How this image to text converter works
The page loads Tesseract.js version 4.1.1 from the cdnjs content delivery network. It is a WebAssembly port of the Tesseract OCR engine, and the recognition runs in a web worker inside your browser. Your image is never uploaded to our server.
What the browser does download, the first time you press Convert, is the engine itself and the English language model (eng.traineddata, about 11 MB compressed) from tessdata.projectnaptha.com, the library's default location. Later conversions reuse the cached files, so the second run is faster than the first. In our test a 900 by 220 pixel screenshot of two printed lines took about six seconds from click to result and came back without a single wrong character.
What decides the accuracy
Image quality
Tesseract was trained on clean printed text. A sharp, evenly lit, straight photo with dark letters on a plain light background reads well. Blur, shadows, skew, low contrast, coloured or patterned backgrounds and very small letters each add errors, and they compound. Crop to the passage you need, photograph pages flat rather than at an angle, and use the original screenshot rather than a resaved copy, because JPEG artefacts around letters confuse the engine.
Printed text versus handwriting
Handwriting is unreliable here. The English model knows printed letterforms; neat block capitals sometimes come through, cursive almost never. Treat the result for handwritten notes as a rough draft to retype, not a transcript to correct.
Language and script
Only the English model is loaded. Other languages written in the Latin alphabet are partly recognised, but accented letters are often dropped or replaced. Arabic, Chinese, Cyrillic, Devanagari and other scripts are not recognised at all. Tesseract has models for many other languages, but this tool does not load them.
Layout
The output is plain text. Columns, tables and forms are read roughly top to bottom and left to right, so a two-column page can interleave its columns, and mathematics, symbols and tables lose their structure.
What the tool does and does not do
| Formats |
Language |
Output |
Processing |
| JPG, PNG, BMP, GIF (non-animated), WebP. No TIFF, no PDF. |
English model only |
Plain text in the result box, copied with the Copy button. No file download. |
In your browser, with Tesseract.js. Engine and model are fetched from a CDN on first use. |
Good candidates are screenshots of error messages or chat threads, printed letters and receipts, book and magazine pages, slides, and signs. Once you have the text, the Text to Speech tool reads it back to you, which is a quick way to catch recognition errors, and the Word Counter gives you its length.