Promotes three previously-manual maintenance tasks to Celery Beat schedules
so the user doesn't have to remember to run them:
- ml.backfill daily
- apply_auto_accept_predictions daily
- recompute_all_centroids weekly
Cadences are env-overridable (ML_BACKFILL_EVERY_SECONDS,
AUTO_ACCEPT_EVERY_SECONDS, CENTROIDS_EVERY_SECONDS).
Each task self-gates so a scheduled run is a no-op when there's nothing
to do:
- ml.backfill: already self-gating — its first paginated query returns
zero rows when no image is missing predictions/embeddings, the loop
breaks, and the task returns. No code change.
- apply_auto_accept_predictions: adds a fast-path NOT EXISTS query that
returns immediately when no WD14 prediction at/above the threshold
exists for an unattached, non-rejected (image, tag) pair. The full
walk only fires when fresh predictions have landed since the last run.
Returns {'skipped_no_candidates': True} on the no-op path.
- recompute_all_centroids: tightens the aggregate query to LEFT JOIN
tag_reference_embedding and skip tags whose stored reference_count
already matches current image_tags membership count. Without this gate
the daily-scheduled sweep would re-enqueue a recompute for every
eligible tag every run, contending with tag_and_embed on the ml queue.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds a Celery task + Settings button that walks every image and applies
general-category WD14 predictions at or above the auto-accept threshold,
without needing the user to open each modal. Same side effects as the
modal's per-image auto-accept flow (creates kind='user' tag if needed,
attaches to image, writes SuggestionFeedback decision='accepted').
Lazy-imports app.services.tag_suggestions so the maintenance worker
doesn't pay its overhead unless this task fires. Amortizes _config()
and _existing_tag_names() across the loop. Commits per-batch (100
images) to keep transactions short and let later batches see freshly
created Tag rows.
Skips integrity-flagged images (get_suggestions already returns empty
for those). Idempotent — re-running just no-ops on already-applied
images via the existing 'tag not in img.tags' guard.
Trigger: Settings -> Maintenance -> "Apply auto-accepts library-wide"
(confirm-gated since it walks the entire library).
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Adds per-image integrity tracking so corrupt files are detected, excluded
from random/showcase/ML/suggestion paths, and recoverable by dropping a
fresh copy in /import — closing the gap that surfaced as the WD14
'6 bytes not processed' OSError.
Schema (migration l26042501)
- image_record.integrity_status: unknown | ok | truncated | unreadable | missing
- image_record.integrity_checked_at: timestamptz
- partial index on status <> 'ok' for cheap report/filter queries
Verifier
- app/services/integrity.py: verify_path() dispatches by extension
- PIL two-stage (verify + load with LOAD_TRUNCATED_IMAGES disabled)
- ffprobe for video, zipfile.testzip for archives
- Truncation-vs-unreadable distinction via PIL message hints
Pipeline
- verify_media_integrity Celery task: per-image, idempotent
- verify_unverified_images sweep: only_unknown by default, skips
paths in active import tasks
- Hooked into the end of import_media_file (new + archive paths) and
the supersede branch
- supersede_image() resets status to 'unknown' so the post-supersede
verify writes a fresh truth
- Supersede-on-replace: a fresh /import/<artist>/<filename> matching
a flagged-corrupt record routes through _supersede_existing,
preserving tags/series/embeddings
Exclusions
- /, /api/random-images, tag_and_embed, ml.backfill enqueue, and
get_suggestions all filter integrity_status IN ('ok', 'unknown') so
flagged rows don't poison the gallery, ML, or suggestion math.
'unknown' is treated as healthy so post-migration data stays visible
until the sweep runs.
UI / report
- Settings -> Maintenance: 'Verify unknown' + 'Force re-verify all'
- GET /api/integrity/failed (paginated list of flagged rows)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
New Celery task sync_character_fandoms_to_images on the maintenance
queue. For every character tag with a non-null fandom_id, runs the
same INSERT ... SELECT ... ON CONFLICT DO NOTHING that set_tag_fandom
already runs inline, attaching the fandom tag to every image that
has the character attached.
