SnipOCR

SnipOCR — automatic local OCR when you take a screenshot.

Works on Windows and macOS. When you capture the screen, SnipOCR:

Detects the new screenshot (clipboard and/or screenshot folder)
Runs on-device OCR (no cloud)
Puts the text on your clipboard
Shows a popup so you can review / edit / re-copy

Platforms

	Windows	macOS
Capture	Snipping Tool (`Win+Shift+S`)	Screenshot (`⌘⇧4`, `⌘⌃⇧4` for clipboard)
OS OCR	Windows.Media.Ocr	Apple Vision
Neural OCR	RapidOCR + ONNX Runtime (same on both)	RapidOCR + ONNX Runtime
Clipboard	win32 clipboard listener	Pasteboard changeCount poll
File saves	`Pictures\Screenshots`	Desktop / `Pictures/Screenshots`
Config	`%APPDATA%\SnipOCR\`	`~/Library/Application Support/SnipOCR/`
Logs	`%LOCALAPPDATA%\SnipOCR\`	`~/Library/Logs/SnipOCR/`

All engines are fully offline after install (RapidOCR downloads ONNX models once on first use).

OCR engines

SnipOCR supports three backends:

Engine id	Name	Platform	Notes
`auto`	OS default	Win / Mac	Windows OCR or Apple Vision
`windows`	Windows OCR	Windows	Language packs via Windows Capability
`macos`	macOS Vision	macOS	Built-in Vision framework
`rapid`	RapidOCR (ONNX)	Win + Mac (+ Linux)	Light PP-OCR models via ONNX Runtime

When to use RapidOCR

Same quality/behavior on Windows and macOS
No Windows OCR language-pack setup
Strong multi-language recognition (Chinese, Japanese, Korean, Latin, Arabic, …)
Optional acceleration: DirectML (Windows), CoreML (macOS), CUDA (NVIDIA)

OS engines remain the lightest install path for pure English UI snips.

Requirements

Python 3.11+
Windows 10/11 or macOS 12+
Platform OCR (optional if you only use RapidOCR):
- Windows: OCR language pack (usually en-US already installed)
- macOS: built-in Vision
RapidOCR (recommended for cross-platform AI OCR):
- rapidocr + onnxruntime (installed via requirements.txt)
- First run downloads small ONNX models (~10–20 MB typical)

Windows — OCR language packs (OS engine only)

Open Windows PowerShell as Administrator:

Get-WindowsCapability -Online | Where-Object { $_.Name -Like 'Language.OCR*' }

Get-WindowsCapability -Online |
  Where-Object { $_.Name -Like 'Language.OCR*en-US*' } |
  Add-WindowsCapability -Online

Or:

.\scripts\install_ocr_lang.ps1 -Language en-US

macOS — Accessibility / notifications (optional)

Screen Recording permission is not required; SnipOCR only reads the clipboard and screenshot files you already captured.
Notifications use osascript (standard user notification).

Optional GPU acceleration (RapidOCR)

# Windows — DirectML (many AMD/Intel/NVIDIA GPUs)
pip uninstall -y onnxruntime onnxruntime-gpu onnxruntime-directml
pip install onnxruntime-directml

# NVIDIA CUDA (any OS with a matching CUDA toolkit)
pip uninstall -y onnxruntime onnxruntime-gpu onnxruntime-directml
pip install onnxruntime-gpu

macOS: the standard onnxruntime wheel may expose CoreML. Leave rapidocr_accel on auto.

Then set in config (or tray → RapidOCR acceleration):

"rapidocr_accel": "dml"

or "coreml" / "cuda" / "cpu".

Quick start

Windows

cd path\to\snipocr
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python main.py

Or:

.\scripts\run.ps1

macOS

cd path/to/snipocr
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python main.py

Or:

chmod +x scripts/run.sh
./scripts/run.sh

A tray / menu-bar icon appears. Take a screenshot — text is OCR’d and copied.

macOS tip: hold Control while capturing (⌘⌃⇧4) to put the snip on the clipboard directly. File-based captures on Desktop are also watched.

Switch to RapidOCR

Tray → OCR engine → RapidOCR (ONNX)
Or edit config:

"ocr_engine": "rapid"

First recognition may download models (one-time). Subsequent snips stay local and fast.

