Repository navigation
VetroLook 2.4.0 — Tokyo Glass and Smart Gallery
Changes since the last published release, 2.2:
- English / Русский on first launch. The question is in English and stays open until you choose. Close, Alt+F4 and Escape cannot dismiss it. Your choice is saved and can be changed later in the menu.
- Tokyo Glass. Cream, charcoal and red, bold typography, rounded frosted surfaces and restored ambient image/video reflection.
- Files sidebar. Recent, Favorites, pinned folders, system-location icons and a compact current-path stack. Pin/unpin or move a folder to the Recycle Bin from its context menu.
- Quick Scope and spring-loaded folders. Search the current folder, subfolders or all indexed locations; drag-hover for about 500 ms to open a folder.
- Local Smart Gallery. The bundled six-class classifier runs in the background. Initial progress can morph into a bottom-right toast; completion expands it into an application question.
- Your confirmation controls filtering. First-pass labels never hide images until Yes, apply. Later preserves labels and asks again next launch. After Yes, new files apply automatically.
- Persistent corrections. Manual Show/Hide decisions take priority, uncertain images stay visible and original files stay untouched.
- Complete model payload in release packages. EXE, MSI and portable ZIP include the same model and policy, playback library and Lensfun database. The legacy clean uninstaller now preserves settings and gallery corrections.
Downloads
- VetroLook-2.4.0-Setup.exe — recommended Windows x64 installer.
- VetroLook-2.4.0-x64.msi — Windows Installer package.
- VetroLook-2.4.0-Portable.zip — portable distribution.
- VetroLook-Clean-Uninstall.exe — legacy installation cleanup; settings and gallery corrections are preserved.
- SHA256SUMS.txt — checksums.
Verification and model limits
Release build, both language choices, blocked dismissal, preference persistence and invalid-setting recovery were checked. Native Smart Gallery integration passed 20 checks; all four popup/toast/application UI scenarios passed.
The classifier was not retrained for this release. Its real-photo false-hidden rate on the reviewed evaluation was 2/4,178 (0.04787%); the one-sided 95% upper bound is 0.15061%, above the original 0.1% target. The original scientific gate remains failed. This model is distributed with publisher acceptance and requires the user's initial application confirmation. See the model card.