docs: update README for vision pipeline and Postgres stack

Reflect current state: Postgres+pgvector replaces SQLite, vision
pipeline (YOLO, CLIP, InsightFace, OCR) is shipped, card-grid browse
views for tags/colors/ratings/people, map view, duplicate detection,
and semantic search are all live. Remove completed items from future
features.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.gitignore

@@ -65,4 +65,4 @@ docker-compose.override.yml
 
 # Thumbnails
 /thumbs/
-/trash/
+/trash/backend/yolov8n.pt

README.md

@@ -6,21 +6,27 @@ A self-hosted, Docker-deployed photo management application inspired by Lightroo
 
 - **Photo Organization**: Browse photos in a timeline view with virtual scrolling for performance
 - **Thumbnail Generation**: Automatic thumbnail generation for all photo formats including RAW
-- **Metadata Extraction**: Full EXIF/XMP metadata extraction and search
+- **Metadata Extraction**: Full EXIF/XMP metadata extraction and GPS mapping
 - **Keyboard Shortcuts**: Lightroom-style keyboard navigation and actions
 - **File Support**: JPEG, PNG, RAW formats (CR2, CR3, NEF, ARW, etc.), HEIC/HEIF, and videos
 - **Heaps**: Temporary collections for organizing photos
-- **Tags & Ratings**: Organize with tags, star ratings, and color labels
+- **Tags & Ratings**: Organize with tags, star ratings, and color labels — each with a card-grid browse view that drills into a full Timeline detail
 - **Dark Mode**: Photography-optimized dark interface
+- **Vision Pipeline**: YOLO object detection, OCR text extraction, CLIP embeddings for semantic search, InsightFace face detection and clustering
+- **People View**: Browse identified people as cards, click to see all photos of a person
+- **Map View**: Browse GPS-tagged photos on an interactive Leaflet map
+- **Duplicate Detection**: Perceptual hash-based duplicate grouping with best-pick UI
+- **Semantic Search**: Natural-language photo search powered by CLIP embeddings
 
 ## Tech Stack
 
 ### Backend
 - Python 3.12 with FastAPI
-- SQLite with SQLAlchemy (async)
+- PostgreSQL + pgvector with SQLAlchemy (async) and Alembic migrations
 - Celery + Redis for background tasks
 - pyvips for fast thumbnail generation
 - ExifTool for metadata extraction
+- ONNX Runtime for vision models (YOLO, CLIP, InsightFace)
 
 ### Frontend
 - React 18 with TypeScript
@@ -161,9 +167,9 @@ The application consists of 5 Docker services:
 
 - **frontend**: React SPA served by Nginx
 - **backend**: FastAPI REST API
-- **worker**: Celery workers for background tasks
+- **worker**: Celery workers for background tasks (thumbnails, metadata, vision pipeline)
 - **redis**: Message broker for Celery
-- **db**: SQLite database (file-based)
+- **db**: PostgreSQL with pgvector extension (for CLIP/face embeddings)
 
 ## Keyboard Shortcuts
 
@@ -172,8 +178,7 @@ The application consists of 5 Docker services:
 | `←` `→` `↑` `↓` | Navigate photos |
 | `Space` | Quick preview |
 | `Enter` | Open loupe view |
-| `P` | Pick photo |
+| `T` | Add to active heap |
-| `X` | Reject photo |
 | `1-5` | Set star rating |
 | `Tab` | Toggle left sidebar |
 | `I` | Toggle metadata panel |
@@ -211,14 +216,12 @@ Source roots are managed by the UI / API (the database owns them). Edit
 - Handles 100,000+ photos efficiently
 - Virtual scrolling for smooth timeline navigation
 - Thumbnail generation at 10+ photos/second
-- SQLite FTS5 for fast full-text search
+- PostgreSQL full-text search with tsvector indexing
+- pgvector for fast nearest-neighbor embedding search
 
-## Future Features (Phase 2)
+## Future Features
 
-- AI-powered scene classification
+- Smart albums (auto-populated by saved filters)
-- Face detection and clustering
-- Smart albums
-- Duplicate detection
 - Export presets
 - Multi-user support
 

backend/Dockerfile

@@ -30,7 +30,7 @@ RUN pip install --no-cache-dir -r requirements.txt
 COPY . .
 
 # Create necessary directories
-RUN mkdir -p /data/thumbs /data/db /data/proxies /app/config
+RUN mkdir -p /data/thumbs /data/db /data/proxies /data/models /app/config
 
 # Expose port
 EXPOSE 8000

backend/alembic.ini

@@ -0,0 +1,48 @@
+# Alembic configuration for PhotoVault.
+#
+# The actual database URL is loaded at runtime by alembic/env.py from the
+# DATABASE_URL environment variable (with the async driver suffix stripped).
+# The placeholder below is only used for `alembic revision --autogenerate`
+# when no env var is set.
+
+[alembic]
+script_location = alembic
+prepend_sys_path = .
+version_path_separator = os
+sqlalchemy.url = postgresql+psycopg2://mulita:mulita@localhost:5432/mulita
+
+[post_write_hooks]
+
+[loggers]
+keys = root,sqlalchemy,alembic
+
+[handlers]
+keys = console
+
+[formatters]
+keys = generic
+
+[logger_root]
+level = WARNING
+handlers = console
+qualname =
+
+[logger_sqlalchemy]
+level = WARNING
+handlers =
+qualname = sqlalchemy.engine
+
+[logger_alembic]
+level = INFO
+handlers =
+qualname = alembic
+
+[handler_console]
+class = StreamHandler
+args = (sys.stderr,)
+level = NOTSET
+formatter = generic
+
+[formatter_generic]
+format = %(levelname)-5.5s [%(name)s] %(message)s
+datefmt = %H:%M:%S

backend/alembic/env.py

@@ -0,0 +1,96 @@
+"""
+Alembic environment for PhotoVault.
+
+Pulls DATABASE_URL from the environment so the same migrations work in
+docker compose and locally. Strips the async driver suffix because Alembic
+runs synchronously via psycopg2.
+
+Future-migration note
+---------------------
+Fresh installs run `Base.metadata.create_all` in `app.database.init_db`
+*before* migrations would normally apply, so any migration that adds a
+column / index / table to an object the model already declares will see
+that object already present. Write migrations defensively:
+
+    op.execute("ALTER TABLE photos ADD COLUMN IF NOT EXISTS new_col TEXT")
+    op.execute("CREATE INDEX IF NOT EXISTS ix_foo ON foo(bar)")
+
+For brand-new tables that the model also declares, the same applies — use
+`op.execute("CREATE TABLE IF NOT EXISTS ...")` or check first.
+"""
+from logging.config import fileConfig
+import os
+import sys
+from pathlib import Path
+
+from sqlalchemy import engine_from_config, pool
+from alembic import context
+
+# Make `app` importable from this script.
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from app.database import Base  # noqa: E402
+# Import all models so they're registered on Base.metadata for autogenerate.
+from app.models import (  # noqa: E402, F401
+    Photo,
+    Folder,
+    SourceRoot,
+    Tag,
+    Heap,
+    HeapPhoto,
+    Embedding,
+)
+
+config = context.config
+
+# Resolve DATABASE_URL from env. Strip async driver suffixes — Alembic
+# uses sync drivers.
+db_url = os.environ.get("DATABASE_URL") or config.get_main_option("sqlalchemy.url")
+if db_url:
+    if "+asyncpg" in db_url:
+        db_url = db_url.replace("+asyncpg", "+psycopg2")
+    elif db_url.startswith("postgresql://"):
+        db_url = db_url.replace("postgresql://", "postgresql+psycopg2://", 1)
+    elif "+aiosqlite" in db_url:
+        db_url = db_url.replace("+aiosqlite", "")
+    config.set_main_option("sqlalchemy.url", db_url)
+
+if config.config_file_name is not None:
+    fileConfig(config.config_file_name)
+
+target_metadata = Base.metadata
+
+
+def run_migrations_offline() -> None:
+    """Run migrations in 'offline' mode (emit SQL only)."""
+    url = config.get_main_option("sqlalchemy.url")
+    context.configure(
+        url=url,
+        target_metadata=target_metadata,
+        literal_binds=True,
+        dialect_opts={"paramstyle": "named"},
+    )
+    with context.begin_transaction():
+        context.run_migrations()
+
+
+def run_migrations_online() -> None:
+    """Run migrations against a live database."""
+    connectable = engine_from_config(
+        config.get_section(config.config_ini_section, {}),
+        prefix="sqlalchemy.",
+        poolclass=pool.NullPool,
+    )
+    with connectable.connect() as connection:
+        context.configure(
+            connection=connection,
+            target_metadata=target_metadata,
+        )
+        with context.begin_transaction():
+            context.run_migrations()
+
+
+if context.is_offline_mode():
+    run_migrations_offline()
+else:
+    run_migrations_online()

backend/alembic/script.py.mako

@@ -0,0 +1,26 @@
+"""${message}
+
+Revision ID: ${up_revision}
+Revises: ${down_revision | comma,n}
+Create Date: ${create_date}
+
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+${imports if imports else ""}
+
+# revision identifiers, used by Alembic.
+revision: str = ${repr(up_revision)}
+down_revision: Union[str, None] = ${repr(down_revision)}
+branch_labels: Union[str, Sequence[str], None] = ${repr(branch_labels)}
+depends_on: Union[str, Sequence[str], None] = ${repr(depends_on)}
+
+
+def upgrade() -> None:
+    ${upgrades if upgrades else "pass"}
+
+
+def downgrade() -> None:
+    ${downgrades if downgrades else "pass"}

backend/alembic/versions/0001_baseline.py

@@ -0,0 +1,27 @@
+"""baseline (empty)
+
+Revision ID: 0001_baseline
+Revises:
+Create Date: 2026-04-10
+
+The current schema is created by SQLAlchemy `Base.metadata.create_all` in
+`app.database.init_db()` on first boot. Alembic only owns deltas from
+PR3 onward. This baseline is intentionally empty so `alembic upgrade head`
+on a fresh DB simply creates the `alembic_version` table and stamps it.
+"""
+from typing import Sequence, Union
+
+
+# revision identifiers, used by Alembic.
+revision: str = "0001_baseline"
+down_revision: Union[str, None] = None
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    pass
+
+
+def downgrade() -> None:
+    pass

backend/alembic/versions/0002_extend_tags_for_vision.py

@@ -0,0 +1,85 @@
+"""extend tags for vision pipeline
+
+Revision ID: 0002_extend_tags
+Revises: 0001_baseline
+Create Date: 2026-04-10
+
+Add kind, source, representative_photo_id to tags table.
+Add confidence, bbox, source to photo_tags association.
+Switch uniqueness from (name) to (name, kind).
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+from sqlalchemy.dialects.postgresql import JSONB
+
+revision: str = "0002_extend_tags"
+down_revision: Union[str, None] = "0001_baseline"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ── tags table ────────────────────────────────────────────────────
+    op.execute("ALTER TABLE tags ADD COLUMN IF NOT EXISTS kind VARCHAR NOT NULL DEFAULT 'user'")
+    op.execute("ALTER TABLE tags ADD COLUMN IF NOT EXISTS source VARCHAR")
+    op.execute("ALTER TABLE tags ADD COLUMN IF NOT EXISTS representative_photo_id VARCHAR REFERENCES photos(id) ON DELETE SET NULL")
+
+    # Create index on kind for filtering
+    op.execute("CREATE INDEX IF NOT EXISTS ix_tags_kind ON tags(kind)")
+
+    # Drop old unique constraint on name (if it exists) and add (name, kind).
+    # SQLAlchemy create_all may have created either — handle both cases.
+    op.execute("""
+        DO $$
+        BEGIN
+            -- Drop the old single-column unique index/constraint if present.
+            IF EXISTS (
+                SELECT 1 FROM pg_indexes
+                WHERE tablename = 'tags' AND indexname = 'ix_tags_name'
+            ) THEN
+                DROP INDEX ix_tags_name;
+            END IF;
+
+            -- Some SQLAlchemy versions create a unique constraint directly.
+            IF EXISTS (
+                SELECT 1 FROM information_schema.table_constraints
+                WHERE table_name = 'tags' AND constraint_name = 'tags_name_key'
+            ) THEN
+                ALTER TABLE tags DROP CONSTRAINT tags_name_key;
+            END IF;
+        END $$;
+    """)
+
+    op.execute("""
+        DO $$
+        BEGIN
+            IF NOT EXISTS (
+                SELECT 1 FROM pg_constraint WHERE conname = 'uq_tags_name_kind'
+            ) THEN
+                ALTER TABLE tags ADD CONSTRAINT uq_tags_name_kind UNIQUE (name, kind);
+            END IF;
+        END $$;
+    """)
+
+    # ── photo_tags table ──────────────────────────────────────────────
+    op.execute("ALTER TABLE photo_tags ADD COLUMN IF NOT EXISTS confidence FLOAT")
+    op.execute("ALTER TABLE photo_tags ADD COLUMN IF NOT EXISTS bbox JSONB")
+    op.execute("ALTER TABLE photo_tags ADD COLUMN IF NOT EXISTS source VARCHAR")
+
+
+def downgrade() -> None:
+    # photo_tags columns
+    op.drop_column("photo_tags", "source")
+    op.drop_column("photo_tags", "bbox")
+    op.drop_column("photo_tags", "confidence")
+
+    # tags: restore old unique constraint
+    op.execute("ALTER TABLE tags DROP CONSTRAINT IF EXISTS uq_tags_name_kind")
+    op.execute("CREATE UNIQUE INDEX IF NOT EXISTS ix_tags_name ON tags(name)")
+
+    # tags columns
+    op.drop_column("tags", "representative_photo_id")
+    op.drop_column("tags", "source")
+    op.drop_column("tags", "kind")

