Image Triage

Image Triage is a PySide6 desktop application for rapidly browsing and culling very large folders of images. It uses a virtualized thumbnail grid, asynchronous folder scanning, background thumbnail generation, and both memory and disk caches so scrolling stays responsive even with tens of thousands of files.

Features

• Left-hand folder tree for fast directory navigation
• Large fixed-column thumbnail layout with 2, 3, or 4 images across
• Virtualized grid that only paints visible rows plus a small buffer
• Progressive loading with immediate placeholders
• Background thumbnail generation with visible-item prioritization
• Memory LRU cache and persistent on-disk thumbnail cache
• Nikon RAW .NEF support for thumbnails and preview
• Fast keyboard workflow
• Full-screen preview that loads asynchronously
• EXIF capture details in preview and compare mode
• Persistent accepted/rejected/rating/tag decisions
• Persistent warm-start folder scan cache for faster folder reopen
• Smart compare that can auto-size to detected exposure brackets
• Right-click file actions including Explorer reveal and app handoff
• Sorting by filename, date modified, or file size

Keyboard Controls

• Arrow keys: move selection
• Page Up / Page Down / Home / End: jump through the grid
• Space or Enter: open full-screen preview
• Delete: move selected file to trash when available
• W: accept the selected image
• X: reject the selected image
• K: move selected file into a _keep subfolder
• M: move selected file to a folder you choose
• 0-5: assign a rating
• T: assign tags
• Esc: close full-screen preview

Run

py -3 -m pip install -e .
py -3 -m image_triage

If the py launcher is not available on your machine, run the commands with your local Python executable instead.

If you are updating an existing install, rerun the install command so dependencies stay in sync:

py -3 -m pip install -e .

Windows MSI Updates

The Windows app includes Help > Check For Updates.... By default it checks the latest GitHub Release for tylermcm/ImageTriage, looks for an .msi asset, downloads it, and launches msiexec so the newer MSI can upgrade the existing install without requiring an uninstall.

For this to work in production:

• Bump project.version in pyproject.toml and __version__ in image_triage/_version.py for every release.
• Keep the upgrade_code in setup_msi.py unchanged.
• Push a matching version tag, for example v1.1.4. The Build And Release Windows MSI GitHub Actions workflow builds the MSI and uploads it to that GitHub Release.
• Prefer attaching a release asset digest or hosting a JSON update manifest with a sha256 value so downloads can be verified before install.

To use a custom feed instead of GitHub Releases, set IMAGE_TRIAGE_UPDATE_FEED_URL to the URL of a JSON document like:

{
  "version": "1.1.4",
  "installer_url": "https://example.com/downloads/ImageTriage-1.1.4.msi",
  "sha256": "0123456789abcdef...",
  "release_notes_url": "https://example.com/releases/1.1.4"
}

Linux AppImage

Build and install a local AppImage from this repo with:

bash packaging/install_linux.sh

That script will:

• create .linux_build_venv
• install the AppImage build dependencies from packaging/linux-build-requirements.txt
• run setup_linux.py to build the AppImage
• install the result under ~/.local/opt/ImageTriage
• create a launcher at ~/.local/bin/image-triage
• create a desktop entry under ~/.local/share/applications

If you already have a built AppImage, install just that file:

bash packaging/install_linux.sh --appimage /path/to/ImageTriage-0.1.0-x86_64.AppImage

Useful options:

• --skip-build: reuse the newest AppImage in ./dist
• --install-dir PATH: change the install target
• --no-desktop: skip desktop entry creation

GitHub Actions also builds an AppImage artifact on pull requests, pushes to main, and manual runs through build-linux-appimage.yml. Pushing a matching v* tag builds the same AppImage and uploads it to that GitHub Release.

The Linux build and the Windows MSI now share the same AI runtime staging logic through freeze_support.py, so both package types bundle the same integrated AICullingPipeline tree, helper launchers, and Python-side AI dependencies.

The active CLI-Culler source lives in the in-repo aiculler package and is the default AI culler implementation used by the app. CLIP/TOPIQ weights remain outside git; set IMAGE_TRIAGE_AICULLER_MODEL_ROOT to point at a local models directory when the defaults do not apply.

On first launch, the app offers to download the AI model into:

~/.cache/image_triage_ai_cache/models/DinoV3

If the user skips that step, the older bundled AI pipeline stays unavailable until the model is installed.

Notes

• Version 1 focuses on scan speed, responsive scrolling, and large visual previews.
• Workflow refinement, metadata polish, and side-by-side compare improvements are natural next steps.