Optional: pre-download default models:

rapidocr download_models

Tray menu

Action	What it does
Enable / Disable SnipOCR	Master switch
Enable OCR all images	OCR any clipboard image (not only screenshot tools)
Show last result	Re-open the last OCR popup
OCR engine	Radio: OS default / Windows or Vision / RapidOCR (ONNX)
RapidOCR model	Auto / Small / Mobile / Server / Tiny
RapidOCR acceleration	Auto / CPU / CoreML / DirectML / CUDA
Quit	Exit

How detection works

By default SnipOCR only OCRs images when:

A screenshot process was recently active
(Windows: ScreenClippingHost / Snipping Tool · macOS: screencapture / Screenshot), or
A new screenshot-like file appears in a known folder, or
(Windows) the clipboard owner is a snipping process

That avoids OCR-ing random images you copy from a browser.

Turn on OCR all images from the tray if you want every clipboard image processed.

Clipboard + file paths are de-duplicated so the same snip is not OCR’d twice.

Settings

Stored at:

Platform	Path
Windows	`%APPDATA%\SnipOCR\config.json`
macOS	`~/Library/Application Support/SnipOCR/config.json`

Core

Key	Default	Meaning
`enabled`	`true`	Master switch
`ocr_engine`	`auto`	`auto` / `windows` / `macos` / `rapid`
`ocr_language`	`en-US`	Preferred language (BCP-47 or RapidOCR alias)
`replace_clipboard_with_text`	`true`	Put OCR text on clipboard
`ocr_all_clipboard_images`	`false`	Skip snip-process filter
`watch_screenshot_folders`	`true`	Watch Desktop / Screenshots folders
`show_popup`	`true`	Show result window
`popup_autohide_seconds`	`8`	Auto-hide popup (`0` = stay)
`show_toast`	`true`	Desktop notifications
`collapse_single_newlines`	`false`	Join soft line breaks into spaces

RapidOCR (when `ocr_engine` is `rapid`)

Key	Default	Meaning
`rapidocr_model_type`	`auto`	`auto` / `small` / `mobile` / `server` / `tiny`
`rapidocr_ocr_version`	`auto`	`auto` / `PP-OCRv6` / `PP-OCRv5` / `PP-OCRv4`
`rapidocr_use_cls`	`true`	Run text-line orientation classifier
`rapidocr_text_score`	`0.5`	Drop lines below this confidence
`rapidocr_accel`	`auto`	`auto` / `cpu` / `dml` / `coreml` / `cuda`
`rapidocr_intra_op_threads`	`-1`	ONNX Runtime intra-op threads (`-1` = library default)
`rapidocr_inter_op_threads`	`-1`	ONNX Runtime inter-op threads

Model tips

Preset	Typical use
`auto`	English → mobile PP-OCRv4; otherwise small PP-OCRv6 (fast multi-lang)
`small`	Default modern path (PP-OCRv6 small when available)
`mobile`	Smaller / older mobile models
`server`	Heavier, higher accuracy
`tiny`	Experimental; may fall back if unsupported for your language

Language mapping (ocr_language → RapidOCR)

Examples	RapidOCR rec model
`en`, `en-US`	`en`
`zh-CN`, `zh`	`ch` (Chinese + often English)
`zh-TW`, `zh-HK`	`chinese_cht`
`ja`, `ja-JP`	`japan`
`ko`, `ko-KR`	`korean`
`de`, `fr`, `es`, `pt`, …	`latin`
`ru`, `uk`, …	`cyrillic`
`ar`, `hi`, `th`, `el`, …	matching script models

If a language + model combo is unavailable, SnipOCR automatically falls back to a working combination.

