Allow users to specify a Tesseract language string (e.g. jpn+jpn_vert)
on a per-extraction basis, overriding the global OCR language setting.
- Add payload column to ai_jobs table (migration) to carry per-call data
- Thread ocrLanguages payload through enqueueJob → processNextJob → extractItemText
- New GET /api/ai-settings/ocr endpoint (requireAuth) returns { ocrMode, ocrLanguages }
- ImageLightbox fetches OCR settings and shows a language input next to the
Extract Text button when mode is hybrid or tesseract (hidden for llm-only)
- MixedView fetches OCR settings and passes them down to EntryTile; kebab
Extract Text on images shows an inline language prompt before dispatching the job
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Poll /api/ai-tagging/fields every 2s after any 202 (queued) response in
ImageLightbox and DoomScrollView so extraction, translation, and description
results appear automatically without a page refresh
- DoomScrollView extract button now turns accent-coloured while a job is
queued instead of flashing red; red is reserved for genuine errors
- Kebab menu "Translate" option is now gated on entry.hasExtractedText
(populated via a batch DB query in the browse API) so it only appears
when there is text to translate
- Tag panel redesigned: toolbar collapses to just the filename when open;
panel header holds hide (›), AI Tagger (✨), and Close (✕) buttons;
sections ordered Description → Text Extraction → Tags; description
state and generate handler moved from TagSelector into ImageLightbox
- VideoPlayerModal receives the same toolbar/panel restructure
- TagSelector gains hideDescription prop so the parent can own description
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Allows users to configure the max_tokens sent to the AI endpoint for
each activity (tagging, description, extraction, translation) individually,
with per-library overrides following the same pattern as model overrides.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Extracted text in the tag panel is now an editable textarea; a Save
button appears when the content is dirty and persists edits to the DB
- Source language input added next to Re-translate button; when filled,
the translation prompt uses "translate from X to Y" for more accurate
results
- New updateExtractedText() helper and PATCH /api/ai-tagging/fields
endpoint to support saving edited text
- translateItemText/translateText accept optional sourceLanguage param
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Show an extract-text button (document icon) in the bottom bar when the
current image has no extracted text yet. Clicking it calls the extract-text
API, shows a spinner while in progress, and on success replaces itself with
the text-lines display button and auto-opens the overlay. Error state briefly
turns the button red. Resets on every item navigation alongside the other
text state. Hidden for videos and items without an itemKey.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add getAiImagePath() to thumbnails.ts (1920px wide, quality 90, no
upscaling) cached separately from display thumbnails via an _ai suffix.
Swap all four image-to-AI code paths in ai-tagger.ts (extract text,
describe, batch tagging x2) to use the new high-res image instead of
the 400px display thumbnail, improving OCR accuracy on dense text.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- move play/pause to clicking the video directly; remove dedicated button
- replace emoji mute icons with flat minimal SVGs
- add view-in-library button in doom scroll that navigates to the file's
directory and opens it in the regular viewer
- add display text overlay button in doom scroll and image lightbox;
shows extracted text (translated by default when available) in a
semi-transparent box at the bottom; toggle between translated/original
- hide tag panel by default in image lightbox and video player modal
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When generating an item description, any already-applied tags are
appended to the system prompt as a source of truth, so the model
can produce a more accurate description aligned with existing tags.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Applied tags are now pinned to the front of each category's tag list,
with unapplied tags continuing in usage order behind them. Both
partitions preserve the existing usage-sort from the API.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When no filters are selected, doom scroll now recursively fetches only
items under the current directory instead of the entire library root.
Navigating to a new directory invalidates the cached listing. Filter-
based doom scroll (search or tags) continues to search library-wide.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add library_ai_settings table with migration for per-library overrides
- Extend AiConfig with editable prompt parts for description, tagging,
extraction, and translation steps; defaults match previous hardcoded values
- Add getEffectiveAiConfig(libraryId) that merges global settings with
library-level overrides (empty override falls through to global)
- Update all ai-tagger functions to use getEffectiveAiConfig and build
prompts from configurable parts
- Add GET/PUT /api/ai-settings/library/[id] for per-library overrides
- Update /api/ai-settings GET/PUT to include prompt fields
- Add Prompts section and Library Overrides section to admin UI
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds automatic image tagging that runs as a post-scan phase, sending
thumbnails to an OpenAI-compatible vision API and applying matching
tags from the user-defined tag vocabulary.
- New ai-tagger module with batch processing, failure tolerance, and
tag validation against existing vocabulary
- Admin settings page (Manage > AI Tagging) for endpoint, model, and
enable toggle with connection testing
- DB migration for ai_tagged_at tracking column and AI config seeds
- Re-tag All support to queue items for reprocessing
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- New /api/game-screenshots route handles GET (list), POST (upload), and DELETE (remove single file)
- Screenshots stored in a screenshots/ subfolder inside each game directory
- Images converted to JPEG via Sharp on upload, named shot-{timestamp}.jpg
- Modal shows a horizontal scrollable strip of 16:9 thumbnail tiles
- Hover a tile to reveal a delete button; uploading placeholders appear per in-flight upload
- Dashed + tile triggers multi-file picker for batch uploads
- Click any thumbnail to open a full-screen lightbox with prev/next arrows, counter, and keyboard nav (←/→/Esc)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Detect Windows (.zip), Linux (.tar.gz), and macOS (.dmg / .app bundle) game archives during scan
- Store GameFile[] with platform metadata in DB instead of plain zipFiles[]
- Stream .app bundles as on-the-fly zip archives via archiver
- Show WIN/LIN/MAC platform badge pills on GameCard and SeriesCard
- Auto-select the download matching the user's OS in GameDetailModal
- Persist cover URL to DB immediately on upload (no re-scan needed)
- Backward-compatible: legacy zipFiles entries map to platform 'windows'
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Formats like .mkv, .avi, .wmv, .ts, .m2ts and .tiff are not natively
supported by browsers and would stall silently in Doom Scroll mode.
Add src/lib/browser-media.ts with BROWSER_VIDEO_EXTENSIONS (.mp4,
.webm, .mov, .m4v), BROWSER_IMAGE_EXTENSIONS (.jpg, .jpeg, .png, .gif,
.webp, .bmp), and an isBrowserPlayable() helper that extracts the
extension without importing Node's path module.
Filter doomScrollItems in MixedView, MoviesView, and TvView using this
helper so only natively renderable files are passed to DoomScrollView.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
When history hit the 100-entry cap, setHistory sliced the array back to
indices 0–99 but setHistoryIndex still returned idx + 1 = 100, making
current = history[100] = undefined. Nothing rendered and no API calls
were made until the user went back (decrementing to index 99, which
held the newly-picked item).
Fix: cap the returned historyIndex at HISTORY_CAP - 1 so it always
points to a valid entry in the sliced array. Extract HISTORY_CAP = 100
as a named constant so the slice and the index cap stay in sync.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
media_key was a lossy shortening of item_key (libraryId:lastSegment) that
introduced a real collision bug: two TV episodes from different series with
the same filename would share the same media_key and each other's tags.
- DB migration converts existing media_tags rows from short format to full
item_key by joining against media_items; ambiguous/orphaned rows are dropped
- media_tags column renamed media_key → item_key
- Removed itemKeyToMediaKey() from scanner; reconcileAndPrune now passes
item_key directly to reKeyMediaItem
- DB reader functions (tv, movies, games) now expose item_key on returned
entities; frontend components use entity.item_key instead of constructing
the short libraryId:id form
- MixedView now constructs the full mixed_file: item_key format
- Tag API renamed mediaKey param → itemKey throughout
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