backend/alembic/versions/0003_pgvector_embeddings.py

@@ -0,0 +1,52 @@
+"""pgvector embeddings
+
+Revision ID: 0003_pgvector_embeddings
+Revises: 0002_extend_tags
+Create Date: 2026-04-10
+
+Rewrite the embeddings table to use pgvector Vector(512) instead of
+LargeBinary. Add composite PK (photo_id, model), created_at, and
+HNSW index on vector column.
+"""
+from typing import Sequence, Union
+
+from alembic import op
+import sqlalchemy as sa
+
+revision: str = "0003_pgvector_embeddings"
+down_revision: Union[str, None] = "0002_extend_tags"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Drop the old placeholder table and recreate with pgvector types.
+    # No data to preserve — it was never populated.
+    op.execute("DROP TABLE IF EXISTS embeddings")
+    op.execute("""
+        CREATE TABLE embeddings (
+            photo_id VARCHAR NOT NULL REFERENCES photos(id) ON DELETE CASCADE,
+            model VARCHAR(64) NOT NULL,
+            vector vector(512),
+            created_at TIMESTAMPTZ DEFAULT now(),
+            PRIMARY KEY (photo_id, model)
+        )
+    """)
+    # HNSW index for cosine similarity search.
+    # Defer creation on large backfills — drop and recreate afterward.
+    op.execute("""
+        CREATE INDEX IF NOT EXISTS ix_embeddings_vector_hnsw
+        ON embeddings USING hnsw (vector vector_cosine_ops)
+    """)
+
+
+def downgrade() -> None:
+    op.execute("DROP TABLE IF EXISTS embeddings")
+    op.execute("""
+        CREATE TABLE embeddings (
+            photo_id VARCHAR NOT NULL REFERENCES photos(id) ON DELETE CASCADE,
+            model VARCHAR,
+            vector BYTEA,
+            PRIMARY KEY (photo_id)
+        )
+    """)

backend/alembic/versions/0004_ocr_text_and_fts.py

@@ -0,0 +1,82 @@
+"""ocr_text table and Postgres FTS
+
+Revision ID: 0004_ocr_fts
+Revises: 0003_pgvector_embeddings
+Create Date: 2026-04-10
+
+Create ocr_text table for storing OCR results. Add a tsvector column
+to photos for unified full-text search (filename + user_title +
+user_notes) with a GIN index. OCR text is rolled up into a materialized
+view or joined at query time.
+"""
+from typing import Sequence, Union
+
+from alembic import op
+
+revision: str = "0004_ocr_fts"
+down_revision: Union[str, None] = "0003_pgvector_embeddings"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ── ocr_text table ────────────────────────────────────────────────
+    op.execute("""
+        CREATE TABLE IF NOT EXISTS ocr_text (
+            id VARCHAR PRIMARY KEY,
+            photo_id VARCHAR NOT NULL REFERENCES photos(id) ON DELETE CASCADE,
+            text TEXT NOT NULL,
+            language VARCHAR(8) DEFAULT '',
+            confidence FLOAT,
+            bbox JSONB,
+            created_at TIMESTAMPTZ DEFAULT now()
+        )
+    """)
+    op.execute("CREATE INDEX IF NOT EXISTS ix_ocr_text_photo_id ON ocr_text(photo_id)")
+
+    # ── tsvector column on photos ─────────────────────────────────────
+    op.execute("ALTER TABLE photos ADD COLUMN IF NOT EXISTS search_vector tsvector")
+    op.execute("CREATE INDEX IF NOT EXISTS ix_photos_search_vector ON photos USING GIN (search_vector)")
+
+    # Trigger to auto-update search_vector on INSERT/UPDATE
+    op.execute("""
+        CREATE OR REPLACE FUNCTION photos_search_vector_update() RETURNS trigger AS $$
+        BEGIN
+            NEW.search_vector :=
+                setweight(to_tsvector('english', coalesce(NEW.filename, '')), 'A') ||
+                setweight(to_tsvector('english', coalesce(NEW.user_title, '')), 'A') ||
+                setweight(to_tsvector('english', coalesce(NEW.user_notes, '')), 'B');
+            RETURN NEW;
+        END
+        $$ LANGUAGE plpgsql;
+    """)
+    op.execute("""
+        DO $$
+        BEGIN
+            IF NOT EXISTS (
+                SELECT 1 FROM pg_trigger WHERE tgname = 'photos_search_vector_trigger'
+            ) THEN
+                CREATE TRIGGER photos_search_vector_trigger
+                BEFORE INSERT OR UPDATE OF filename, user_title, user_notes
+                ON photos
+                FOR EACH ROW
+                EXECUTE FUNCTION photos_search_vector_update();
+            END IF;
+        END $$;
+    """)
+
+    # Backfill existing rows
+    op.execute("""
+        UPDATE photos SET search_vector =
+            setweight(to_tsvector('english', coalesce(filename, '')), 'A') ||
+            setweight(to_tsvector('english', coalesce(user_title, '')), 'A') ||
+            setweight(to_tsvector('english', coalesce(user_notes, '')), 'B')
+    """)
+
+
+def downgrade() -> None:
+    op.execute("DROP TRIGGER IF EXISTS photos_search_vector_trigger ON photos")
+    op.execute("DROP FUNCTION IF EXISTS photos_search_vector_update()")
+    op.execute("DROP INDEX IF EXISTS ix_photos_search_vector")
+    op.execute("ALTER TABLE photos DROP COLUMN IF EXISTS search_vector")
+    op.execute("DROP TABLE IF EXISTS ocr_text")

backend/alembic/versions/0005_face_embeddings.py

@@ -0,0 +1,41 @@
+"""face_embeddings table
+
+Revision ID: 0005_face_embeddings
+Revises: 0004_ocr_fts
+Create Date: 2026-04-10
+
+Create face_embeddings table with pgvector Vector(128) for SFace
+recognition embeddings and HNSW index.
+"""
+from typing import Sequence, Union
+
+from alembic import op
+
+revision: str = "0005_face_embeddings"
+down_revision: Union[str, None] = "0004_ocr_fts"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    op.execute("""
+        CREATE TABLE IF NOT EXISTS face_embeddings (
+            id VARCHAR PRIMARY KEY,
+            photo_id VARCHAR NOT NULL REFERENCES photos(id) ON DELETE CASCADE,
+            bbox JSONB,
+            vector vector(128),
+            cluster_id VARCHAR REFERENCES tags(id) ON DELETE SET NULL,
+            quality FLOAT,
+            created_at TIMESTAMPTZ DEFAULT now()
+        )
+    """)
+    op.execute("CREATE INDEX IF NOT EXISTS ix_face_embeddings_photo_id ON face_embeddings(photo_id)")
+    op.execute("CREATE INDEX IF NOT EXISTS ix_face_embeddings_cluster_id ON face_embeddings(cluster_id)")
+    op.execute("""
+        CREATE INDEX IF NOT EXISTS ix_face_embeddings_vector_hnsw
+        ON face_embeddings USING hnsw (vector vector_cosine_ops)
+    """)
+
+
+def downgrade() -> None:
+    op.execute("DROP TABLE IF EXISTS face_embeddings")

backend/alembic/versions/0006_face_embeddings_512d.py

@@ -0,0 +1,39 @@
+"""face_embeddings vector 128 -> 512
+
+Revision ID: 0006_face_512d
+Revises: 0005_face_embeddings
+Create Date: 2026-04-10
+
+Resize face_embeddings.vector from Vector(128) to Vector(512) for
+ArcFace embeddings (InsightFace). Drops existing data and HNSW index,
+recreates both.
+"""
+from typing import Sequence, Union
+
+from alembic import op
+
+revision: str = "0006_face_512d"
+down_revision: Union[str, None] = "0005_face_embeddings"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # Drop index, truncate (old 128-d vectors are incompatible), resize
+    op.execute("DROP INDEX IF EXISTS ix_face_embeddings_vector_hnsw")
+    op.execute("DELETE FROM face_embeddings")
+    op.execute("ALTER TABLE face_embeddings ALTER COLUMN vector TYPE vector(512)")
+    op.execute("""
+        CREATE INDEX IF NOT EXISTS ix_face_embeddings_vector_hnsw
+        ON face_embeddings USING hnsw (vector vector_cosine_ops)
+    """)
+
+
+def downgrade() -> None:
+    op.execute("DROP INDEX IF EXISTS ix_face_embeddings_vector_hnsw")
+    op.execute("DELETE FROM face_embeddings")
+    op.execute("ALTER TABLE face_embeddings ALTER COLUMN vector TYPE vector(128)")
+    op.execute("""
+        CREATE INDEX IF NOT EXISTS ix_face_embeddings_vector_hnsw
+        ON face_embeddings USING hnsw (vector vector_cosine_ops)
+    """)

backend/app/config.py

@@ -29,6 +29,48 @@ class PerformanceSettings(BaseModel):
     db_pool_size: int = 20
     db_pool_recycle: int = 3600
 
+class EmbedderSettings(BaseModel):
+    """CLIP / SigLIP embedding model settings"""
+    name: str = "openclip_vitb32"
+    batch_size: int = 8
+
+class OCRSettings(BaseModel):
+    """PaddleOCR / rapidocr settings"""
+    enabled: bool = True
+    languages: list[str] = ["en"]
+    min_confidence: float = 0.5
+
+class DetectorSettings(BaseModel):
+    """YOLOv8n object detection settings"""
+    enabled: bool = True
+    min_confidence: float = 0.35
+    max_detections: int = 50
+
+class FacesSettings(BaseModel):
+    """YuNet + SFace face detection/recognition settings"""
+    enabled: bool = True
+    min_face_size: int = 40
+    recognition_threshold: float = 0.65
+    cluster_eps: float = 0.5
+
+class ClassifierSettings(BaseModel):
+    """CLIP zero-shot content classification settings"""
+    enabled: bool = True
+    min_confidence: float = 0.3
+
+class VisionSettings(BaseModel):
+    """AI vision pipeline settings. Disabled when running on SQLite
+    (pgvector is required for embedding storage)."""
+    enabled: bool = True
+    backend: str = "onnx"           # "onnx" | "rocm" (future)
+    models_dir: str = "/data/models"
+    embedder: EmbedderSettings = EmbedderSettings()
+    ocr: OCRSettings = OCRSettings()
+    detector: DetectorSettings = DetectorSettings()
+    faces: FacesSettings = FacesSettings()
+    classifier: ClassifierSettings = ClassifierSettings()
+    worker_concurrency: int = 2
+
 class MulitaConfig(BaseModel):
     """Main configuration from YAML file. Source roots and the discard
     workflow are owned by the database now — only operational settings
@@ -36,12 +78,15 @@ class MulitaConfig(BaseModel):
     thumbnails: ThumbnailSettings = ThumbnailSettings()
     scanner: ScannerSettings = ScannerSettings()
     performance: PerformanceSettings = PerformanceSettings()
+    vision: VisionSettings = VisionSettings()
 
 class Settings(BaseSettings):
     """Application settings"""
-    # Database
+    # Database — Postgres + pgvector by default. The SQLite escape hatch
+    # remains supported via the docker-compose.sqlite.yml override and by
+    # setting DATABASE_URL=sqlite+aiosqlite:///... in .env for local dev.
     database_url: str = Field(
-        default="sqlite+aiosqlite:///data/db/mulita.db",
+        default="postgresql+asyncpg://mulita:mulita@db:5432/mulita",
         env="DATABASE_URL"
     )
     
@@ -125,6 +170,10 @@ class Settings(BaseSettings):
     def performance(self) -> PerformanceSettings:
         return self.config.performance
 
