Image to Text (OCR)
Extract text from an image, screenshot or scanned page with OCR. Paste a screenshot with Ctrl+V, drop or select an image, and copy the recognized text. Reads English, Japanese (including vertical Japanese), Chinese, Korean, major European languages, Vietnamese, Thai, Arabic and Hindi. Recognition runs in your browser with Tesseract, so your image is never uploaded. Only the OCR engine and language data are downloaded the first time.
What does Image to Text do?
This tool reads the text in an image and turns it into text you can edit and copy, so you do not have to retype it. It works with screenshots, photos of documents, scanned pages, slides and signs. It uses Tesseract, the open-source OCR engine, compiled to WebAssembly (tesseract.js), and runs entirely in your browser.
Supported languages:
| Language | Use it for |
|---|---|
| Auto (English / Japanese) | English or Japanese when you are not sure of the direction. Reads the image as English + Japanese, and if the result looks poor, also as vertical Japanese and keeps the better one |
| English | English text, numbers and code |
| Japanese, horizontal | Japanese written left to right (横書き) |
| Japanese, vertical | Japanese written top to bottom, such as novels and newspapers (縦書き) |
| English + Japanese | Japanese text mixed with English words |
| Chinese, Simplified / Traditional | Chinese written left to right (简体中文 / 繁體中文) |
| Chinese, Simplified / Traditional, vertical | Chinese written top to bottom (竖排 / 直排) |
| Korean | Korean (한국어) |
| French, German, Spanish, Portuguese, Italian, Russian | Text in these languages, including accented letters and Cyrillic |
| Vietnamese, Thai, Arabic, Hindi | Text in these languages. Arabic is shown right to left |
When you open the tool for the first time, the language is Auto if your browser is set to Japanese, and English otherwise.
How to use
- Choose the Language of the text in the image.
- Paste a screenshot with Ctrl+V (Cmd+V on a Mac), drop an image onto the drop area, or click it to choose a file.
- Wait for recognition to finish. The first time, the OCR engine (about 4 MB) and the language data (about 1–3 MB per language) are downloaded; the progress bar shows how far it has got.
- The text appears in Recognized text. Fix any mistakes in the box and click the copy button.
Turn on Highlight to draw boxes around the recognized words or lines on the image. The color shows how confident the engine is: green (80% or more), orange (60–80%) and red (below 60%). Hover over a box to see its text and confidence, which helps you find the words to check.
If the result is poor, change the language and click Recognize again. With Auto, the text below the result shows which reading was used, for example Auto: Japanese (vertical).
Tips for better accuracy
- Resolution: the text should be at least about 20 pixels tall. Small images are enlarged automatically, but a sharper original works better. Zoom in before taking a screenshot of small text.
- Contrast: dark text on a plain, light background works best. Shadows, textured backgrounds, glare and colored text reduce accuracy.
- Straight and flat: photograph pages straight on, without tilt or curved lines, and crop away everything that is not text.
- One language and direction: choose the language of the text rather than Auto when you know it, and a vertical language for 縦書き or vertical Chinese. A page that mixes vertical and horizontal text is read more reliably if you crop each part separately.
- Columns and tables: text is read block by block; for complex layouts, crop one column or table at a time.
Limitations
- Handwriting is not supported well. Tesseract is trained on printed text, so handwritten notes usually give poor results.
- Decorative fonts, very large or very small text, furigana (ruby) and text on photos or illustrations may be misread.
- Only the text is extracted. Formatting such as fonts, colors and table structure is not kept.
- Images are scaled to at most 4,000 pixels on the long side.
FAQ
Is my image uploaded to a server?
No. The image is processed in your browser and never leaves your device. Only the OCR engine and the language data are downloaded from the jsDelivr CDN, at fixed versions that are checked against known hashes before use. The image is not saved with the tool; only the selected language and the last recognized text (up to 20,000 characters) are kept.
Why is the first run slow?
The OCR engine and language data (about 5–11 MB in total) are downloaded the first time. Your browser caches them, so later runs start much faster. Auto may read the image twice (horizontal and vertical), so it takes longer than choosing a language.
How does Auto choose between horizontal and vertical?
Auto first reads the image as English + Japanese written horizontally. If the mean confidence is below 60%, or the result looks like vertical text read the wrong way (many lines of only one or two characters), it reads the image again as vertical Japanese and keeps the result with the higher mean confidence. Auto only covers English and Japanese; choose another language from the list for other text.
Can I paste a screenshot directly?
Yes. Take a screenshot to the clipboard (for example Win+Shift+S on Windows or Cmd+Ctrl+Shift+4 on a Mac), click this tool, and press Ctrl+V / Cmd+V.
Why are there spaces between Japanese characters?
Tesseract separates every character with a space. This tool removes spaces between Japanese and Chinese characters automatically, while keeping the spaces between English words. Spaces in Korean are kept, because Korean separates words with spaces.
Can it read PDFs?
Not directly. Take a screenshot of the page or export it as an image, then paste or drop the image.