mg-chao/snow-apps

Snow Apps is worth a look if your daily work involves capturing a region of the screen, marking it up, extracting text, or flipping through a folder of images, and you'd rather use one integrated suite than a stack of separate utilities. It bundles Snow Shot (screenshot/annotation/OCR/screen recording) with Snow Image Viewer, and installs like a normal desktop app on Windows x64 and Apple Silicon macOS 15+.

The problem

You see something on screen worth keeping: a bug reproduction, a UI detail, a paragraph in an image, a receipt. The old workflow is to press the OS screenshot shortcut, paste into an editor, draw boxes and arrows, then run a separate OCR tool or retype the text, and finally save the file somewhere retrievable. Snow Shot tries to compress that into one capture-and-annotate step with OCR built in, so the text and the marked-up image come out of the same session. Snow Image Viewer covers the follow-up: opening a saved capture or a folder of images without bouncing between apps.

What's different

The interesting part is the packaging rather than a single novel feature. Screenshot tools, annotation editors, OCR, and image viewers are each common; combining them in one C++/Qt codebase with a Rust OCR component (the simd-paddle-ocr-rs submodule, itself an Apache-2.0 project with its own release cadence) is the distinguishing choice. The OCR submodule ships engine defaults rather than requiring every user to fetch models first, which lowers the setup cost. The trade-off is that this is a multi-license repository: snow-crates/, snow_image/, snow_image_viewer/, snow_shot/ and the FFI layer are GPL-3.0-or-later, while the Qt design toolkit and OCR crate are Apache-2.0. If you plan to embed or fork components, check LICENSE.md scope rules first, because the GPL portions constrain downstream use more than the Apache ones.

Another practical wrinkle: Windows ARM64 builds and cross-compilation are documented separately, including OCR preparation and release validation prerequisites. That suggests the ARM64 path is more involved than the x64 one. And the repository takes fix-only pull requests; new features go through community discussion first. If you were hoping to shape the roadmap by submitting a feature PR, this isn't the project for that.

Who should try it

Good fit: a Windows x64 or Apple Silicon macOS user who takes screenshots regularly for documentation, support tickets, or teaching, and wants annotation plus OCR in one place. Also a reasonable fit if you already use Homebrew or WinGet and prefer package-manager installs with documented upgrade and removal paths. Poor fit: Linux users (no install path documented here), Intel Macs, older macOS versions, anyone who needs a fully permissive license for redistribution, or someone who wants to upstream features rather than report bugs. Windows ARM64 users should read the dedicated build doc before assuming a smooth install.

Requirements and limitations

On Windows, install the x64 package from Snow Shot releases or via WinGet; the WinGet package includes beta versions and the default OCR resources. A Scoop path is described, but its acceptance into the official Extras bucket is pending, so treat it as not-yet-available until that submission lands. On macOS, the Homebrew cask requires Apple Silicon and macOS 15+, reuses a local signing identity, and may prompt for Keychain access and privacy permissions.

Ongoing upkeep differs by channel: the Scoop instructions explicitly say to quit the app and use Scoop rather than the built-in updater, and settings/history persist across upgrades and ordinary uninstall. Building from source means initializing the OCR submodule separately, and OCR changes must be committed to that submodule's own repository before the parent revision is updated. Check the macOS Homebrew installation and migration note if you already have an installation or use a custom app directory.

HotRepo's take

HotRepo recorded this project on the daily board on 2026-10-24 (6,226 stars, +39) and 2026-10-25 (6,234 stars, +41). That two-day sample is too thin to judge momentum either way; treat it as a snapshot, not a trend. The concrete thing to verify before committing is the license split: confirm which components you actually need and whether the GPL-3.0-or-later terms fit your use.