+    @property
+    def vision(self) -> VisionSettings:
+        return self.config.vision
+
     class Config:
         env_file = ".env"
         case_sensitive = False

backend/app/database.py

@@ -1,38 +1,50 @@
 """
-Database configuration and session management
+Database configuration and session management.
+
+Schema management strategy
+--------------------------
+Postgres (default): Alembic owns schema deltas. `alembic upgrade head` is
+run before the app starts (in the container CMD). `init_db()` calls
+`create_all` afterward as the source of truth for fresh installs — it is
+idempotent for existing tables and creates any tables defined on
+`Base.metadata` that don't yet exist. Future Alembic migrations should be
+written defensively (`IF NOT EXISTS` etc.) so they remain safe to run on a
+fresh DB where `create_all` has already laid down the same objects.
+
+SQLite (escape hatch via docker-compose.sqlite.yml): no Alembic. The
+historical inline ALTER TABLE block stays in place so existing dev
+installs keep upgrading.
 """
 from sqlalchemy.ext.asyncio import AsyncSession, create_async_engine, async_sessionmaker
 from sqlalchemy.orm import declarative_base
-from sqlalchemy import event, text
+from sqlalchemy import text
 import logging
-import os
 from pathlib import Path
 
 from app.config import settings
 
 logger = logging.getLogger(__name__)
 
-# Create database directory if it doesn't exist
+_is_sqlite = settings.database_url.startswith("sqlite")
-db_path = Path(settings.database_url.replace("sqlite+aiosqlite:///", ""))
+_is_postgres = settings.database_url.startswith("postgresql")
-db_path.parent.mkdir(parents=True, exist_ok=True)
 
-# Create async engine
+if _is_sqlite:
-# SQLite doesn't support pool configuration
+    db_path = Path(settings.database_url.replace("sqlite+aiosqlite:///", ""))
-if "sqlite" in settings.database_url:
+    db_path.parent.mkdir(parents=True, exist_ok=True)
     engine = create_async_engine(
         settings.database_url,
-        echo=False,  # Set to True for SQL debugging
+        echo=False,
         connect_args={
-            "check_same_thread": False,  # SQLite specific
+            "check_same_thread": False,
-            "timeout": 30
+            "timeout": 30,
-        }
+        },
     )
 else:
     engine = create_async_engine(
         settings.database_url,
-        echo=False,  # Set to True for SQL debugging
+        echo=False,
         pool_size=settings.performance.db_pool_size,
-        pool_recycle=settings.performance.db_pool_recycle
+        pool_recycle=settings.performance.db_pool_recycle,
     )
 
 # Create async session factory
@@ -59,26 +71,32 @@ async def init_db():
         # Import all models to register them with Base
         from app.models import Photo, Folder, SourceRoot, Tag, PhotoTag, Heap, HeapPhoto, Embedding
 
+        # Postgres: ensure pgvector is available before create_all touches
+        # any Vector columns (added in later PRs but the extension is cheap
+        # and idempotent to create now).
+        if _is_postgres:
+            await conn.execute(text("CREATE EXTENSION IF NOT EXISTS vector"))
+
         # Create all tables. Note: create_all only creates *missing* tables —
         # it does NOT add new columns to existing tables when the model gains
-        # them. Anything new on an existing table needs an explicit ALTER
+        # them. On Postgres, Alembic handles deltas; on SQLite, the inline
-        # below.
+        # ALTER block below is the legacy fallback.
         await conn.run_sync(Base.metadata.create_all)
 
-        # Enable WAL mode for SQLite (better concurrency)
+        gps_columns_added = False
-        if "sqlite" in settings.database_url:
+
+        if _is_sqlite:
+            # WAL mode for better concurrency.
             await conn.execute(text("PRAGMA journal_mode=WAL"))
             await conn.execute(text("PRAGMA synchronous=NORMAL"))
             await conn.execute(text("PRAGMA cache_size=10000"))
             await conn.execute(text("PRAGMA temp_store=MEMORY"))
 
-        # ── Idempotent column adds ────────────────────────────────────────
+            # ── Idempotent column adds (SQLite only) ─────────────────────
-        # The project does not use Alembic; we lean on create_all + a small
+            # SQLite supports ADD COLUMN but not "IF NOT EXISTS" for
-        # set of inline ALTER TABLE statements for the columns we've added
+            # columns, so introspect via PRAGMA first. Each entry is
-        # post-launch. SQLite supports ADD COLUMN but not "IF NOT EXISTS"
+            # (column_name, ALTER statement). Add new columns at the
-        # for columns, so introspect via PRAGMA first. Each entry is
+            # bottom. On Postgres these live in Alembic migrations.
-        # (column_name, ALTER statement). Add new columns at the bottom.
-        if "sqlite" in settings.database_url:
             existing_cols = {
                 row[1]
                 for row in (
@@ -94,18 +112,12 @@ async def init_db():
                 ("latitude", "ALTER TABLE photos ADD COLUMN latitude REAL"),
                 ("longitude", "ALTER TABLE photos ADD COLUMN longitude REAL"),
             ]
-            # Track whether the GPS columns were just added so we can kick
-            # off a one-shot backfill of existing photos at the end of init.
-            gps_columns_added = False
             for col_name, alter_sql in pending_alters:
                 if col_name not in existing_cols:
                     logger.info(f"Adding photos.{col_name} column")
                     await conn.execute(text(alter_sql))
                     if col_name in ("latitude", "longitude"):
                         gps_columns_added = True
-            # Indexes for the new duplicate-detection columns. CREATE INDEX
-            # IF NOT EXISTS is supported on SQLite so this is safe to run
-            # every startup.
             await conn.execute(
                 text("CREATE INDEX IF NOT EXISTS ix_photos_phash ON photos(phash)")
             )
@@ -124,11 +136,12 @@ async def init_db():
 
         logger.info("Database initialized successfully")
 
-        # If we just introduced the GPS columns on an existing install, kick
+        # If we just introduced the GPS columns on an existing SQLite
-        # off a one-shot backfill so the Map view is populated without a
+        # install, kick off a one-shot backfill so the Map view is
-        # manual full re-scan. Imported lazily to avoid pulling Celery into
+        # populated without a manual full re-scan. Postgres installs are
-        # the import graph for non-worker processes that don't need it.
+        # always fresh (no SQLite→PG migration path), so this code path
-        if "sqlite" in settings.database_url and gps_columns_added:
+        # is SQLite-only.
+        if _is_sqlite and gps_columns_added:
             try:
                 from app.tasks.scan import backfill_gps
                 backfill_gps.delay()
@@ -137,8 +150,9 @@ async def init_db():
                 logger.warning(f"Could not queue backfill_gps task: {e}")
 
 async def create_fts_table():
-    """Create Full-Text Search table for SQLite"""
+    """Create Full-Text Search table for SQLite. On Postgres this is
-    if "sqlite" in settings.database_url:
+    replaced by a tsvector column on the photos table (added in PR5)."""
+    if _is_sqlite:
         async with engine.begin() as conn:
             # Create FTS5 virtual table for full-text search
             await conn.execute(text("""
@@ -151,4 +165,4 @@ async def create_fts_table():
                     tokenize='unicode61'
                 )
             """))
-            logger.info("FTS5 table created successfully")
+            logger.info("FTS5 table created successfully")

backend/app/main.py

@@ -11,7 +11,7 @@ import os
 
 from app.config import settings
 from app.database import init_db
-from app.routers import photos, folders, heaps, tags, discard, library
+from app.routers import photos, folders, heaps, tags, discard, library, search
 from app.services.scanner import start_initial_scan, bootstrap_default_source_root
 from app.services.cleanup import cleanup_data_integrity
 
@@ -91,6 +91,7 @@ app.include_router(heaps.router, prefix="/api/v1/heaps", tags=["heaps"])
 app.include_router(tags.router, prefix="/api/v1/tags", tags=["tags"])
 app.include_router(discard.router, prefix="/api/v1/discard", tags=["discard"])
 app.include_router(library.router, prefix="/api/v1/library", tags=["library"])
+app.include_router(search.router, prefix="/api/v1/photos/search", tags=["search"])
 
 @app.get("/")
 async def root():

backend/app/models/__init__.py

@@ -6,6 +6,8 @@ from app.models.folders import Folder, SourceRoot
 from app.models.tags import Tag, PhotoTag
 from app.models.heaps import Heap, HeapPhoto
 from app.models.embeddings import Embedding
+from app.models.ocr_text import OCRText
+from app.models.face_embedding import FaceEmbedding
 
 __all__ = [
     'Photo',
@@ -15,5 +17,7 @@ __all__ = [
     'PhotoTag',
     'Heap',
     'HeapPhoto',
-    'Embedding'
+    'Embedding',
+    'OCRText',
+    'FaceEmbedding',
 ]

backend/app/models/embeddings.py

@@ -1,17 +1,19 @@
 """
-Embedding model definition (placeholder for AI features)
+Embedding model — stores CLIP/SigLIP image embeddings via pgvector.
+
+Composite PK (photo_id, model) allows re-embedding with newer models
+without clobbering old vectors.
 """
-from sqlalchemy import Column, String, ForeignKey, LargeBinary
+from sqlalchemy import Column, String, ForeignKey, DateTime, func
-import uuid
+from pgvector.sqlalchemy import Vector
 
 from app.database import Base
 
+
 class Embedding(Base):
-    """
-    Placeholder table for future AI embeddings (CLIP, face recognition, etc.)
-    """
     __tablename__ = 'embeddings'
-    
+
     photo_id = Column(String, ForeignKey('photos.id', ondelete='CASCADE'), primary_key=True)
-    model = Column(String)  # e.g., 'clip-vit-b32', 'face-recognition', etc.
+    model = Column(String(64), primary_key=True)  # e.g. 'openclip_vitb32'
-    vector = Column(LargeBinary)  # raw float32 bytes for embedding vector
+    vector = Column(Vector(512))                   # OpenCLIP ViT-B/32 → 512-d
+    created_at = Column(DateTime(timezone=True), server_default=func.now())

backend/app/models/face_embedding.py

@@ -0,0 +1,24 @@
+"""
+Face embedding model — stores per-face detection + recognition vectors.
+
+cluster_id FKs to tags.id where kind='face_cluster'. Null means
+unclustered (will be assigned by recluster_faces).
+"""
+from sqlalchemy import Column, String, Float, ForeignKey, DateTime, func
+from sqlalchemy.dialects.postgresql import JSONB
+from pgvector.sqlalchemy import Vector
+import uuid
+
+from app.database import Base
+
+
+class FaceEmbedding(Base):
+    __tablename__ = 'face_embeddings'
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    photo_id = Column(String, ForeignKey('photos.id', ondelete='CASCADE'), nullable=False, index=True)
+    bbox = Column(JSONB)           # [x1, y1, x2, y2] normalized 0-1
+    vector = Column(Vector(512))   # ArcFace → 512-d
+    cluster_id = Column(String, ForeignKey('tags.id', ondelete='SET NULL'), nullable=True, index=True)
+    quality = Column(Float)
+    created_at = Column(DateTime(timezone=True), server_default=func.now())

backend/app/models/ocr_text.py

@@ -0,0 +1,20 @@
+"""
+OCR text model — stores text regions extracted from photos via rapidocr.
+"""
+from sqlalchemy import Column, String, Float, ForeignKey, Text, DateTime, func
+from sqlalchemy.dialects.postgresql import JSONB
+import uuid
+
+from app.database import Base
+
+
+class OCRText(Base):
+    __tablename__ = 'ocr_text'
+
+    id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
+    photo_id = Column(String, ForeignKey('photos.id', ondelete='CASCADE'), nullable=False, index=True)
+    text = Column(Text, nullable=False)
+    language = Column(String(8), default='')
+    confidence = Column(Float)
+    bbox = Column(JSONB)  # [x1, y1, x2, y2] normalized 0-1
+    created_at = Column(DateTime(timezone=True), server_default=func.now())

backend/app/models/tags.py

@@ -1,8 +1,14 @@
 """
-Tag model definitions
+Tag model definitions.
+
+Tags are unified across user-created tags, ML-detected objects, scene
+labels, and face clusters via the `kind` column. The `photo_tags`
+association carries per-photo ML metadata (confidence, bounding box,
+source model).
 """
-from sqlalchemy import Column, String, ForeignKey, Table, Index
+from sqlalchemy import Column, String, Float, ForeignKey, Table, Index, UniqueConstraint
 from sqlalchemy.orm import relationship
+from sqlalchemy.dialects.postgresql import JSONB
 import uuid
 
 from app.database import Base
@@ -13,20 +19,41 @@ photo_tags = Table(
     Base.metadata,
     Column('photo_id', String, ForeignKey('photos.id', ondelete='CASCADE'), primary_key=True),
     Column('tag_id', String, ForeignKey('tags.id', ondelete='CASCADE'), primary_key=True),
+    # ML metadata — null for user-applied tags
+    Column('confidence', Float, nullable=True),
+    Column('bbox', JSONB, nullable=True),       # [x1, y1, x2, y2] normalized 0-1
+    Column('source', String, nullable=True),     # e.g. "vision:yolov8n", "vision:sface"
     Index('ix_photo_tags_photo_id', 'photo_id'),
     Index('ix_photo_tags_tag_id', 'tag_id'),
 )
 
 class Tag(Base):
     __tablename__ = 'tags'
-    
+    __table_args__ = (
+        UniqueConstraint('name', 'kind', name='uq_tags_name_kind'),
+    )
+
     id = Column(String, primary_key=True, default=lambda: str(uuid.uuid4()))
-    name = Column(String, unique=True, nullable=False, index=True)
+    name = Column(String, nullable=False, index=True)
     color = Column(String)  # Hex color code for UI display
-    
+
+    # Tag classification
+    kind = Column(String, nullable=False, default='user', index=True)
+    # kind values: 'user' | 'object' | 'scene' | 'face_cluster'
+
+    # Which model produced this tag (null for user-created)
+    source = Column(String, nullable=True)
+    # e.g. "vision:yolov8n", "vision:sface", null
+
+    # For face clusters: the photo used as the cluster representative thumbnail
+    representative_photo_id = Column(
+        String, ForeignKey('photos.id', ondelete='SET NULL'), nullable=True
+    )
+
     # Relationships
     photos = relationship("Photo", secondary=photo_tags, backref="tags")
 
