feat(suggestions): auto-accept threshold + retroactive blocklist sweep

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>

app/celery_app.py

@@ -71,6 +71,7 @@ def make_celery(app=None):
             'app.tasks.scan.cleanup_old_tasks': {'queue': 'maintenance'},
             'app.tasks.scan.update_system_stats': {'queue': 'maintenance'},
             'app.tasks.scan.update_batch_stats': {'queue': 'maintenance'},
+            'app.tasks.maintenance.sweep_blocklisted_tag_from_images': {'queue': 'maintenance'},
 
             # Import tasks - handled by worker (heavy processing)
             'app.tasks.import_file.*': {'queue': 'import'},

app/main.py

@@ -743,9 +743,57 @@ def list_tags(image_id):
 
 @main.get("/image/<int:image_id>/suggestions")
 def get_image_suggestions(image_id):
+    """Return the modal's suggestion shape AND auto-apply general-category
+    suggestions above the auto_accept_general_threshold config. Each
+    auto-applied tag is attached to the image and a SuggestionFeedback row
+    logged, and the applied tags are surfaced in `auto_accepted` so the UI
+    can show them for review.
+    """
     from app.services.tag_suggestions import get_suggestions
-    suggestions = get_suggestions(image_id)
+    from app.models import SuggestionFeedback
-    return jsonify({'ok': True, 'suggestions': suggestions})
+
+    img = ImageRecord.query.get(image_id)
+    if img is None:
+        return jsonify({'ok': False, 'error': 'image_not_found'}), 404
+
+    grouped = get_suggestions(image_id)
+    candidates = grouped.pop('auto_accept_candidates', []) or []
+
+    auto_accepted = []
+    for cand in candidates:
+        name = cand['name']
+        # General-category auto-accepts become kind='user' tags (same shape
+        # as a manual accept of a general suggestion via accept_image_suggestion).
+        tag = Tag.query.filter_by(kind='user', name=name).first()
+        if tag is None:
+            tag = Tag(kind='user', name=name)
+            db.session.add(tag)
+            db.session.flush()
+        if tag not in img.tags:
+            img.tags.append(tag)
+        db.session.add(SuggestionFeedback(
+            image_id=image_id,
+            tag_name=name,
+            suggestion_source=cand.get('source', 'wd14'),
+            confidence=cand.get('confidence', 0.0),
+            decision='accepted',
+        ))
+        auto_accepted.append({
+            'id': tag.id,
+            'name': tag.name,
+            'display_name': tag.display_name,
+            'kind': tag.kind,
+            'confidence': cand.get('confidence', 0.0),
+            'source': cand.get('source', 'wd14'),
+        })
+    if auto_accepted:
+        db.session.commit()
+
+    return jsonify({
+        'ok': True,
+        'suggestions': grouped,
+        'auto_accepted': auto_accepted,
+    })
 
 
 @main.post("/api/suggestions/bulk")
@@ -952,6 +1000,19 @@ def list_suggestion_blocklist():
     return jsonify(ok=True, names=[r.name for r in rows])
 
 
+def _enqueue_blocklist_sweep(name: str) -> None:
+    """Enqueue the Celery task that removes blocklisted name from image_tags.
+    Best-effort: if enqueue fails, log and continue so the blocklist add
+    itself still succeeds."""
+    try:
+        from app.tasks.maintenance import sweep_blocklisted_tag_from_images
+        sweep_blocklisted_tag_from_images.apply_async(args=[name], queue='maintenance')
+    except Exception as e:
+        current_app.logger.warning(
+            "failed to enqueue blocklist sweep for %r: %s", name, e,
+        )
+
+
 @main.post("/api/suggestions/blocklist")
 def add_suggestion_blocklist():
     from app.models import TagSuggestionBlocklistEntry
@@ -964,6 +1025,9 @@ def add_suggestion_blocklist():
         return jsonify(ok=True, added=False, name=name)
     db.session.add(TagSuggestionBlocklistEntry(name=name))
     db.session.commit()
+    # Fire-and-forget retroactive cleanup: remove the now-blocklisted name
+    # from every image that still has it attached.
+    _enqueue_blocklist_sweep(name)
     return jsonify(ok=True, added=True, name=name)
 
 
@@ -982,20 +1046,66 @@ def remove_suggestion_blocklist():
     return jsonify(ok=True, removed=False, name=name)
 
