Skip to content
StemStudio
AI Image OCR

AI Image OCR

The image is processed in your browser; the OCR runtime and recognition data are fetched on the first run or when the cache is unavailable.

Input

Input

Supported: PNG / JPEG / WebP / BMP / GIF (up to 20 MB, 20 megapixels, and 8,000 px per edge; PDF and handwriting are out of scope)

Choose one image containing text. It is decoded in this browser.

Japanese uses a secondary vertical-text model. Choose Japanese + English for mixed images.

Local OCR model setup: The OCR runtime and recognition data are fetched on the first run or when the cache is unavailable. Your image is not sent to StemStudio; Tesseract.js 7.0.0 runs in a browser Worker. Cached data may remain in IndexedDB or the browser HTTP cache. When the selected language is not cached, the first run needs about 7.8 MB of OCR data. It cannot run without access to those assets. See the model source, licence, and data conditions

Recognize printed Japanese or English text in an image, inspect it, copy it, or save a UTF-8 text file. The image stays in your browser; the OCR runtime and recognition data are fetched on the first run or when the cache is unavailable.

1Choose imageAn imagewith text2Recognize locallyOCR modelin a Worker3Save textProofread, copyor .txtThe image stays in the browser; OCR data is served as a separate asset.
Image OCR

How to use

  1. 1Choose an imageSelect one browser-readable PNG, JPEG, WebP, or similar image containing text.
  2. 2Choose a language and runPick Japanese, English, or a mixed language, then watch the progress.
  3. 3Save the textProofread the result, then copy it or download a .txt file.

What this tool does

Read image text

An OCR model returns plain recognized text from the image you select.

Keep the image local

No upload is needed. Tesseract.js runs in a Worker and the selected image is not sent to StemStudio.

Inspect and save

Review the text, copy it, or save it as a UTF-8 .txt file.

Related tools

All tools

Limits and good uses

Does it read handwriting or PDFs?

This first version targets one browser-readable image containing printed text. Handwriting, PDF parsing, and searchable-PDF creation are out of scope.

What about vertical Japanese?

A secondary Japanese model attempts vertical text, but does not guarantee it. Orientation, resolution, and lighting affect accuracy.

Is the image uploaded?

The image is processed in this browser. On the first run or after cache eviction, the browser fetches the OCR runtime and recognition data from this site.

Why is confidence not an accuracy percentage?

The displayed confidence is an engine-wide indication. It does not mean every character is correct.

Frequently asked questions

Does it preserve tables or coordinates?

No. The first version returns recognized plain text only. Table, coordinate, and layout reconstruction are separate future work.

Can I cancel it?

Yes. Cancel terminates the active Worker. A previous run cannot overwrite a newer result.

Is an empty result an error?

No. It is a completed empty result. Try a sharper, better-lit image with less skew.