+
 class PhotoTag:
     """Helper class for photo-tag associations (not a table model)"""
-    pass
+    pass

backend/app/routers/search.py

@@ -0,0 +1,74 @@
+"""
+Search API router — unified hybrid search endpoint.
+"""
+from typing import Optional
+from fastapi import APIRouter, Depends
+from pydantic import BaseModel
+from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import select
+
+from app.database import get_db
+from app.models import Photo
+from app.services.search import hybrid_search
+
+router = APIRouter()
+
+
+class SearchRequest(BaseModel):
+    q: Optional[str] = None
+    filters: Optional[dict] = None
+    limit: int = 50
+    offset: int = 0
+
+
+@router.post("")
+async def search_photos(body: SearchRequest, db: AsyncSession = Depends(get_db)):
+    """Unified search endpoint. Every query runs hybrid (FTS + semantic)
+    by default — the user never picks a mode.
+
+    Filters:
+      - tag_ids: list of tag IDs (any kind: user, object, face_cluster)
+      - date_from / date_to: ISO date strings
+    """
+    filters = body.filters or {}
+
+    results = await hybrid_search(
+        db=db,
+        q=body.q,
+        tag_ids=filters.get("tag_ids"),
+        date_from=filters.get("date_from"),
+        date_to=filters.get("date_to"),
+        limit=body.limit,
+        offset=body.offset,
+    )
+
+    if not results:
+        return {"results": [], "total": 0}
+
+    # Hydrate with photo data
+    photo_ids = [r["photo_id"] for r in results]
+    stmt = select(Photo).where(Photo.id.in_(photo_ids))
+    rows = (await db.execute(stmt)).scalars().all()
+    photo_map = {p.id: p for p in rows}
+
+    hydrated = []
+    for r in results:
+        photo = photo_map.get(r["photo_id"])
+        if not photo:
+            continue
+        hydrated.append({
+            "id": photo.id,
+            "filename": photo.filename,
+            "filepath": photo.filepath,
+            "media_type": photo.media_type,
+            "width": photo.width,
+            "height": photo.height,
+            "taken_at": photo.taken_at.isoformat() if photo.taken_at else None,
+            "rating": photo.rating,
+            "color_label": photo.color_label,
+            "thumb_small": photo.thumb_small,
+            "thumb_medium": photo.thumb_medium,
+            "score": r["score"],
+        })
+
+    return {"results": hydrated, "total": len(hydrated)}

backend/app/routers/tags.py

@@ -1,10 +1,15 @@
 """
-Tags API router
+Tags API router.
+
+Unified across user tags, ML-detected objects, and face clusters via
+the `kind` query parameter. Default behaviour (no kind filter) returns
+all tags — the frontend's "Hide auto-generated tags" toggle filters
+client-side or passes `kind=user`.
 """
 from typing import Optional
-from fastapi import APIRouter, Depends, HTTPException
+from fastapi import APIRouter, Depends, HTTPException, Query
 from pydantic import BaseModel
-from sqlalchemy import select, func, insert, delete
+from sqlalchemy import select, func, update
 from sqlalchemy.ext.asyncio import AsyncSession
 
 from app.database import get_db
@@ -19,6 +24,7 @@ router = APIRouter()
 class TagCreate(BaseModel):
     name: str
     color: Optional[str] = None
+    kind: str = "user"
 
 
 class TagUpdate(BaseModel):
@@ -26,24 +32,35 @@ class TagUpdate(BaseModel):
     color: Optional[str] = None
 
 
+class TagMerge(BaseModel):
+    target_id: str  # tag to merge INTO
+
+
 # ── Endpoints ─────────────────────────────────────────────────────────────
 
 @router.get("")
-async def list_tags(db: AsyncSession = Depends(get_db)):
+async def list_tags(
-    """List all tags with their photo counts."""
+    kind: Optional[str] = Query(None, description="Filter by kind: user, object, scene, face_cluster"),
+    db: AsyncSession = Depends(get_db),
+):
+    """List all tags with their photo counts, optionally filtered by kind."""
     count_subq = (
         select(
             photo_tags.c.tag_id,
             func.count(photo_tags.c.photo_id).label("photo_count"),
+            func.min(photo_tags.c.photo_id).label("first_photo_id"),
         )
         .group_by(photo_tags.c.tag_id)
         .subquery()
     )
     stmt = (
-        select(Tag, count_subq.c.photo_count)
+        select(Tag, count_subq.c.photo_count, count_subq.c.first_photo_id)
         .outerjoin(count_subq, Tag.id == count_subq.c.tag_id)
-        .order_by(Tag.name.asc())
     )
+    if kind:
+        stmt = stmt.where(Tag.kind == kind)
+    stmt = stmt.order_by(Tag.name.asc())
+
     result = await db.execute(stmt)
     rows = result.all()
 
@@ -52,38 +69,48 @@ async def list_tags(db: AsyncSession = Depends(get_db)):
             "id": tag.id,
             "name": tag.name,
             "color": tag.color,
+            "kind": tag.kind,
+            "source": tag.source,
+            "representative_photo_id": tag.representative_photo_id or first_photo_id,
             "photo_count": int(count or 0),
         }
-        for tag, count in rows
+        for tag, count, first_photo_id in rows
     ]
 
 
 @router.post("", status_code=201)
 async def create_tag(body: TagCreate, db: AsyncSession = Depends(get_db)):
-    """Create a new tag. Names are unique — re-creating an existing name
+    """Create a new tag. The (name, kind) pair is unique — re-creating an
-    returns the existing row instead of erroring (idempotent for the
+    existing pair returns the existing row (idempotent for autocomplete)."""
-    autocomplete UI flow)."""
     name = (body.name or "").strip()
     if not name:
         raise HTTPException(status_code=400, detail="Tag name is required")
 
-    existing = await db.execute(select(Tag).where(Tag.name == name))
+    existing = await db.execute(
+        select(Tag).where(Tag.name == name, Tag.kind == body.kind)
+    )
     found = existing.scalar_one_or_none()
     if found:
-        return {"id": found.id, "name": found.name, "color": found.color, "photo_count": 0}
+        return {
+            "id": found.id, "name": found.name, "color": found.color,
+            "kind": found.kind, "photo_count": 0,
+        }
 
-    tag = Tag(name=name, color=body.color)
+    tag = Tag(name=name, color=body.color, kind=body.kind)
     db.add(tag)
     await db.commit()
     await db.refresh(tag)
-    return {"id": tag.id, "name": tag.name, "color": tag.color, "photo_count": 0}
+    return {
+        "id": tag.id, "name": tag.name, "color": tag.color,
+        "kind": tag.kind, "photo_count": 0,
+    }
 
 
 @router.patch("/{tag_id}")
 async def update_tag(
     tag_id: str, body: TagUpdate, db: AsyncSession = Depends(get_db)
 ):
-    """Rename or recolor a tag."""
+    """Rename or recolor a tag (works for any kind — user, object, face_cluster)."""
     result = await db.execute(select(Tag).where(Tag.id == tag_id))
     tag = result.scalar_one_or_none()
     if not tag:
@@ -99,7 +126,52 @@ async def update_tag(
 
     await db.commit()
     await db.refresh(tag)
-    return {"id": tag.id, "name": tag.name, "color": tag.color}
+    return {"id": tag.id, "name": tag.name, "color": tag.color, "kind": tag.kind}
+
+
+@router.post("/{tag_id}/merge")
+async def merge_tag(
+    tag_id: str, body: TagMerge, db: AsyncSession = Depends(get_db)
+):
+    """Merge tag_id INTO target_id. All photo associations from the source
+    tag are moved to the target, then the source tag is deleted.
+
+    Useful for merging auto-detected face clusters (e.g. "Person 3" → "Alice")
+    or merging duplicate object labels."""
+    if tag_id == body.target_id:
+        raise HTTPException(status_code=400, detail="Cannot merge a tag into itself")
+
+    source = (await db.execute(select(Tag).where(Tag.id == tag_id))).scalar_one_or_none()
+    target = (await db.execute(select(Tag).where(Tag.id == body.target_id))).scalar_one_or_none()
+    if not source:
+        raise HTTPException(status_code=404, detail="Source tag not found")
+    if not target:
+        raise HTTPException(status_code=404, detail="Target tag not found")
+
+    # Move photo associations: update tag_id from source → target.
+    # Skip any that would violate the PK (photo already tagged with target).
+    existing_target_photos = select(photo_tags.c.photo_id).where(
+        photo_tags.c.tag_id == body.target_id
+    )
+    await db.execute(
+        update(photo_tags)
+        .where(
+            photo_tags.c.tag_id == tag_id,
+            photo_tags.c.photo_id.notin_(existing_target_photos),
+        )
+        .values(tag_id=body.target_id)
+    )
+    # Delete remaining source associations (duplicates that couldn't move)
+    from sqlalchemy import delete as sa_delete
+    await db.execute(
+        sa_delete(photo_tags).where(photo_tags.c.tag_id == tag_id)
+    )
+
+    # Delete source tag
+    await db.delete(source)
+    await db.commit()
+
+    return {"merged_into": target.id, "target_name": target.name}
 
 
 @router.delete("/{tag_id}", status_code=204)