 
+@main.get("/api/suggestions/config/auto-accept-threshold")
+def get_auto_accept_threshold():
+    from app.models import TagSuggestionConfig
+    row = TagSuggestionConfig.query.filter_by(key='auto_accept_general_threshold').first()
+    try:
+        value = float(row.value) if row else 0.95
+    except (TypeError, ValueError):
+        value = 0.95
+    return jsonify(ok=True, threshold=value)
+
+
+@main.post("/api/suggestions/config/auto-accept-threshold")
+def set_auto_accept_threshold():
+    from app.models import TagSuggestionConfig
+    payload = request.get_json(force=True, silent=True) or {}
+    try:
+        threshold = float(payload.get('threshold'))
+    except (TypeError, ValueError):
+        return jsonify(ok=False, error='invalid_threshold'), 400
+    if threshold < 0:
+        threshold = 0.0
+    row = TagSuggestionConfig.query.filter_by(key='auto_accept_general_threshold').first()
+    if row is None:
+        row = TagSuggestionConfig(
+            key='auto_accept_general_threshold',
+            value=str(threshold),
+            description='General-category suggestions at or above this confidence are auto-applied by the modal suggestions load. Set to 1.1 or greater to disable.',
+        )
+        db.session.add(row)
+    else:
+        row.value = str(threshold)
+    db.session.commit()
+    return jsonify(ok=True, threshold=threshold)
+
+
 @main.post("/api/suggestions/blocklist/bulk")
 def bulk_replace_suggestion_blocklist():
-    """Replace the entire blocklist with the provided names. Textarea-save path."""
+    """Replace the entire blocklist with the provided names. Textarea-save path.
+    Enqueues a cleanup sweep for every name that wasn't already blocklisted —
+    names that were already in the list skip the sweep since they've presumably
+    already been swept when first added."""
     from app.models import TagSuggestionBlocklistEntry
     payload = request.get_json(force=True, silent=True) or {}
     names = payload.get('names', [])
     if not isinstance(names, list):
         return jsonify(ok=False, error='names_must_be_list'), 400
     cleaned = sorted({str(n).strip() for n in names if str(n) and str(n).strip()})
+
+    previous = {row.name for row in TagSuggestionBlocklistEntry.query.all()}
+    newly_added = [n for n in cleaned if n not in previous]
+
     TagSuggestionBlocklistEntry.query.delete()
     for name in cleaned:
         db.session.add(TagSuggestionBlocklistEntry(name=name))
     db.session.commit()
-    return jsonify(ok=True, count=len(cleaned), names=cleaned)
+
+    for name in newly_added:
+        _enqueue_blocklist_sweep(name)
+
+    return jsonify(ok=True, count=len(cleaned), names=cleaned, newly_added=len(newly_added))
 
 
 @main.post("/image/<int:image_id>/tags/add")

app/services/tag_suggestions.py

@@ -57,6 +57,10 @@ _DEFAULTS = {
     'threshold_meta':             0.5,
     'threshold_embedding':        0.85,
     'min_reference_images':       5.0,
+    # Above this confidence, general-category suggestions are auto-applied
+    # to the image by the route layer (see app.main.get_image_suggestions).
+    # Use 1.1 or greater to effectively disable. Default is conservative.
+    'auto_accept_general_threshold': 0.95,
     'wd14_model_version':         'wd-eva02-large-tagger-v3',
     'siglip_model_version':       'siglip-so400m-patch14-384',
 }
@@ -246,17 +250,34 @@ def get_suggestions(
     if existing_all is None:
         existing_all = _existing_tag_names()
 
+    # Separate out general-category suggestions above the auto-accept
+    # threshold. The route layer applies these to the image directly;
+    # we just identify them here so the service remains side-effect-free.
+    try:
+        auto_accept_threshold = float(cfg.get('auto_accept_general_threshold',
+                                              _DEFAULTS['auto_accept_general_threshold']))
+    except (TypeError, ValueError):
+        auto_accept_threshold = _DEFAULTS['auto_accept_general_threshold']
+    auto_accept_candidates: list[dict] = []
+
     grouped: dict[str, list[dict]] = {'character': [], 'copyright': [], 'general': []}
     for s in merged:
         if s['category'] not in grouped:
             continue  # meta / rating / artist / unknown are dropped
         s['exists_in_db'] = s['name'] in existing_all
+        if s['category'] == 'general' and s['confidence'] >= auto_accept_threshold:
+            auto_accept_candidates.append(s)
+            continue
         grouped[s['category']].append(s)
 
     for cat in grouped:
         grouped[cat].sort(key=lambda x: x['confidence'], reverse=True)
         grouped[cat] = grouped[cat][:top_k_per_category]
 
