How to use
- Choose one PDF or multiple images, then select Traditional Chinese, English, or mixed recognition.
- Prepare the model on first use, then choose pages, DPI, layout mode, and preprocessing.
- Review confidence and correct each page before exporting TXT, Markdown, JSON, or a searchable image PDF.
Privacy
Input and files are processed in the current browser tab. Mingmist does not receive the content unless you independently share the result elsewhere.
Limitations
Up to 80 pages/images are processed sequentially with one worker to limit memory risk. Tesseract.js does not parse PDFs directly, so pages are first rendered with PDF.js. Searchable PDF output rasterizes pages and removes original text, links, forms, attachments, annotations, and digital signatures. Manual corrections apply to text exports only and do not recalculate word coordinates.
FAQ
Are files uploaded?
No. PDFs and images are processed only in the current browser tab. First use downloads only OCR runtime assets and language data.
Why is the first run slower?
The OCR core and language data must be downloaded. Tesseract.js stores language data in browser IndexedDB, and the service worker caches runtime assets on demand.
Why does searchable PDF output need a font?
PDF standard fonts cannot encode CJK text. When results contain non-Latin characters, provide a TTF or OTF font that supports them.
Is recognition guaranteed to be correct?
No. Scan quality, fonts, tables, vertical text, columns, and handwriting affect results. Important content must be reviewed manually.