backend/app/services/search.py

@@ -0,0 +1,140 @@
+"""
+Unified search service — hybrid FTS + semantic (RRF) search.
+
+Phase 1 (PR4): semantic-only via pgvector cosine similarity.
+Phase 2 (PR5): adds FTS via tsvector, enables RRF fusion.
+"""
+import logging
+from typing import Optional
+
+import numpy as np
+from sqlalchemy import select, text, func
+from sqlalchemy.ext.asyncio import AsyncSession
+
+from app.models import Photo
+from app.models.embeddings import Embedding
+from app.config import settings
+
+logger = logging.getLogger(__name__)
+
+
+async def hybrid_search(
+    db: AsyncSession,
+    q: Optional[str] = None,
+    tag_ids: Optional[list[str]] = None,
+    date_from: Optional[str] = None,
+    date_to: Optional[str] = None,
+    limit: int = 50,
+    offset: int = 0,
+) -> list[dict]:
+    """Run hybrid search (FTS + semantic) with RRF fusion.
+
+    Currently semantic-only; FTS leg added in PR5.
+    """
+    model_name = settings.vision.embedder.name
+    results = {}
+
+    # ── Semantic search (CLIP text → pgvector cosine) ─────────────────
+    if q:
+        try:
+            from app.services.vision.registry import registry
+            embedder = registry.get_embedder()
+            query_vec = embedder.embed_text(q)
+
+            # pgvector cosine distance: <=> returns distance (lower = closer)
+            vec_str = "[" + ",".join(str(float(v)) for v in query_vec) + "]"
+            stmt = text("""
+                SELECT e.photo_id,
+                       (e.vector <=> :qvec::vector) AS distance
+                FROM embeddings e
+                WHERE e.model = :model
+                ORDER BY e.vector <=> :qvec::vector
+                LIMIT 200
+            """)
+            rows = (await db.execute(stmt, {"qvec": vec_str, "model": model_name})).fetchall()
+
+            for rank, (photo_id, distance) in enumerate(rows):
+                if photo_id not in results:
+                    results[photo_id] = {"semantic_rank": rank, "fts_rank": None}
+                else:
+                    results[photo_id]["semantic_rank"] = rank
+
+        except Exception as e:
+            logger.warning("Semantic search failed (models may not be loaded): %s", e)
+
+    # ── FTS search (photos.search_vector + ocr_text) ────────────────
+    if q:
+        try:
+            fts_stmt = text("""
+                SELECT id, ts_rank(search_vector, plainto_tsquery('english', :q)) AS rank
+                FROM photos
+                WHERE search_vector @@ plainto_tsquery('english', :q)
+                UNION
+                SELECT o.photo_id AS id,
+                       MAX(o.confidence) AS rank
+                FROM ocr_text o
+                WHERE to_tsvector('english', o.text) @@ plainto_tsquery('english', :q)
+                GROUP BY o.photo_id
+                ORDER BY rank DESC
+                LIMIT 200
+            """)
+            fts_rows = (await db.execute(fts_stmt, {"q": q})).fetchall()
+            for rank, (photo_id, score) in enumerate(fts_rows):
+                if photo_id not in results:
+                    results[photo_id] = {"semantic_rank": None, "fts_rank": rank}
+                else:
+                    results[photo_id]["fts_rank"] = rank
+        except Exception as e:
+            logger.warning("FTS search failed: %s", e)
+
+    # ── RRF fusion ────────────────────────────────────────────────────
+    k = 60
+    scored = []
+    for photo_id, ranks in results.items():
+        score = 0.0
+        if ranks["semantic_rank"] is not None:
+            score += 1.0 / (k + ranks["semantic_rank"])
+        if ranks.get("fts_rank") is not None:
+            score += 1.0 / (k + ranks["fts_rank"])
+        scored.append((photo_id, score))
+
+    scored.sort(key=lambda x: -x[1])
+
+    # If no text query, fall back to recent photos
+    if not q:
+        if tag_ids:
+            from app.models.tags import photo_tags
+            # Subquery to get distinct photo_ids matching the tag filter
+            sub = select(photo_tags.c.photo_id).where(
+                photo_tags.c.tag_id.in_(tag_ids)
+            ).distinct().subquery()
+            stmt = select(Photo.id).join(sub, Photo.id == sub.c.photo_id)
+        else:
+            stmt = select(Photo.id)
+        stmt = stmt.order_by(Photo.added_at.desc())
+        if date_from:
+            stmt = stmt.where(Photo.taken_at >= date_from)
+        if date_to:
+            stmt = stmt.where(Photo.taken_at <= date_to)
+        stmt = stmt.offset(offset).limit(limit)
+        rows = (await db.execute(stmt)).fetchall()
+        return [{"photo_id": row[0], "score": 0.0} for row in rows]
+
+    # Apply filters to scored results
+    photo_ids = [pid for pid, _ in scored]
+    if not photo_ids:
+        return []
+
+    # Filter by tags if requested
+    if tag_ids:
+        from app.models.tags import photo_tags
+        stmt = select(photo_tags.c.photo_id).where(
+            photo_tags.c.photo_id.in_(photo_ids),
+            photo_tags.c.tag_id.in_(tag_ids),
+        ).distinct()
+        valid_ids = {row[0] for row in (await db.execute(stmt)).fetchall()}
+        scored = [(pid, s) for pid, s in scored if pid in valid_ids]
+
+    # Paginate
+    page = scored[offset : offset + limit]
+    return [{"photo_id": pid, "score": score} for pid, score in page]

backend/app/services/vision/__init__.py

@@ -0,0 +1,7 @@
+"""
+Vision pipeline services — embedding, OCR, object detection, face recognition.
+
+All inference is done through the ModelRegistry singleton, which lazy-loads
+ONNX Runtime sessions on first use and caches them for the lifetime of the
+worker process.
+"""

backend/app/services/vision/base.py

@@ -0,0 +1,105 @@
+"""
+Abstract base classes for vision backends.
+
+Each ABC defines the contract a backend must satisfy. The default
+implementation is ONNXBackend (onnx_backend.py). A ROCm backend can be
+added later by subclassing these ABCs and registering via
+settings.vision.backend.
+"""
+from abc import ABC, abstractmethod
+from dataclasses import dataclass
+import numpy as np
+
+
+@dataclass
+class DetectionBox:
+    """A single object detection result."""
+    label: str
+    confidence: float
+    bbox: list[float]  # [x1, y1, x2, y2] normalized 0-1
+
+
+@dataclass
+class OCRResult:
+    """A single OCR text region."""
+    text: str
+    confidence: float
+    bbox: list[float]  # [x1, y1, x2, y2] normalized 0-1
+    language: str = ""
+
+
+@dataclass
+class FaceDetection:
+    """A detected face with its recognition embedding."""
+    bbox: list[float]  # [x1, y1, x2, y2] normalized 0-1
+    embedding: np.ndarray  # float32 vector (128-d for SFace)
+    quality: float
+
+
+@dataclass
+class ClassificationResult:
+    """A content-type classification."""
+    label: str
+    confidence: float
+
+
+class Embedder(ABC):
+    """Generates image and text embeddings (e.g. OpenCLIP ViT-B/32)."""
+
+    @abstractmethod
+    def embed_image(self, image: np.ndarray) -> np.ndarray:
+        """Return a normalized float32 embedding vector for an RGB image."""
+        ...
+
+    @abstractmethod
+    def embed_text(self, text: str) -> np.ndarray:
+        """Return a normalized float32 embedding vector for a text query."""
+        ...
+
+    @property
+    @abstractmethod
+    def dim(self) -> int:
+        """Dimensionality of the output embedding."""
+        ...
+
+
+class OCREngine(ABC):
+    """Extracts text from images (e.g. rapidocr-onnxruntime)."""
+
+    @abstractmethod
+    def run(self, image: np.ndarray) -> list[OCRResult]:
+        """Return OCR results for an RGB image."""
+        ...
+
+
+class ObjectDetector(ABC):
+    """Detects objects in images (e.g. YOLOv8n)."""
+
+    @abstractmethod
+    def detect(self, image: np.ndarray) -> list[DetectionBox]:
+        """Return detections for an RGB image."""
+        ...
+
+
+class ContentClassifier(ABC):
+    """Classifies images into content types (screenshot, document, etc.)."""
+
+    @abstractmethod
+    def classify(self, image: np.ndarray) -> list[ClassificationResult]:
+        """Return content type classifications for an RGB image."""
+        ...
+
+
+class FaceProcessor(ABC):
+    """Detects faces and extracts recognition embeddings (e.g. YuNet + SFace)."""
+
+    @abstractmethod
+    def process(self, image: np.ndarray) -> list[FaceDetection]:
+        """Return face detections with embeddings for an RGB image."""
+        ...
+
+    @property
+    @abstractmethod
+    def embedding_dim(self) -> int:
+        """Dimensionality of face embedding vectors."""
+        ...

backend/app/services/vision/bootstrap_models.py

@@ -0,0 +1,85 @@
+"""
+Download vision model weights on first worker boot.
+
+Run as:  python -m app.services.vision.bootstrap_models
+
+Or called from the vision worker entrypoint before Celery starts.
+Downloads are idempotent — existing files are skipped.
+
+For models that require export (OpenCLIP, YOLOv8n), see export_models.py.
+Those must be exported once on any machine with pip, then placed in
+the models volume before the worker starts.
+"""
+import logging
+import os
+from pathlib import Path
+from urllib.request import urlretrieve
+
+from app.config import settings
+
+logger = logging.getLogger(__name__)
+
+# (relative_path, url, description)
+# Models with url=None must be pre-exported via export_models.py.
+# InsightFace (RetinaFace + ArcFace) auto-downloads via the insightface
+# package on first use — no manual download entries needed.
+DOWNLOADS = []
+
+# Models that need manual export via export_models.py
+EXPORTS = [
+    ("embed/visual.onnx", "OpenCLIP ViT-B/32 visual encoder"),
+    ("embed/textual.onnx", "OpenCLIP ViT-B/32 textual encoder"),
+    ("detect/yolov8n.onnx", "YOLOv8n object detector"),
+]
+
+
+def bootstrap(models_dir: str | None = None):
+    """Ensure all model files are present. Download what we can, warn about
+    files that need manual export."""
+    base = Path(models_dir or settings.vision.models_dir)
+    base.mkdir(parents=True, exist_ok=True)
+
+    # Download auto-downloadable models
+    for rel_path, url, desc in DOWNLOADS:
+        dest = base / rel_path
+        dest.parent.mkdir(parents=True, exist_ok=True)
+
+        if dest.exists():
+            logger.debug("Already exists: %s (%s)", dest, desc)
+            continue
+
+        logger.info("Downloading %s → %s", desc, dest)
+        try:
+            urlretrieve(url, str(dest))
+            size_kb = dest.stat().st_size / 1024
+            logger.info("Downloaded %s (%.0f KB)", desc, size_kb)
+        except Exception as e:
+            logger.error("Failed to download %s: %s", desc, e)
+            if dest.exists():
+                dest.unlink()
+
+    # Check for manually-exported models
+    missing = []
+    for rel_path, desc in EXPORTS:
+        dest = base / rel_path
+        if not dest.exists():
+            missing.append((rel_path, desc))
+
+    if missing:
+        logger.warning(
+            "Missing %d model(s) that require manual export via export_models.py:",
+            len(missing),
+        )
+        for rel_path, desc in missing:
+            logger.warning("  %s — %s", base / rel_path, desc)
+        logger.warning(
+            "Run: python -m app.services.vision.export_models --models-dir %s",
+            base,
+        )
+    else:
+        logger.info("All model files present in %s", base)
+
+
+if __name__ == "__main__":
+    logging.basicConfig(level=logging.INFO)
+    bootstrap()

backend/app/services/vision/classify.py

@@ -0,0 +1,106 @@
+"""
+CLIP zero-shot content-type classifier.
+
+Uses the native OpenCLIP PyTorch text encoder for high-quality text
+embeddings (the ONNX text encoder has degraded quality due to the
+eot_indices workaround). Image embeddings use the ONNX visual encoder
+which works well.
+"""
+import logging
+
+import numpy as np
+import torch
+
+from app.config import VisionSettings
+from app.services.vision.base import ContentClassifier, ClassificationResult
+
+logger = logging.getLogger(__name__)
+
+CATEGORY_PROMPTS = {
+    "screenshot": [
+        "a screenshot of a computer screen",
+        "a screenshot of a phone screen",
+        "a screen capture of a user interface",
+    ],
+    "document": [
+        "a scanned document",
+        "a photo of a document with printed text",
+        "a photo of a page of text on paper",
+    ],
+    "receipt": [
+        "a photo of a receipt",
+        "a photo of a bill or invoice",
+    ],
+    "meme": [
+        "an internet meme with text overlay",
+        "a funny image with caption text",
+    ],
+    "artwork": [
+        "a painting or drawing",
+        "a sketch or illustration",
+        "digital art or graphic design",
+    ],
+    "photograph": [
+        "a photograph taken with a camera",
+        "a real photo of a real scene or person",
+        "a candid photograph",
+    ],
+}
+
+
+class CLIPContentClassifier(ContentClassifier):
+    """Zero-shot content classifier using CLIP text-image similarity.
+    Uses native PyTorch for text encoding, ONNX for image encoding."""
+
+    def __init__(self, settings: VisionSettings):
+        import open_clip
+
+        self._min_confidence = settings.classifier.min_confidence
+
+        # Load native model for text encoding only
+        logger.info("Loading OpenCLIP text encoder for content classification")
+        model, _, _ = open_clip.create_model_and_transforms(
+            "ViT-B-32", pretrained="laion2b_s34b_b79k"
+        )
+        model.eval()
+        self._model = model
+        self._tokenizer = open_clip.get_tokenizer("ViT-B-32")
+
+        # Get the ONNX image embedder from the registry
+        from app.services.vision.registry import registry
+        self._embedder = registry.get_embedder()
+
+        # Pre-compute text embeddings for each category
+        self._category_embeddings: dict[str, np.ndarray] = {}
+        for category, prompts in CATEGORY_PROMPTS.items():
+            tokens = self._tokenizer(prompts)
+            with torch.no_grad():
+                text_features = model.encode_text(tokens)
+                text_features /= text_features.norm(dim=-1, keepdim=True)
+                avg = text_features.mean(dim=0)
+                avg /= avg.norm()
+            self._category_embeddings[category] = avg.numpy().astype(np.float32)
+
+        logger.info("Content classifier ready with %d categories", len(self._category_embeddings))
+
+    def classify(self, image: np.ndarray) -> list[ClassificationResult]:
+        img_vec = self._embedder.embed_image(image)
+
+        # Cosine similarity against each category
+        scores = {}
+        for category, cat_vec in self._category_embeddings.items():
+            scores[category] = float(np.dot(img_vec, cat_vec))
+
+        # Sort by score descending
+        ranked = sorted(scores.items(), key=lambda x: -x[1])
+        best_cat, best_score = ranked[0]
+        second_score = ranked[1][1]
+
+        margin = best_score - second_score
+        # Normalize: 0.01 margin → ~0.5 confidence, 0.03+ → ~1.0
+        confidence = min(1.0, margin * 30)
+
+        if confidence >= self._min_confidence:
+            return [ClassificationResult(label=best_cat, confidence=confidence)]
+
+        return []