+    # Stash candidates in a way the shape documents but legacy callers can
+    # ignore. We keep the three core keys at the top level for backwards
+    # compat and add a 4th list.
+    grouped['auto_accept_candidates'] = auto_accept_candidates
     return grouped
 
 
@@ -308,6 +329,11 @@ def get_bulk_suggestions(
             existing_all=existing_all,
         )
         for category, items in grouped.items():
+            # `auto_accept_candidates` is returned by get_suggestions for the
+            # single-image route to apply; skip in bulk (bulk has no single
+            # image to attach to).
+            if category == 'auto_accept_candidates':
+                continue
             for item in items:
                 name = item['name']
                 stat = tag_stats.get(name)

app/static/js/view-modal.js

@@ -242,13 +242,109 @@ document.addEventListener('DOMContentLoaded', () => {
     try {
       const r = await fetch(`/image/${imageId}/suggestions`);
       const j = await r.json();
-      if (j.ok) renderSuggestions(j.suggestions || {}, imageId);
+      if (!j.ok) { clearSuggestions(); return; }
-      else clearSuggestions();
+      // Render regular suggestions first — it wipes innerHTML. Then prepend
+      // the auto-accepted block so both coexist.
+      renderSuggestions(j.suggestions || {}, imageId);
+      renderAutoAccepted(j.auto_accepted || [], imageId);
+      if ((j.auto_accepted || []).length > 0) {
+        suggestionsSection.style.display = '';
+      }
+      // Auto-accepted tags are now attached to the image — refresh the
+      // pill list so the user sees them in the Tags section.
+      if ((j.auto_accepted || []).length > 0 && typeof loadTags === 'function') {
+        await loadTags(imageId);
+      }
     } catch {
       clearSuggestions();
     }
   }
 
+  function renderAutoAccepted(items, imageId) {
+    if (!suggestionsList) return;
+    // Remove any prior auto-accepted block from the current modal view.
+    suggestionsList.querySelectorAll('.suggestions-auto-accepted').forEach(el => el.remove());
+    if (!items || items.length === 0) return;
+
+    const block = document.createElement('div');
+    block.className = 'suggestions-auto-accepted';
+    const label = document.createElement('div');
+    label.className = 'suggestions-category-label';
+    label.textContent = `Auto-accepted (${items.length})`;
+    block.appendChild(label);
+
+    const chips = document.createElement('div');
+    chips.className = 'suggestions-chips';
+    for (const item of items) {
+      chips.appendChild(buildAutoAcceptedChip(item, imageId));
+    }
+    block.appendChild(chips);
+    suggestionsList.prepend(block);
+  }
+
+  function buildAutoAcceptedChip(item, imageId) {
+    const chip = document.createElement('div');
+    chip.className = 'suggestion-chip suggestion-chip-auto';
+    chip.dataset.tagId = String(item.id);
+    chip.dataset.tagName = item.name;
+    chip.dataset.source = item.source || 'wd14';
+    chip.dataset.confidence = String(item.confidence || 0);
+    chip.title = `Auto-accepted from ${item.source || 'wd14'} at ${Math.round((item.confidence || 0) * 100)}%`;
+    const pct = Math.round((item.confidence || 0) * 100);
+    chip.innerHTML = `
+      <span class="sugg-body">
+        <span class="sugg-label">${escapeHtml(item.display_name || item.name)}</span>
+        <span class="sugg-confidence">${pct}%</span>
+      </span>
+      <button type="button" class="sugg-btn sugg-undo" aria-label="Undo (remove from this image)" title="Remove from this image">✕</button>
+      <button type="button" class="sugg-btn sugg-block" aria-label="Blocklist this tag" title="Never suggest '${escapeHtml(item.name)}' again — removes from all images">⊘</button>
+    `;
+    chip.querySelector('.sugg-undo')?.addEventListener('click', () => undoAutoAccepted(chip, imageId));
+    chip.querySelector('.sugg-block')?.addEventListener('click', () => blocklistSuggestion(chip, imageId));
+    return chip;
+  }
+
+  async function undoAutoAccepted(chip, imageId) {
+    if (chip.dataset.disabled === 'true') return;
+    chip.dataset.disabled = 'true';
+    chip.style.pointerEvents = 'none';
+    const name = chip.dataset.tagName;
+    try {
+      // Remove the tag from this image (uses the same remove endpoint the
+      // manual tag-pill '×' button uses).
+      const fd = new FormData();
+      fd.append('name', name);
+      const r = await fetch(`/image/${imageId}/tags/remove`, { method: 'POST', body: fd });
+      const j = await r.json();
+      if (!j.ok) {
+        chip.dataset.disabled = 'false';
+        chip.style.pointerEvents = '';
+        if (tagActionFeedback) tagActionFeedback.textContent = `Couldn't undo: ${j.error || 'error'}`;
+        return;
+      }
+      // Log the rejection for ML feedback so future threshold tuning sees this.
+      try {
+        await fetch(`/image/${imageId}/suggestions/reject`, {
+          method: 'POST',
+          headers: { 'Content-Type': 'application/json' },
+          body: JSON.stringify({
+            tag_name: name,
+            source: chip.dataset.source || 'wd14',
+            confidence: Number(chip.dataset.confidence) || 0,
+          }),
+        });
+      } catch {
+        // best-effort
+      }
+      if (tagActionFeedback) tagActionFeedback.textContent = `Removed '${name}'`;
+      animateChipOut(chip);
+      if (typeof loadTags === 'function') loadTags(imageId);
+    } catch {
+      chip.dataset.disabled = 'false';
+      chip.style.pointerEvents = '';
+    }
+  }
+
   function animateChipOut(chip) {
     chip.classList.add('leaving');
     setTimeout(() => { chip.remove(); }, 200);