Why this exists: migration j26042101 extracted the '(Fandom)' suffix
from pre-refactor character names into tag.fandom_id, but it did NOT
walk image_tags to attach the fandom rows to every image that had
the character. Post-refactor auto-apply (in add_tag /
accept_image_suggestion / set_tag_fandom) only fires on NEW
add/accept events — pre-existing character attachments from before
the migration still show the character pill in the modal but no
fandom pill, because the fandom row was never added to image_tags.
Surfaced by Settings → Maintenance → 'Sync fandoms to images'.
Safe, additive, idempotent. Run once after the migration to close
the gap; subsequent set-fandom operations continue to maintain the
invariant inline.
Verified locally: seeded a character+fandom pair with only the
character attached to image 1; task added the fandom to image_tags;
final state has both rows.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Two capabilities that work together to turn the suggestion system from
a 'review every noisy chip' UX into a 'curated tags only, configurable
auto-apply' UX for a solo-user library.
Auto-accept threshold
Service (tag_suggestions.py):
- New default auto_accept_general_threshold = 0.95 in _DEFAULTS
- get_suggestions splits general-category hits at/above the threshold
into a separate auto_accept_candidates list; the core 'character'/
'copyright'/'general' keys stay the same shape for existing callers
- get_bulk_suggestions ignores the new key (no side effects in bulk)
Route (main.py GET /image/<id>/suggestions):
- For each candidate: find-or-create Tag(kind='user', name=name), attach
to image, log SuggestionFeedback(source='wd14'|etc, decision='accepted')
- Response adds 'auto_accepted: [{id, name, display_name, kind,
confidence, source}, ...]' so the modal can review + undo
Config endpoints (main.py):
GET /api/suggestions/config/auto-accept-threshold
POST /api/suggestions/config/auto-accept-threshold {threshold}
Values > 1.0 effectively disable the feature
UI (view-modal.js):
- New renderAutoAccepted block at top of suggestions section with
green-tinted chips carrying ✕ (undo for this image, logs rejection)
and ⊘ (blocklist + remove from all images)
- loadSuggestions refreshes the tag pill list when auto_accepted has
items so the user sees them in the Tags section too
Settings:
- Maintenance tab gains a number input + save button backed by
/api/suggestions/config/auto-accept-threshold
Retroactive blocklist sweep
New Celery task app.tasks.maintenance.sweep_blocklisted_tag_from_images
on the maintenance queue. Enqueued automatically when a name is added
via POST /api/suggestions/blocklist or appears new in the bulk replace.
Task body: finds kind='user' Tag matching the name, deletes its
image_tags rows, deletes the Tag itself. Scope limited to kind='user'
so deliberate character/fandom/artist tags sharing a blocklisted name
aren't silently destroyed.
Verification (local dev):
- Threshold GET/POST round-trip correct (default 0.95, persisted in
tag_suggestion_config)
- Image 633 at threshold=0.9: 4 general tags auto-applied, each with
a SuggestionFeedback row, returned in auto_accepted
- Blocklist add of 'sweeptestonly' with an attached test tag: Redis
maintenance queue depth = 1, task body removes 1 image_tags row +
the Tag row
- get_bulk_suggestions still works (auto_accept_candidates key skipped)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
tag.fandom_id is now authoritative. set_tag_fandom auto-applies the
fandom tag inline; add_tag already did. There's no drift to sync on
a schedule anymore.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Consolidated merge of feat/tag-suggestions branch. Original 64-commit history
was lost to git-object corruption in a Nextcloud-synced checkout; this single
commit captures the equivalent diff.
Includes:
- pgvector-backed tag suggestion infra (WD14 + SigLIP centroids, ml-worker
container, Celery tasks, suggestion service, accept/reject endpoints + modal
UI with green/red chip buttons)
- Character/fandom integrity: title-case normalization on every write path,
fandom-id backfill, maintenance task + settings button, migrations g26041901
+ h26041901 to canonicalize legacy rows with case-only duplicate merging
- Tag-underscores + modal polish: WD14 name canonicalization at emit + accept
+ add/bulk-add paths, migration i26041901 for legacy-row rename-or-merge
across character/fandom/NULL kinds, suggestion-accept refresh parity via
awaited loadTags, persistent chip tint
bvandeusen
Bryan Van Deusen