Grabbing an error message from a screenshot

Crop to the message and choose Single block or Sparse text. Code and file paths contain characters OCR confuses, such as l and 1 or O and 0, so check those before you search for the error.

Digitizing a printed letter

Use Automatic, join broken lines in the body, then fix the address block and the sign-off by hand. If the letter runs over several pages, add every page and extract them together.

Limits, and When to Use Something Else

  • Four languages. English, German, French and Spanish each have their own model; pick the one that matches your text, because a mismatched model misreads letters such as ß, é or ñ. Other Latin-script languages are often partly readable, and scripts such as Chinese, Arabic or Cyrillic are not recognized.
  • Handwriting. Tesseract is trained on printed text. Neat block capitals sometimes work; cursive and quick notes usually do not.
  • Tables and columns. They come out as lines of text without the original structure. Crop one column at a time if the order gets mixed up.
  • iPhone HEIC photos. Many browsers cannot open HEIC files. If the tool says the image could not be opened, export it as JPG or PNG first, or take a screenshot of it.
  • PDFs. Try the PDF to Text Converter first. If the PDF has a text layer, extraction is exact and quick. Only a scanned PDF needs OCR: export or screenshot its pages as images and add them here.
  • Size. Images up to 50 MB can be added.

Troubleshooting

The result is gibberish

Check the preview first. Sideways images need rotating, and light text on a dark background needs Invert. If the page has a colored or patterned background, try Threshold: Auto.

The result is empty or only a few words

The layout is probably wrong for the picture. Switch to Sparse text for screenshots of apps and signs, or crop tightly and use Single block. Very small text needs a sharper source image.

Letters are broken or filled in

The threshold is too high or too low. Set Threshold to Manual and move the level a little in each direction while watching the preview.

The first extraction takes a while

The engine and model download once and then start up. On slower devices, starting the engine still takes a few seconds even when the files are cached.

Your Images Stay on Your Device

Many online OCR services upload your image to their servers. Here, the engine and the language model you choose are the only downloads, and they come from this site. Your images, the extracted text and the results stay in the page and are discarded when you close it. That makes the tool suitable for receipts, letters, ID documents and confidential screenshots.

Frequently Asked Questions

Is my image uploaded when I extract text?

No. Tesseract runs in your browser as WebAssembly, so the image and the recognized text never leave your device. The only downloads are the OCR engine and the language model you choose, which come from writepadpro.com and are then cached by your browser.

Why does a screenshot work better than a photo?

A screenshot contains the exact pixels of the text: straight lines, even contrast and no blur. A photo adds perspective, shadows, glare and camera noise, all of which make letters harder to recognize. When you can take a screenshot instead of photographing a screen, do.

What do the layout options do?

They tell Tesseract what to expect: Automatic for whole pages, Single column for one column of text, Single block for one paragraph, Single line for a code or title, and Sparse text for scattered text such as app screenshots, signs and diagrams. The wrong layout can leave out text or merge lines.

Why does Join broken lines leave hyphens like “Thurs-day”?

Because the tool cannot know whether a hyphen at the end of a line splits one word or belongs to a compound such as “well-known.” It keeps the hyphen and closes the gap so that compounds stay correct. Delete the hyphens in split words yourself.

Can I extract text from a scanned PDF?

Not directly. Try the PDF to Text Converter first, because many PDFs contain real text. If it reports pages with no selectable text, export or screenshot those pages as images and add them to Image to Text.

Which languages can the OCR tool read?

English, German, French and Spanish. Choose the language of your text in the Text language menu so the matching Tesseract model is used; each model also reads digits and common punctuation. Other Latin-script languages are partly readable, but accented letters may be missed, and non-Latin scripts are not recognized.

Try These Tools

Related Articles