backend/app/services/vision/clustering.py

@@ -0,0 +1,42 @@
+"""
+Face embedding clustering using DBSCAN with cosine distance.
+
+Called by the periodic `recluster_faces` Celery task (PR7).
+"""
+import logging
+
+import numpy as np
+from sklearn.cluster import DBSCAN
+
+logger = logging.getLogger(__name__)
+
+
+def cluster_faces(
+    embeddings: np.ndarray,
+    eps: float = 0.35,
+    min_samples: int = 2,
+) -> np.ndarray:
+    """Cluster face embeddings using DBSCAN with cosine metric.
+
+    Args:
+        embeddings: (N, D) float32 array of L2-normalized face embeddings.
+        eps: Maximum cosine distance between two samples to be in the
+             same neighborhood. Lower = tighter clusters.
+        min_samples: Minimum cluster size.
+
+    Returns:
+        (N,) int array of cluster labels. -1 = noise / unclustered.
+    """
+    if len(embeddings) < min_samples:
+        return np.full(len(embeddings), -1, dtype=int)
+
+    db = DBSCAN(eps=eps, min_samples=min_samples, metric="cosine")
+    labels = db.fit_predict(embeddings)
+
+    n_clusters = len(set(labels) - {-1})
+    n_noise = (labels == -1).sum()
+    logger.info(
+        "Face clustering: %d embeddings → %d clusters, %d noise",
+        len(embeddings), n_clusters, n_noise,
+    )
+    return labels

backend/app/services/vision/detect.py

@@ -0,0 +1,140 @@
+"""
+YOLOv8n object detector using raw ONNX Runtime.
+
+Expects {models_dir}/detect/yolov8n.onnx, exported from ultralytics
+via bootstrap_models.py. We do NOT ship ultralytics at runtime to
+avoid dragging in torch.
+"""
+import logging
+from pathlib import Path
+
+import numpy as np
+import onnxruntime as ort
+
+from app.config import VisionSettings
+from app.services.vision.base import ObjectDetector, DetectionBox
+
+logger = logging.getLogger(__name__)
+
+_INPUT_SIZE = 640
+
+# COCO class names (80 classes)
+COCO_LABELS = [
+    "person", "bicycle", "car", "motorcycle", "airplane", "bus", "train",
+    "truck", "boat", "traffic light", "fire hydrant", "stop sign",
+    "parking meter", "bench", "bird", "cat", "dog", "horse", "sheep",
+    "cow", "elephant", "bear", "zebra", "giraffe", "backpack", "umbrella",
+    "handbag", "tie", "suitcase", "frisbee", "skis", "snowboard",
+    "sports ball", "kite", "baseball bat", "baseball glove", "skateboard",
+    "surfboard", "tennis racket", "bottle", "wine glass", "cup", "fork",
+    "knife", "spoon", "bowl", "banana", "apple", "sandwich", "orange",
+    "broccoli", "carrot", "hot dog", "pizza", "donut", "cake", "chair",
+    "couch", "potted plant", "bed", "dining table", "toilet", "tv",
+    "laptop", "mouse", "remote", "keyboard", "cell phone", "microwave",
+    "oven", "toaster", "sink", "refrigerator", "book", "clock", "vase",
+    "scissors", "teddy bear", "hair drier", "toothbrush",
+]
+
+
+def _preprocess(image: np.ndarray) -> tuple[np.ndarray, float, float]:
+    """Letterbox-resize + normalize to NCHW float32. Returns input tensor
+    and scale factors for mapping boxes back to original coords."""
+    from PIL import Image
+
+    img = Image.fromarray(image).convert("RGB")
+    orig_w, orig_h = img.size
+
+    scale = min(_INPUT_SIZE / orig_w, _INPUT_SIZE / orig_h)
+    new_w = int(orig_w * scale)
+    new_h = int(orig_h * scale)
+    img = img.resize((new_w, new_h), Image.BICUBIC)
+
+    # Paste onto gray canvas
+    canvas = np.full((_INPUT_SIZE, _INPUT_SIZE, 3), 114, dtype=np.uint8)
+    pad_x = (_INPUT_SIZE - new_w) // 2
+    pad_y = (_INPUT_SIZE - new_h) // 2
+    canvas[pad_y : pad_y + new_h, pad_x : pad_x + new_w] = np.array(img)
+
+    blob = canvas.astype(np.float32) / 255.0
+    blob = blob.transpose(2, 0, 1)[np.newaxis]  # NCHW
+    return blob, scale, pad_x, pad_y
+
+
+def _postprocess(
+    outputs: np.ndarray,
+    scale: float,
+    pad_x: int,
+    pad_y: int,
+    orig_w: int,
+    orig_h: int,
+    conf_threshold: float,
+    max_detections: int,
+) -> list[DetectionBox]:
+    """Parse YOLOv8 output (1, 84, N) → list of DetectionBox."""
+    # outputs shape: (1, 84, N) where 84 = 4 box coords + 80 class scores
+    preds = outputs[0]  # (84, N)
+    preds = preds.T  # (N, 84)
+
+    boxes_xywh = preds[:, :4]
+    scores = preds[:, 4:]
+
+    class_ids = np.argmax(scores, axis=1)
+    confidences = scores[np.arange(len(scores)), class_ids]
+
+    mask = confidences >= conf_threshold
+    boxes_xywh = boxes_xywh[mask]
+    class_ids = class_ids[mask]
+    confidences = confidences[mask]
+
+    if len(confidences) == 0:
+        return []
+
+    # Sort by confidence, take top N
+    order = np.argsort(-confidences)[:max_detections]
+    boxes_xywh = boxes_xywh[order]
+    class_ids = class_ids[order]
+    confidences = confidences[order]
+
+    results = []
+    for i in range(len(confidences)):
+        cx, cy, w, h = boxes_xywh[i]
+        # Remove letterbox padding and rescale to original image
+        x1 = (cx - w / 2 - pad_x) / scale
+        y1 = (cy - h / 2 - pad_y) / scale
+        x2 = (cx + w / 2 - pad_x) / scale
+        y2 = (cy + h / 2 - pad_y) / scale
+        # Normalize to 0-1
+        bbox = [
+            max(0, x1 / orig_w),
+            max(0, y1 / orig_h),
+            min(1, x2 / orig_w),
+            min(1, y2 / orig_h),
+        ]
+        label = COCO_LABELS[class_ids[i]] if class_ids[i] < len(COCO_LABELS) else f"class_{class_ids[i]}"
+        results.append(DetectionBox(label=label, confidence=float(confidences[i]), bbox=bbox))
+
+    return results
+
+
+class YOLOv8Detector(ObjectDetector):
+    def __init__(self, settings: VisionSettings):
+        model_path = Path(settings.models_dir) / "detect" / "yolov8n.onnx"
+
+        opts = ort.SessionOptions()
+        opts.inter_op_num_threads = 2
+        opts.intra_op_num_threads = 2
+
+        logger.info("Loading YOLOv8n from %s", model_path)
+        self._session = ort.InferenceSession(str(model_path), opts, providers=["CPUExecutionProvider"])
+        self._conf_threshold = settings.detector.min_confidence
+        self._max_detections = settings.detector.max_detections
+
+    def detect(self, image: np.ndarray) -> list[DetectionBox]:
+        orig_h, orig_w = image.shape[:2]
+        blob, scale, pad_x, pad_y = _preprocess(image)
+        input_name = self._session.get_inputs()[0].name
+        outputs = self._session.run(None, {input_name: blob})[0]
+        return _postprocess(
+            outputs, scale, pad_x, pad_y, orig_w, orig_h,
+            self._conf_threshold, self._max_detections,
+        )

backend/app/services/vision/embed.py

@@ -0,0 +1,86 @@
+"""
+OpenCLIP ViT-B/32 embedder using ONNX Runtime.
+
+Expects two ONNX files under {models_dir}/embed/:
+  - visual.onnx   (image encoder)
+  - textual.onnx  (text encoder)
+
+These are exported from open_clip via bootstrap_models.py.
+"""
+import logging
+from pathlib import Path
+
+import numpy as np
+import onnxruntime as ort
+
+from app.config import VisionSettings
+from app.services.vision.base import Embedder
+
+logger = logging.getLogger(__name__)
+
+# OpenCLIP ViT-B/32 preprocessing constants (ImageNet norm)
+_MEAN = np.array([0.48145466, 0.4578275, 0.40821073], dtype=np.float32)
+_STD = np.array([0.26862954, 0.26130258, 0.27577711], dtype=np.float32)
+_INPUT_SIZE = 224
+
+
+def _preprocess_image(image: np.ndarray) -> np.ndarray:
+    """Resize, center-crop, normalize an RGB uint8 image to NCHW float32."""
+    from PIL import Image
+
+    img = Image.fromarray(image).convert("RGB")
+    # Resize shortest edge to _INPUT_SIZE, then center crop
+    w, h = img.size
+    scale = _INPUT_SIZE / min(w, h)
+    img = img.resize((int(w * scale), int(h * scale)), Image.BICUBIC)
+    w, h = img.size
+    left = (w - _INPUT_SIZE) // 2
+    top = (h - _INPUT_SIZE) // 2
+    img = img.crop((left, top, left + _INPUT_SIZE, top + _INPUT_SIZE))
+
+    arr = np.array(img, dtype=np.float32) / 255.0
+    arr = (arr - _MEAN) / _STD
+    arr = arr.transpose(2, 0, 1)  # HWC → CHW
+    return arr[np.newaxis]  # NCHW
+
+
+class OpenCLIPEmbedder(Embedder):
+    def __init__(self, settings: VisionSettings):
+        model_dir = Path(settings.models_dir) / "embed"
+        visual_path = model_dir / "visual.onnx"
+        textual_path = model_dir / "textual.onnx"
+
+        opts = ort.SessionOptions()
+        opts.inter_op_num_threads = 2
+        opts.intra_op_num_threads = 2
+
+        logger.info("Loading visual encoder from %s", visual_path)
+        self._visual = ort.InferenceSession(str(visual_path), opts, providers=["CPUExecutionProvider"])
+
+        logger.info("Loading textual encoder from %s", textual_path)
+        self._textual = ort.InferenceSession(str(textual_path), opts, providers=["CPUExecutionProvider"])
+
+    def embed_image(self, image: np.ndarray) -> np.ndarray:
+        inp = _preprocess_image(image)
+        input_name = self._visual.get_inputs()[0].name
+        out = self._visual.run(None, {input_name: inp})[0][0]
+        out = out / np.linalg.norm(out)
+        return out.astype(np.float32)
+
+    def embed_text(self, text: str) -> np.ndarray:
+        import open_clip
+        tokenizer = open_clip.get_tokenizer("ViT-B-32")
+        tokens = tokenizer([text]).numpy().astype(np.int64)
+        # Compute EOT indices outside ONNX (avoids ArgMax(13) op)
+        eot_indices = tokens.argmax(axis=-1).astype(np.int64)
+        inputs = self._textual.get_inputs()
+        out = self._textual.run(None, {
+            inputs[0].name: tokens,
+            inputs[1].name: eot_indices,
+        })[0][0]
+        out = out / np.linalg.norm(out)
+        return out.astype(np.float32)
+
+    @property
+    def dim(self) -> int:
+        return 512