Example config for cross-platform neural OCR:

{
  "enabled": true,
  "ocr_engine": "rapid",
  "ocr_language": "en-US",
  "rapidocr_model_type": "auto",
  "rapidocr_ocr_version": "auto",
  "rapidocr_use_cls": true,
  "rapidocr_text_score": 0.5,
  "rapidocr_accel": "auto",
  "replace_clipboard_with_text": true,
  "show_popup": true,
  "show_toast": true
}

Logs:

Platform	Path
Windows	`%LOCALAPPDATA%\SnipOCR\snipocr.log`
macOS	`~/Library/Logs/SnipOCR/snipocr.log`

Project layout

snipocr/
  main.py
  assets/                 # brand logo + icons
  app/
    platform_util.py      # OS detection, paths, hints
    clipboard_io.py       # cross-platform facade
    clipboard_io_win.py
    clipboard_io_mac.py
    clipboard_watcher.py
    screenshot_watcher.py
    snip_detector.py
    service.py
    result_ui.py
    tray.py
    settings.py
    notifications.py
    paths.py
    ocr/
      base.py             # OCREngine protocol + OCRResult
      windows_ocr.py      # Windows.Media.Ocr
      macos_ocr.py        # Apple Vision
      rapid_ocr.py        # RapidOCR + ONNX Runtime
  scripts/
    run.ps1
    run.sh
    install_ocr_lang.ps1
    generate_logo.py

Troubleshooting

Windows — “No Windows OCR engine”

Install the OCR language pack (see above). Confirm:

# Windows PowerShell 5.x
[Windows.Media.Ocr.OcrEngine, Windows.Foundation, ContentType = WindowsRuntime] | Out-Null
[Windows.Media.Ocr.OcrEngine]::AvailableRecognizerLanguages

Or switch to RapidOCR: tray → OCR engine → RapidOCR (ONNX).

macOS — Vision import errors

pip install 'pyobjc-framework-Vision' 'pyobjc-framework-Quartz' 'pyobjc-framework-Cocoa'

RapidOCR — “not installed” / import errors

pip install -U rapidocr onnxruntime

First use needs network once to download models (ModelScope CDN). After that, fully offline.

Pre-download:

rapidocr download_models
rapidocr check

RapidOCR is slow on first snip

Cold start loads ONNX sessions; later snips are much faster.
Prefer rapidocr_model_type: "auto" or "small".
Enable GPU EP if available (dml / coreml / cuda).
Huge screenshots are downscaled (max side 2000 px).

Snips do nothing

Is the tray icon enabled?
Did you use the OS screenshot tool?
Try Enable OCR all images once to test the OCR path.
Check the log file paths above.
On macOS, prefer ⌘⌃⇧4 (clipboard) or ensure Desktop/Screenshots is writable.

Clipboard still has the image

If no text was found, the image is left on the clipboard on purpose.

Privacy

OCR runs locally (Windows.Media.Ocr, Apple Vision, or RapidOCR/ONNX)
No cloud OCR API in the default path
RapidOCR may download model weights once from RapidAI/ModelScope on first use
OCR text is not written to disk by default (only logs metadata)

License

MIT — use freely.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

SnipOCR

Platforms

OCR engines

When to use RapidOCR

Requirements

Windows — OCR language packs (OS engine only)

macOS — Accessibility / notifications (optional)

Optional GPU acceleration (RapidOCR)

Quick start

Windows

macOS

Switch to RapidOCR

Tray menu

How detection works

Settings

Core

RapidOCR (when `ocr_engine` is `rapid`)

Project layout

Troubleshooting

Windows — “No Windows OCR engine”

macOS — Vision import errors

RapidOCR — “not installed” / import errors

RapidOCR is slow on first snip

Snips do nothing

Clipboard still has the image

Privacy

License

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
Latest commit History 9 Commits
.agents		.agents
app		app
assets		assets
scripts		scripts
.gitignore		.gitignore
README.md		README.md
main.py		main.py
requirements.txt		requirements.txt

Folders and files

Latest commit

History

Repository files navigation

SnipOCR

Platforms

OCR engines

When to use RapidOCR

Requirements

Windows — OCR language packs (OS engine only)

macOS — Accessibility / notifications (optional)

Optional GPU acceleration (RapidOCR)

Quick start

Windows

macOS

Switch to RapidOCR

Tray menu

How detection works

Settings

Core

RapidOCR (when ocr_engine is rapid)

Project layout

Troubleshooting

Windows — “No Windows OCR engine”

macOS — Vision import errors

RapidOCR — “not installed” / import errors

RapidOCR is slow on first snip

Snips do nothing

Clipboard still has the image

Privacy

License

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

RapidOCR (when `ocr_engine` is `rapid`)