What it does Umi-OCR is a desktop GUI for offline text recognition. It runs entirely local—no cloud APIs, no subscription tiers. You can screenshot a region, drop a folder of images, or feed it PDFs and e-books (XPS, EPUB, MOBI, FB2, CBZ). It spits out plain text, JSONL, Markdown, or CSV. There’s also a QR/barcode scanner and generator tucked in, supporting 19 protocols from Aztec to DataMatrix.

The interesting bit The “ignore region” feature is the quiet killer: draw rectangles over watermarks or page headers in batch jobs, and the OCR engine skips entire text blocks inside those zones. The README even warns you to draw the boxes generously—no pixel-perfect nonsense required. For document nerds, it can turn scanned PDFs into double-layer searchable PDFs while stripping out recurring page-foot clutter.

Key highlights

• Two swappable offline engines: RapidOCR (compatibility) and PaddleOCR (speed)
• CLI and HTTP APIs for automation; Scoop install on Windows
• Layout-aware post-processing: handles multi-column, vertical text, and code indentation preservation
• Cross-platform: Windows 7+ x64 and Linux x64
• GPU-accelerated Qt/QML UI with fallback renderers if your graphics driver throws a tantrum

Caveats

• Windows and Linux only; macOS is not mentioned in the build instructions
• The “formula recognition” feature appears to be experimental or incomplete—it’s referenced only via a GitHub issue link
• Engine plugins live in a separate repo; switching isn’t quite plug-and-play

Verdict Grab this if you process documents in bulk and are tired of cloud OCR quotas or privacy policies. Skip it if you need mobile support, real-time video OCR, or macOS-native builds.