backend/app/services/vision/export_models.py

@@ -0,0 +1,177 @@
+"""
+Export / download all vision model weights to ONNX format.
+
+Run ONCE on any machine with Python + pip (doesn't need GPU):
+
+    pip install open-clip-torch ultralytics onnx
+    python -m app.services.vision.export_models [--models-dir /data/models]
+
+This produces:
+    embed/visual.onnx        (~350 MB)
+    embed/textual.onnx       (~250 MB)
+    detect/yolov8n.onnx      (~12 MB)
+
+YuNet and SFace are downloaded by bootstrap_models.py at worker boot
+(Apache 2.0, lightweight, no export step needed).
+
+After export, copy the /data/models directory into your Docker volume:
+    docker cp /data/models mulita-worker:/data/models
+Or mount a host path in docker-compose.yml.
+"""
+import argparse
+import logging
+import sys
+from pathlib import Path
+
+logger = logging.getLogger(__name__)
+
+
+def export_openclip(models_dir: Path):
+    """Export OpenCLIP ViT-B/32 to two ONNX files (visual + textual)."""
+    import torch
+    import open_clip
+
+    out_dir = models_dir / "embed"
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    visual_path = out_dir / "visual.onnx"
+    textual_path = out_dir / "textual.onnx"
+
+    if visual_path.exists() and textual_path.exists():
+        logger.info("OpenCLIP ONNX files already exist, skipping export")
+        return
+
+    logger.info("Loading OpenCLIP ViT-B-32 laion2b_s34b_b79k...")
+    model, _, preprocess = open_clip.create_model_and_transforms(
+        "ViT-B-32", pretrained="laion2b_s34b_b79k"
+    )
+    model.eval()
+
+    # Use dynamo=False to get the legacy TorchScript exporter which
+    # produces IR version 9 (compatible with onnxruntime 1.17.x).
+    # The new torch.onnx.export default (dynamo=True) emits IR 10.
+    export_kwargs = dict(opset_version=14, dynamo=False)
+
+    # ── Visual encoder ────────────────────────────────────────────────
+    if not visual_path.exists():
+        logger.info("Exporting visual encoder → %s", visual_path)
+        dummy_image = torch.randn(1, 3, 224, 224)
+        torch.onnx.export(
+            model.visual,
+            dummy_image,
+            str(visual_path),
+            input_names=["image"],
+            output_names=["embedding"],
+            dynamic_axes={"image": {0: "batch"}},
+            **export_kwargs,
+        )
+        size_mb = visual_path.stat().st_size / 1e6
+        logger.info("Visual encoder exported (%.1f MB)", size_mb)
+
+    # ── Textual encoder ───────────────────────────────────────────────
+    if not textual_path.exists():
+        logger.info("Exporting textual encoder → %s", textual_path)
+        tokenizer = open_clip.get_tokenizer("ViT-B-32")
+        dummy_text = tokenizer(["a photo"]).to(torch.int64)
+
+        class TextEncoder(torch.nn.Module):
+            """Wrap the CLIP text encoder to avoid argmax in the ONNX graph.
+            OpenCLIP uses argmax to find the EOT token position, but ORT
+            ARM64 doesn't support ArgMax(13). We pre-compute the EOT index
+            from the token sequence and pass it directly."""
+            def __init__(self, clip_model):
+                super().__init__()
+                self.transformer = clip_model.transformer
+                self.token_embedding = clip_model.token_embedding
+                self.positional_embedding = clip_model.positional_embedding
+                self.ln_final = clip_model.ln_final
+                self.text_projection = clip_model.text_projection
+
+            def forward(self, text, eot_indices):
+                x = self.token_embedding(text)
+                x = x + self.positional_embedding
+                x = x.permute(1, 0, 2)  # NLD -> LND
+                x = self.transformer(x)
+                x = x.permute(1, 0, 2)  # LND -> NLD
+                x = self.ln_final(x)
+                # Take the feature at the EOT token. The EOT index is
+                # passed in as a separate input (computed outside ONNX)
+                # to avoid ArgMax(13) which ORT ARM64 doesn't support.
+                x = x[torch.arange(x.shape[0]), eot_indices]
+                x = x @ self.text_projection
+                return x
+
+        text_enc = TextEncoder(model)
+        text_enc.eval()
+
+        # Compute EOT indices from dummy tokens (argmax of token ids)
+        dummy_eot = dummy_text.argmax(dim=-1)
+
+        torch.onnx.export(
+            text_enc,
+            (dummy_text, dummy_eot),
+            str(textual_path),
+            input_names=["text", "eot_indices"],
+            output_names=["embedding"],
+            dynamic_axes={"text": {0: "batch"}, "eot_indices": {0: "batch"}},
+            **export_kwargs,
+        )
+        size_mb = textual_path.stat().st_size / 1e6
+        logger.info("Textual encoder exported (%.1f MB)", size_mb)
+
+
+def export_yolov8n(models_dir: Path):
+    """Export YOLOv8n to ONNX."""
+    out_dir = models_dir / "detect"
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    onnx_path = out_dir / "yolov8n.onnx"
+
+    if onnx_path.exists():
+        logger.info("YOLOv8n ONNX already exists, skipping export")
+        return
+
+    logger.info("Exporting YOLOv8n → %s", onnx_path)
+
+    from ultralytics import YOLO
+
+    model = YOLO("yolov8n.pt")
+    model.export(format="onnx", imgsz=640, simplify=True)
+
+    # ultralytics exports to cwd as yolov8n.onnx — move to target
+    exported = Path("yolov8n.onnx")
+    if exported.exists():
+        exported.rename(onnx_path)
+
+    size_mb = onnx_path.stat().st_size / 1e6
+    logger.info("YOLOv8n exported (%.1f MB)", size_mb)
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Export vision model weights to ONNX")
+    parser.add_argument(
+        "--models-dir",
+        type=Path,
+        default=Path("/data/models"),
+        help="Directory to write model files (default: /data/models)",
+    )
+    args = parser.parse_args()
+
+    logging.basicConfig(
+        level=logging.INFO,
+        format="%(asctime)s %(levelname)s %(message)s",
+    )
+
+    models_dir = args.models_dir
+    models_dir.mkdir(parents=True, exist_ok=True)
+
+    logger.info("Exporting models to %s", models_dir)
+
+    export_openclip(models_dir)
+    export_yolov8n(models_dir)
+
+    logger.info("Done. Run bootstrap_models.py next to download YuNet + SFace.")
+
+
+if __name__ == "__main__":
+    main()

backend/app/services/vision/faces.py

@@ -0,0 +1,148 @@
+"""
+Face detection (YuNet) + recognition (SFace) using OpenCV DNN.
+
+YuNet is loaded via cv2.FaceDetectorYN which handles the multi-scale
+anchor decoding and NMS internally. SFace recognition uses raw ONNX
+Runtime for the 128-d embedding.
+
+Both models are from opencv_zoo (Apache 2.0 license).
+Expects {models_dir}/face/:
+  - yunet.onnx     (~233 KB)
+  - sface.onnx     (~37 MB, 128-d embeddings)
+"""
+import logging
+from pathlib import Path
+
+import numpy as np
+import cv2
+import onnxruntime as ort
+
+from app.config import VisionSettings
+from app.services.vision.base import FaceProcessor, FaceDetection
+
+logger = logging.getLogger(__name__)
+
+
+def _align_face(image: np.ndarray, landmarks: np.ndarray) -> np.ndarray:
+    """Align and crop a 112x112 face patch using 5-point landmarks."""
+    left_eye = landmarks[0]
+    right_eye = landmarks[1]
+
+    dx = right_eye[0] - left_eye[0]
+    dy = right_eye[1] - left_eye[1]
+    angle = np.degrees(np.arctan2(dy, dx))
+    eye_center = ((left_eye[0] + right_eye[0]) / 2, (left_eye[1] + right_eye[1]) / 2)
+    eye_dist = np.sqrt(dx * dx + dy * dy)
+
+    M = cv2.getRotationMatrix2D(eye_center, angle, 1.0)
+    rotated = cv2.warpAffine(image, M, (image.shape[1], image.shape[0]))
+
+    # Crop around face center
+    scale = 64.0 / max(eye_dist, 1e-6)
+    cx, cy = eye_center
+    half = 56.0 / scale
+    x1 = max(0, int(cx - half))
+    y1 = max(0, int(cy - half * 0.8))
+    x2 = min(rotated.shape[1], int(cx + half))
+    y2 = min(rotated.shape[0], int(cy + half * 1.2))
+    crop = rotated[y1:y2, x1:x2]
+
+    if crop.size == 0:
+        return np.zeros((112, 112, 3), dtype=np.float32)
+
+    return cv2.resize(crop, (112, 112)).astype(np.float32)
+
+
+class YuNetSFaceProcessor(FaceProcessor):
+    def __init__(self, settings: VisionSettings):
+        face_dir = Path(settings.models_dir) / "face"
+        yunet_path = str(face_dir / "yunet.onnx")
+        sface_path = str(face_dir / "sface.onnx")
+
+        # YuNet via OpenCV's FaceDetectorYN — handles anchor decoding + NMS
+        self._detector = cv2.FaceDetectorYN.create(
+            yunet_path,
+            "",
+            (640, 640),
+            settings.faces.recognition_threshold,
+            0.3,  # NMS threshold
+            5000,  # top_k
+        )
+        logger.info("YuNet face detector loaded via OpenCV")
+
+        # SFace via ONNX Runtime
+        opts = ort.SessionOptions()
+        opts.inter_op_num_threads = 2
+        opts.intra_op_num_threads = 2
+        ort.set_default_logger_severity(3)
+        self._recognizer = ort.InferenceSession(sface_path, opts, providers=["CPUExecutionProvider"])
+        logger.info("SFace recognizer loaded via ONNX Runtime")
+
+        self._min_face_size = settings.faces.min_face_size
+
+    def process(self, image: np.ndarray) -> list[FaceDetection]:
+        orig_h, orig_w = image.shape[:2]
+
+        # Convert RGB → BGR for OpenCV
+        bgr = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
+
+        # Set input size to actual image dimensions
+        self._detector.setInputSize((orig_w, orig_h))
+
+        # Detect faces
+        _, faces_raw = self._detector.detect(bgr)
+
+        if faces_raw is None or len(faces_raw) == 0:
+            return []
+
+        results = []
+        for face in faces_raw:
+            # face: [x, y, w, h, right_eye_x, right_eye_y, left_eye_x, left_eye_y,
+            #        nose_x, nose_y, right_mouth_x, right_mouth_y, left_mouth_x, left_mouth_y, score]
+            x, y, w, h = int(face[0]), int(face[1]), int(face[2]), int(face[3])
+            score = float(face[14])
+
+            # Filter small faces
+            face_size = max(w, h)
+            if face_size < self._min_face_size:
+                continue
+
+            # Normalized bbox
+            bbox = [
+                max(0, x / orig_w),
+                max(0, y / orig_h),
+                min(1, (x + w) / orig_w),
+                min(1, (y + h) / orig_h),
+            ]
+
+            # Extract 5-point landmarks for alignment
+            landmarks = np.array([
+                [face[4], face[5]],   # right eye
+                [face[6], face[7]],   # left eye
+                [face[8], face[9]],   # nose
+                [face[10], face[11]], # right mouth
+                [face[12], face[13]], # left mouth
+            ], dtype=np.float32)
+
+            # Align face for recognition
+            face_crop = _align_face(image, landmarks)
+
+            # SFace expects (1, 3, 112, 112) float32, BGR
+            face_bgr = cv2.cvtColor(face_crop.astype(np.uint8), cv2.COLOR_RGB2BGR)
+            face_blob = (face_bgr.astype(np.float32) / 255.0).transpose(2, 0, 1)[np.newaxis]
+
+            rec_input = self._recognizer.get_inputs()[0].name
+            embedding = self._recognizer.run(None, {rec_input: face_blob})[0][0]
+            embedding = embedding / np.linalg.norm(embedding)
+
+            results.append(FaceDetection(
+                bbox=bbox,
+                embedding=embedding.astype(np.float32),
+                quality=score,
+            ))
+
+        return results
+
+    @property
+    def embedding_dim(self) -> int:
+        return 128

backend/app/services/vision/insightface_processor.py

@@ -0,0 +1,70 @@
+"""
+Face detection + recognition using InsightFace (RetinaFace + ArcFace).
+
+Uses the buffalo_l model pack which auto-downloads on first use (~300MB).
+Produces 512-d ArcFace embeddings. Non-commercial research license —
+fine for homelab self-hosting.
+"""
+import logging
+from pathlib import Path
+
+import numpy as np
+
+from app.config import VisionSettings
+from app.services.vision.base import FaceProcessor, FaceDetection
+
+logger = logging.getLogger(__name__)
+
+
+class InsightFaceProcessor(FaceProcessor):
+    def __init__(self, settings: VisionSettings):
+        from insightface.app import FaceAnalysis
+
+        model_root = str(Path(settings.models_dir) / "face" / "insightface")
+        logger.info("Loading InsightFace buffalo_l from %s", model_root)
+
+        self._app = FaceAnalysis(
+            name="buffalo_l",
+            root=model_root,
+            providers=["CPUExecutionProvider"],
+        )
+        self._app.prepare(ctx_id=-1, det_size=(640, 640))
+        self._min_det_score = settings.faces.recognition_threshold
+
+    def process(self, image: np.ndarray) -> list[FaceDetection]:
+        orig_h, orig_w = image.shape[:2]
+
+        # InsightFace expects BGR
+        bgr = image[:, :, ::-1].copy()
+
+        faces = self._app.get(bgr)
+
+        if not faces:
+            return []
+
+        results = []
+        for face in faces:
+            if face.det_score < self._min_det_score:
+                continue
+
+            # face.bbox is [x1, y1, x2, y2] in pixel coords
+            x1, y1, x2, y2 = face.bbox
+            bbox = [
+                max(0, float(x1) / orig_w),
+                max(0, float(y1) / orig_h),
+                min(1, float(x2) / orig_w),
+                min(1, float(y2) / orig_h),
+            ]
+
+            embedding = face.normed_embedding  # already L2-normalized, 512-d
+            results.append(FaceDetection(
+                bbox=bbox,
+                embedding=embedding.astype(np.float32),
+                quality=float(face.det_score),
+            ))
+
+        return results
+
+    @property
+    def embedding_dim(self) -> int:
+        return 512

