English | 简体中文

FleetingGleam

A local-first photo culling app for macOS and Android. It groups the bursts in thousands of photos, picks the best frame in each group and shows you why. Your photos never leave your own devices.

Website: fleetinggleam.fleetinggleam-web.workers.dev · Download: latest release

What it does

• Groups your bursts. Frames of the same moment sit together. Faces, orientation and the time gap between shots all count.
• Picks the best frame. Every photo gets a score and the best of each group is marked with a gold star. The inspector on the right lists where the score came from: sharpness, exposure, AI impression, faces and more.
• Compare and decide. Zoom into the frames of one group side by side, then press P to keep, R to review, X to reject, and ⌘Z to undo.
• Scores by kind of scene. Portraits are judged by faces, sport by focus, and a night shot is not marked down for being dark.
• Find photos by describing them. Type "a small boat by the sea" or "two people". English and Chinese both work.
• Learns your taste. Point it at a folder of photos you picked in the past and it learns your preferences. Keep one style for weddings and another for street.
• Organise. Keepers are copied to a new folder (on Android, to your photo album). On a Mac the rest can go to the Trash, where they can be recovered, and put back in one click.
• Works on a NAS. Add your NAS in Settings (server and account; SMB); the app connects with its own built-in SMB client (no Finder, no mounted volume on the desktop), a new project browses the NAS's shares and picks any folder, and photos are mirrored into the app's cache before analysis.
• Cloud backup of your taste. Your settings and what the taste models learned can be backed up automatically to your own Cloudflare R2 bucket, and restored on another computer.
• Take your taste with you. Settings and everything the styles learned export to one file (embeddings and decisions only, never photos or paths) and import on another computer.

A score is a hint, not a verdict. The scoring constants are tuned by hand and have not yet been calibrated against real human labels. The app can tell you which frame is technically stronger, but a good photograph is sometimes imperfect. The final call is always yours.

Privacy

• Photos are processed on your computer and never uploaded. There is no account.
• The app goes online only to download the model files it needs, the first time a feature uses them, and to reach a NAS or a backup bucket that you set up yourself. Otherwise it works fully offline.
• A few models under 10 MB come with the app. The large ones are neither in this repository nor in the installer. They are downloaded straight from public sources such as Hugging Face, and every file is checked against a SHA-256 built into the app. Where each file comes from is listed in docs/MODELS.md; how it looks to a user is in the getting started guide (Chinese).

Quick start

Using the app

Download the .dmg from Releases. It needs a Mac with Apple Silicon (M-series) and macOS 15 or later; Intel Macs are not supported yet. This build is not notarised by Apple, so before opening it the first time, run this once in Terminal:

xattr -cr /Applications/FleetingGleam.app

Step by step instructions are in the getting started guide (Chinese).

For Android, download FleetingGleam-android.apk from the same Release and open it on the phone. It needs Android 12 or later on a 64-bit ARM phone. The app is not in an app store, so allow your browser to install apps when Android asks.

A Windows version is planned and does not exist yet.

Developing

You need: macOS 15+, the Xcode command line tools (for swiftc), Rust stable, Node.js 22 and pnpm 11 (the exact version is pinned by packageManager in the root package.json).

pnpm install

pnpm dev:web        # landing page, http://localhost:3150 (English at /, Chinese at /zh)
pnpm dev:desktop    # desktop app (tauri dev); compiles Rust and Swift, slow the first time

Only want to work on the interface, without compiling Rust or downloading models? The desktop app has a browser preview mode that uses drawn sample photos and a simulated analysis:

pnpm desktop dev    # then open http://localhost:1420/?mock

The Apple Vision helper for macOS (optional; the app runs without it) has to be built once:

sh apps/desktop/scripts/build-vision.sh

More about developing and the checks to run before you commit: CONTRIBUTING.md (Chinese).

Repository layout

apps/
  desktop/        Tauri 2 + React + Rust desktop app (macOS first)
    src/          the interface (TypeScript / React)
    src-tauri/    the Rust backend; native/ and vision/ hold the macOS parts written in Swift
  android/        native Android app (Kotlin + Jetpack Compose); see apps/android/README.md
  web/            landing page (TanStack Start + Tailwind v4 + Motion), in English and Chinese
packages/
  core/           the shared Rust core (analysis, scoring, NAS, backup) used by both the Mac and the Android app
docs/             documentation, see the table below
branding/         logo and icon sources
DESIGN.md         the design system: colour, type and motion rules

Documentation

Most of the documentation is written in Chinese for now. Issues and pull requests in English are welcome.

Document	What is in it
Getting started (Chinese)	Install, the first model download, and the whole flow from import to organising
CONTRIBUTING.md (Chinese)	Contributing: environment, checks, conventions
docs/RELEASING.md (Chinese)	How to publish a version and how the GitHub Actions work
docs/THIRD_PARTY.md (Chinese)	The third-party models and libraries we use, and their licenses
docs/MODELS.md (Chinese)	Where every model file is downloaded from, its size and license, and how it is verified
docs/RESEARCH.md (Chinese)	Research notes on culling methods and models, including the aesthetic-model survey
docs/REVIEW-2026-10-20.md (Chinese)	Code review of the desktop app: prioritised list of what to improve
docs/REVIEW-2026-10-21.md (Chinese)	Second review round: backend, frontend, landing page and CI
docs/ROADMAP.md (Chinese)	The plan for the macOS and Android apps
apps/android/README.md (Chinese)

License

The source code of this project is released under the GNU Affero General Public License, version 3 or later (AGPL-3.0-or-later).

• You are free to use, study, modify and share it. If you distribute a modified version, or offer it to others as a network service, you must make the corresponding source code available under the same license.
• Not covered by this license:
  • The names "FleetingGleam" and "惊鸿", and the logo and icons in branding/. Please do not use them to suggest that your derivative is the official one.
  • The sample photos in the landing page screenshots come from Lorem Picsum under the Unsplash license and are not part of this project.
  • The models the app uses each have their own license, see docs/THIRD_PARTY.md.

Contributing

Issues and pull requests are welcome, in English or Chinese. Please read CONTRIBUTING.md and the code of conduct first.