app/static/style.css

@@ -1020,7 +1020,46 @@ header {
   border-color: rgba(255, 180, 0, 0.6);
 }
 
-/* Settings: blocklist textarea */
+/* Auto-accepted suggestion block (top of suggestions section) */
+.suggestions-auto-accepted {
+  margin-bottom: 0.75rem;
+  padding: 0.5rem;
+  background: rgba(139, 231, 139, 0.04);
+  border: 1px solid rgba(139, 231, 139, 0.2);
+  border-radius: 6px;
+}
+.suggestions-auto-accepted .suggestions-category-label {
+  color: #8be78b;
+  font-weight: 600;
+}
+.suggestion-chip.suggestion-chip-auto {
+  border-color: rgba(139, 231, 139, 0.3);
+  background: rgba(139, 231, 139, 0.08);
+}
+.suggestion-chip .sugg-btn.sugg-undo {
+  color: #e78b8b;
+  border-color: rgba(231, 139, 139, 0.35);
+}
+.suggestion-chip .sugg-btn.sugg-undo:hover {
+  background: rgba(231, 139, 139, 0.12);
+  border-color: rgba(231, 139, 139, 0.6);
+}
+
+/* Settings: threshold number input + blocklist textarea */
+.settings-input {
+  padding: 0.4rem 0.6rem;
+  border: 1px solid rgba(255, 255, 255, 0.2);
+  border-radius: 6px;
+  background: rgba(255, 255, 255, 0.08);
+  color: var(--text);
+  font-size: 0.95rem;
+}
+.settings-input:focus {
+  outline: none;
+  border-color: var(--btn-primary);
+  background: rgba(255, 255, 255, 0.12);
+}
+
 .settings-textarea {
   width: 100%;
   padding: 0.5rem 0.75rem;

app/tasks/maintenance.py

@@ -1,4 +1,58 @@
-"""Maintenance-queue Celery tasks. Currently empty — the
+"""Maintenance-queue Celery tasks.
-sync_character_fandoms task was removed on 2026-04-21 as part of the
+
-bare-name refactor (tag.fandom_id is now authoritative; no drift to sync).
+Currently holds the blocklist cleanup task that retroactively removes
+blocklisted tag names from the library. The earlier sync_character_fandoms
+task was removed on 2026-04-21 as part of the bare-name refactor.
 """
+import logging
+
+from celery import shared_task
+from sqlalchemy import text
+
+from app import db
+from app.models import Tag
+
+log = logging.getLogger(__name__)
+
+
+@shared_task(
+    name='app.tasks.maintenance.sweep_blocklisted_tag_from_images',
+    soft_time_limit=60,
+    time_limit=120,
+)
+def sweep_blocklisted_tag_from_images(name: str) -> dict:
+    """Remove a blocklisted tag name from every image that has it attached,
+    then delete the Tag row itself.
+
+    Scope is limited to kind='user' tags — that's the kind WD14's general
+    category gets materialized as when accepted, which is the vast majority
+    of blocklist hits. Character / fandom / artist / series / post / archive
+    tags sharing the same name are left alone: those are deliberate, curated
+    entities, and removing them silently because of a blocklist text match
+    would be destructive.
+
+    Returns a summary dict so the Celery result is introspectable.
+    """
+    tag = Tag.query.filter_by(kind='user', name=name).first()
+    if tag is None:
+        log.info("sweep_blocklisted_tag_from_images: no kind='user' tag named %r; nothing to do", name)
+        return {'name': name, 'tag_found': False, 'image_tags_deleted': 0, 'tag_deleted': False}
+
+    result = db.session.execute(
+        text("DELETE FROM image_tags WHERE tag_id = :tid"),
+        {'tid': tag.id},
+    )
+    rowcount = result.rowcount or 0
+    db.session.delete(tag)
+    db.session.commit()
+
+    log.info(
+        "sweep_blocklisted_tag_from_images: removed %r (tag_id=%d) from %d images",
+        name, tag.id, rowcount,
+    )
+    return {
+        'name': name,
+        'tag_found': True,
+        'image_tags_deleted': rowcount,
+        'tag_deleted': True,
+    }

