ocr🦫 3000!

A fast, beautiful OCR command-line tool for macOS. Powered by Apple Vision framework's VNRecognizeTextRequest (Revision 3).

Pipe-friendly by default — results go to stdout. Write to files with -o and the format is auto-detected from the extension.

Install

swift build -c release
cp .build/release/macocr /usr/local/bin/

Usage

macocr photo.jpg                          # OCR to stdout
macocr scan.png -o results.json -c        # JSON with confidence scores
macocr document.pdf -f -l de -o out.md    # Fast German PDF to Markdown
macocr *.png -a --format json > all.json  # Batch with auto-detect
macocr receipt.jpg -w "IBAN" -w "BIC"     # Custom dictionary

Features

Recognition

Flag	Description
-f, --fast	Fast recognition (lower quality, near-instant)
-l, --languages	Set languages in priority order (-l en -l de)
-a, --auto-detect	Automatically detect document language
--no-correction	Disable language correction
-w, --words	Custom dictionary words
--min-height	Minimum text height relative to image (0.0-1.0)
-n, --candidates	Number of recognition candidates per line

Output

Flag	Description
-o, --output	Write to file (format from extension: txt/md/json/html)
--format	Force output format
-c, --confidence	Include confidence scores
-b, --bounds	Include bounding boxes

PDF

Flag	Description
--dpi	PDF rasterization DPI (default: 200)

Info

Flag	Description
--list-languages	Show all supported recognition languages
--version	Show version

Output Formats

• txt — plain text, one line per recognition (default)
• md — Markdown with headers per file
• json — structured JSON with optional confidence/bounds metadata
• html — styled HTML document with dark mode support

Benchmark: ocr🦫 3000! vs Tesseract 5

Tested on Tesseract's own test images. Full details in benchmark/BENCHMARK.md.

Accuracy (Character Error Rate — lower is better)

                    phototest    eurotext    book page 1    book page 2
                    (clean)      (multi-lang) (scanned)     (scanned)
  ─────────────────────────────────────────────────────────────────────
  Tesseract 5    │  100.0%       98.8%        90.6%          92.2%
  ocr🦫 accurate │  100.0%       99.8%  ★     92.5%  ★       93.0%  ★
  ocr🦫 fast     │  100.0%       97.6%        38.5%          98.5%

Performance (seconds — lower is better)

                    phototest    eurotext    book page 1    book page 2    4MB grayscale
  ──────────────────────────────────────────────────────────────────────────────────────
  Tesseract 5    │  0.206        0.573       2.033          2.178          2.245
  ocr🦫 accurate │  0.870        0.431  ★    0.938  ★       1.084  ★      1.067  ★
  ocr🦫 fast     │  0.156  ★     0.164  ★    0.474  ★       0.478  ★      0.144  ★

Summary

Engine	Avg Accuracy	Avg Speed	Best For
Tesseract 5	95.4%	1.25s	Cross-platform, Linux servers
ocr🦫 accurate	96.3%	0.83s	Best quality on macOS
ocr🦫 fast	83.7%	0.32s	Quick previews, clean input

Supported Languages

30 languages in accurate mode including English, German, French, Spanish, Chinese, Japanese, Korean, Russian, Arabic, and more. Run macocr --list-languages for the full list.

Requirements

macOS 14+, Swift 5.9+

License

MIT