Highlights
- Added GPU precision models for automatic 2D/3D/Real classification in Core and Video.
- Book now saves your reading direction. Right-to-left reading also applies to the page gallery, page-area clicks and progress controls, with improved two-page placement.
- Landscape and portrait Reels use the same full-height vertical page transition and make better use of the available picture area.
- Customize supported player keyboard shortcuts and optionally use mouse side buttons to go back. Existing bindings are preserved.
Browsing and playback
- Improved Home card sizing and spacing, mixed cover shapes, and items without artwork.
- Use Left/Right to seek while the Core or Video playback queue has focus, including Explorer players. Use Up/Down to navigate the Reels sidebar.
- Reels retains automatic next-item playback and original-aspect settings after restart. Previous/next navigation and returning to the player have also been refined, while saved playlist order is preserved.
- Repeat switches stay synchronized across Core, Video, Voice and Explorer. Leaving Reels or a Gallery preview preserves your ordinary repeat preference.
- Two-page Book reading keeps page pairs in order and centers a single remaining page.
- Hover previews preserve the current Mini playback and do not add to view counts.
- Improved typing and focus when editing thumbnail filename suffixes.
- Fixed duplicate-review details becoming unresponsive when unrelated background updates refresh the card.
Automatic classification and model files
- With automatic tagging enabled, GPU precision classification is used by default on supported hardware. You can turn it off in Settings; the existing lightweight models remain the fallback when it is unavailable.
- Model updates preserve manual tags. Interrupted automatic classification can resume the remaining items.
- These additional large model files increase the download and installed size of this update.
- Recoverable errors during GPU classification now switch the remaining items to the lightweight CPU models, while preserving completed GPU results.