app/templates/settings.html

@@ -458,6 +458,23 @@
 
     </section>
 
+    <section class="settings-section">
+      <h3>Auto-accept threshold (general tags)</h3>
+      <p class="settings-hint">
+        When the modal loads suggestions for an image, any general-category
+        tag with confidence at or above this threshold is auto-applied and
+        shown in a reviewable "Auto-accepted" block at the top of the
+        suggestions section. Lower values accept more tags (riskier);
+        values above 1.0 effectively disable the feature. Default is 0.95.
+      </p>
+      <div class="mt-sm">
+        <input type="number" id="autoAcceptThresholdInput" class="settings-input"
+               min="0" max="1.1" step="0.01" style="width: 7rem;" />
+        <button type="button" id="autoAcceptThresholdSaveBtn" class="btn secondary-btn">Save threshold</button>
+        <span id="autoAcceptThresholdStatus" class="settings-hint" style="margin-left: 0.5rem;"></span>
+      </div>
+    </section>
+
     <section class="settings-section">
       <h3>Suggestion blocklist</h3>
       <p class="settings-hint">
@@ -465,8 +482,11 @@
         post-canonicalization WD14 output (e.g. <code>1girl</code>,
         <code>breasts</code>). Blocklisted names never appear in the
         modal's auto-suggestions — either WD14 or embedding-similarity
-        sourced. You can also add entries inline by clicking the ⊘ button
+        sourced. Adding a name also enqueues a cleanup task that removes
-        on any suggestion chip.
+        the tag from every image that has it attached (kind='user' only;
+        deliberate character / artist / fandom tags sharing the same name
+        are left alone). You can also add entries inline by clicking the
+        ⊘ button on any suggestion chip.
       </p>
       <textarea id="suggestionBlocklistArea" class="settings-textarea" rows="10"
                 placeholder="1girl&#10;2girls&#10;breasts&#10;nipples&#10;penis"></textarea>
@@ -521,6 +541,54 @@
   saveBtn.addEventListener('click', save);
   load();
 })();
+
+(function(){
+  const input   = document.getElementById('autoAcceptThresholdInput');
+  const saveBtn = document.getElementById('autoAcceptThresholdSaveBtn');
+  const status  = document.getElementById('autoAcceptThresholdStatus');
+  if (!input || !saveBtn) return;
+
+  async function load() {
+    try {
+      const r = await fetch('/api/suggestions/config/auto-accept-threshold');
+      const j = await r.json();
+      if (j.ok) input.value = String(j.threshold);
+    } catch { /* leave empty */ }
+  }
+
+  async function save() {
+    saveBtn.disabled = true;
+    status.textContent = 'Saving…';
+    const threshold = parseFloat(input.value);
+    if (isNaN(threshold) || threshold < 0) {
+      status.textContent = 'Enter a number >= 0 (use > 1 to disable).';
+      saveBtn.disabled = false;
+      return;
+    }
+    try {
+      const r = await fetch('/api/suggestions/config/auto-accept-threshold', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({ threshold }),
+      });
+      const j = await r.json();
+      if (j.ok) {
+        status.textContent = `Saved (threshold = ${j.threshold}).`;
+        input.value = String(j.threshold);
+      } else {
+        status.textContent = 'Save failed: ' + (j.error || 'unknown');
+      }
+    } catch (e) {
+      status.textContent = 'Save failed: ' + e;
+    } finally {
+      saveBtn.disabled = false;
+      setTimeout(() => { status.textContent = ''; }, 4000);
+    }
+  }
+
+  saveBtn.addEventListener('click', save);
+  load();
+})();
 </script>
 {% endif %}