OCR to PDF turns scans, photos of documents, multi-page TIFFs and scanned PDFs into a PDF whose text can be searched, selected and copied. Every page goes through four steps, in this order: straightening (perspective correction of skewed photos or deskewing of scans, with corners adjustable by hand), automatic orientation, image isolation (photos, stamps, logos, which are not mistaken for text) and text recognition with Tesseract 5. Everything happens on the computer; for damaged prints and handwriting a local vision model can re-read the page.
Italian and English are included; the other 120 languages are downloaded once, with hash verification, and then stay on the computer. Probatio recognises the page's script (Latin, Cyrillic, Arabic, Greek, Chinese…) and suggests the right languages.
A sheet photographed in perspective becomes rectangular again, a skewed scan straight, an upside-down or sideways page is turned. Automatic straightening is designed to apply only to documents, not to ordinary photos.
Optional: a vision model served by Ollama on your computer re-reads each page and corrects Tesseract. Probatio only accepts a local address, so images never leave the machine. The model's text is used only where it agrees with Tesseract and never on a blank page, because a generative model can make things up.
Faithful: the original image, and an untouched JPEG enters the PDF identical byte for byte. Compact: crisp text and separate figures, a much smaller file. Black and white: the lightest, for text only. Text only: adds the text to the original PDF without touching it, which stays intact at the start of the new file.
If the image has low resolution, Probatio warns that “Faithful” will stay grainy and suggests the modes that redraw the text crisply. Pages of a PDF that are already digital, with real text, are not redone.
Progress, pages, a preview with words coloured by confidence, recognised text and an activity console, in a window of its own.
The PDF and, on request, the text (.txt) and the log (.ocr.json) with the files' hashes and, for every page, straightening, rotation, languages, confidence and the local model used, with the number of words changed.