backend/app/services/vision/ocr.py

@@ -0,0 +1,45 @@
+"""
+OCR engine using rapidocr-onnxruntime (PP-OCRv4 weights).
+
+No PaddlePaddle dependency — pure ONNX Runtime. Language packs are
+downloaded automatically by rapidocr on first use.
+"""
+import logging
+
+import numpy as np
+
+from app.config import VisionSettings
+from app.services.vision.base import OCREngine, OCRResult
+
+logger = logging.getLogger(__name__)
+
+
+class RapidOCREngine(OCREngine):
+    def __init__(self, settings: VisionSettings):
+        from rapidocr_onnxruntime import RapidOCR
+
+        self._min_confidence = settings.ocr.min_confidence
+        self._engine = RapidOCR()
+        logger.info("RapidOCR engine initialized")
+
+    def run(self, image: np.ndarray) -> list[OCRResult]:
+        result, _ = self._engine(image)
+        if not result:
+            return []
+
+        out = []
+        for box, text, score in result:
+            if score < self._min_confidence:
+                continue
+            # box is [[x1,y1],[x2,y2],[x3,y3],[x4,y4]] — take bounding rect
+            xs = [p[0] for p in box]
+            ys = [p[1] for p in box]
+            h, w = image.shape[:2]
+            bbox = [
+                min(xs) / w,
+                min(ys) / h,
+                max(xs) / w,
+                max(ys) / h,
+            ]
+            out.append(OCRResult(text=text, confidence=float(score), bbox=bbox))
+        return out

backend/app/services/vision/onnx_backend.py

@@ -0,0 +1,41 @@
+"""
+ONNX Runtime backend — default CPU inference for all vision models.
+
+Each create_* method returns a concrete implementation of the
+corresponding ABC from base.py. Models are loaded from ONNX files
+under settings.vision.models_dir, downloaded on first boot by
+bootstrap_models.py.
+"""
+import logging
+
+from app.config import VisionSettings
+from app.services.vision.base import Embedder, OCREngine, ObjectDetector, FaceProcessor, ContentClassifier
+
+logger = logging.getLogger(__name__)
+
+
+class ONNXBackend:
+    """Factory for ONNX Runtime-based vision model instances."""
+
+    def __init__(self, vision_settings: VisionSettings):
+        self._settings = vision_settings
+
+    def create_embedder(self) -> Embedder:
+        from app.services.vision.embed import OpenCLIPEmbedder
+        return OpenCLIPEmbedder(self._settings)
+
+    def create_ocr(self) -> OCREngine:
+        from app.services.vision.ocr import RapidOCREngine
+        return RapidOCREngine(self._settings)
+
+    def create_detector(self) -> ObjectDetector:
+        from app.services.vision.detect import YOLOv8Detector
+        return YOLOv8Detector(self._settings)
+
+    def create_face_processor(self) -> FaceProcessor:
+        from app.services.vision.insightface_processor import InsightFaceProcessor
+        return InsightFaceProcessor(self._settings)
+
+    def create_classifier(self) -> ContentClassifier:
+        from app.services.vision.classify import CLIPContentClassifier
+        return CLIPContentClassifier(self._settings)

backend/app/services/vision/registry.py

@@ -0,0 +1,84 @@
+"""
+ModelRegistry — singleton that lazy-loads vision models per worker process.
+
+Usage from Celery tasks:
+
+    from app.services.vision.registry import registry
+    embedder = registry.get_embedder()
+    vec = embedder.embed_image(img)
+
+Models are created on first access and cached for the worker's lifetime.
+The registry reads settings.vision to decide which backend to use and
+where model weights live.
+"""
+import logging
+from functools import lru_cache
+
+from app.config import settings
+from app.services.vision.base import Embedder, OCREngine, ObjectDetector, FaceProcessor, ContentClassifier
+
+logger = logging.getLogger(__name__)
+
+
+class ModelRegistry:
+    """Central access point for all vision models."""
+
+    def __init__(self):
+        self._vision = settings.vision
+
+    @lru_cache(maxsize=1)
+    def get_embedder(self) -> Embedder:
+        logger.info("Loading embedder: %s (backend=%s)", self._vision.embedder.name, self._vision.backend)
+        return self._load_backend().create_embedder()
+
+    @lru_cache(maxsize=1)
+    def get_ocr(self) -> OCREngine:
+        logger.info("Loading OCR engine (backend=%s)", self._vision.backend)
+        return self._load_backend().create_ocr()
+
+    @lru_cache(maxsize=1)
+    def get_detector(self) -> ObjectDetector:
+        logger.info("Loading object detector (backend=%s)", self._vision.backend)
+        return self._load_backend().create_detector()
+
+    @lru_cache(maxsize=1)
+    def get_face_processor(self) -> FaceProcessor:
+        logger.info("Loading face processor (backend=%s)", self._vision.backend)
+        return self._load_backend().create_face_processor()
+
+    @lru_cache(maxsize=1)
+    def get_classifier(self) -> ContentClassifier:
+        logger.info("Loading content classifier (backend=%s)", self._vision.backend)
+        return self._load_backend().create_classifier()
+
+    @lru_cache(maxsize=1)
+    def _load_backend(self):
+        """Import and instantiate the configured backend."""
+        backend_name = self._vision.backend
+        if backend_name == "onnx":
+            from app.services.vision.onnx_backend import ONNXBackend
+            return ONNXBackend(self._vision)
+        elif backend_name == "rocm":
+            from app.services.vision.rocm_backend import ROCmBackend
+            return ROCmBackend(self._vision)
+        else:
+            raise ValueError(f"Unknown vision backend: {backend_name}")
+
+    def warmup(self):
+        """Pre-load all enabled models. Called from Celery worker_process_init
+        on the vision queue to avoid cold-start latency on the first task."""
+        logger.info("Warming up vision models...")
+        self.get_embedder()
+        if self._vision.ocr.enabled:
+            self.get_ocr()
+        if self._vision.detector.enabled:
+            self.get_detector()
+        if self._vision.faces.enabled:
+            self.get_face_processor()
+        if self._vision.classifier.enabled:
+            self.get_classifier()
+        logger.info("Vision model warmup complete")
+
+
+# Module-level singleton. Import this from tasks.
+registry = ModelRegistry()

backend/app/services/vision/rocm_backend.py

@@ -0,0 +1,25 @@
+"""
+ROCm backend — GPU-accelerated inference for Radeon 760M-class hardware.
+
+Stub: raises NotImplementedError on all factory methods. To enable,
+set `vision.backend: rocm` in mulita.yml once ROCm support is implemented.
+"""
+from app.config import VisionSettings
+from app.services.vision.base import Embedder, OCREngine, ObjectDetector, FaceProcessor
+
+
+class ROCmBackend:
+    def __init__(self, vision_settings: VisionSettings):
+        self._settings = vision_settings
+
+    def create_embedder(self) -> Embedder:
+        raise NotImplementedError("ROCm backend not yet implemented — use 'onnx'")
+
+    def create_ocr(self) -> OCREngine:
+        raise NotImplementedError("ROCm backend not yet implemented — use 'onnx'")
+
+    def create_detector(self) -> ObjectDetector:
+        raise NotImplementedError("ROCm backend not yet implemented — use 'onnx'")
+
+    def create_face_processor(self) -> FaceProcessor:
+        raise NotImplementedError("ROCm backend not yet implemented — use 'onnx'")

backend/app/tasks/celery.py

@@ -9,7 +9,7 @@ celery_app = Celery(
     'mulita',
     broker=settings.celery_broker_url,
     backend=settings.celery_result_backend,
-    include=['app.tasks.scan', 'app.tasks.thumbs']
+    include=['app.tasks.scan', 'app.tasks.thumbs', 'app.tasks.vision']
 )
 
 # Configure Celery
@@ -22,6 +22,13 @@ celery_app.conf.update(
     task_routes={
         'app.tasks.thumbs.*': {'queue': 'high'},
         'app.tasks.scan.*': {'queue': 'low'},
+        'app.tasks.vision.*': {'queue': 'vision'},
+        'embed_photo': {'queue': 'vision'},
+        'ocr_photo': {'queue': 'vision'},
+        'detect_objects': {'queue': 'vision'},
+        'extract_faces': {'queue': 'vision'},
+        'classify_content': {'queue': 'vision'},
+        'vision_fanout': {'queue': 'vision'},
     },
     task_default_queue='default',
     task_default_exchange='default',

backend/app/tasks/thumbs.py

@@ -317,6 +317,15 @@ async def _generate_thumbnails_async(photo_id: str, task):
             await session.commit()
             
             logger.info(f"Thumbnails generated for photo {photo_id}")
+
+            # Dispatch vision pipeline (embedding, OCR, detection, faces)
+            # after thumbs are ready so vision tasks have images to read.
+            try:
+                from app.tasks.vision import vision_fanout
+                vision_fanout.delay(photo_id)
+            except Exception as e:
+                logger.warning(f"Could not dispatch vision_fanout for {photo_id}: {e}")
+
             return {'status': 'success', 'photo_id': photo_id}
             
         except Exception as e:

backend/app/tasks/vision.py

@@ -0,0 +1,479 @@
+"""
+Celery tasks for the vision pipeline — embedding, OCR, object detection,
+face recognition.
+
+All tasks run on the dedicated `vision` queue with limited concurrency
+(memory-bound CPU inference). They read thumbnails generated by
+generate_thumbnails, so they MUST run after thumbs complete.
+
+DB access uses sync psycopg2 sessions (not asyncpg) because Celery
+forks workers and asyncpg connections can't be shared across forks.
+"""
+import logging
+from pathlib import Path
+
+import numpy as np
+from celery import shared_task
+from sqlalchemy import create_engine, text as sa_text, select, delete
+from sqlalchemy.orm import Session, sessionmaker
+from PIL import Image
+
+from app.models.embeddings import Embedding
+from app.config import settings
+
+logger = logging.getLogger(__name__)
+
+
+def _get_sync_session() -> Session:
+    """Create a sync DB session for use in Celery workers."""
+    sync_url = settings.database_url.replace("+asyncpg", "+psycopg2").replace("+aiosqlite", "")
+    engine = create_engine(sync_url, pool_pre_ping=True)
+    return sessionmaker(bind=engine)()
+
+
+def _load_thumb(photo_id: str, size: str = "medium") -> np.ndarray | None:
+    """Load a thumbnail as an RGB numpy array."""
+    thumb_path = Path(f"/data/thumbs/{photo_id}/{size}.webp")
+    if not thumb_path.exists():
+        logger.warning("Thumbnail not found: %s", thumb_path)
+        return None
+    img = Image.open(thumb_path).convert("RGB")
+    return np.array(img)
+
+
+@shared_task(name='embed_photo', queue='vision')
+def embed_photo(photo_id: str):
+    """Generate CLIP embedding for a photo and store in pgvector."""
+    if not settings.vision.enabled:
+        return {'status': 'skipped', 'reason': 'vision disabled'}
+
+    image = _load_thumb(photo_id, "medium")  # 640px
+    if image is None:
+        return {'status': 'error', 'message': 'thumbnail not found'}
+
+    from app.services.vision.registry import registry
+    embedder = registry.get_embedder()
+    vector = embedder.embed_image(image)
+
+    model_name = settings.vision.embedder.name
+
+    session = _get_sync_session()
+    try:
+        session.execute(
+            delete(Embedding).where(
+                Embedding.photo_id == photo_id,
+                Embedding.model == model_name,
+            )
+        )
+        emb = Embedding(
+            photo_id=photo_id,
+            model=model_name,
+            vector=vector.tolist(),
+        )
+        session.add(emb)
+        session.commit()
+    finally:
+        session.close()
+
+    logger.info("Embedded photo %s with %s", photo_id, model_name)
+    return {'status': 'success', 'photo_id': photo_id}
+
+
+@shared_task(name='vision_fanout', queue='vision')
+def vision_fanout(photo_id: str):
+    """Dispatch all enabled vision tasks for a photo."""
+    if not settings.vision.enabled:
+        return {'status': 'skipped', 'reason': 'vision disabled'}
+
+    embed_photo.delay(photo_id)
+
+    if settings.vision.ocr.enabled:
+        ocr_photo.delay(photo_id)
+    if settings.vision.detector.enabled:
+        detect_objects.delay(photo_id)
+    if settings.vision.faces.enabled:
+        extract_faces.delay(photo_id)
+    if settings.vision.classifier.enabled:
+        classify_content.delay(photo_id)
+
+    return {'status': 'dispatched', 'photo_id': photo_id}
+
+
+@shared_task(name='ocr_photo', queue='vision')
+def ocr_photo(photo_id: str):
+    """Run OCR on a photo and store text regions."""
+    if not settings.vision.enabled or not settings.vision.ocr.enabled:
+        return {'status': 'skipped', 'reason': 'OCR disabled'}
+
+    image = _load_thumb(photo_id, "large")  # 1280px for better OCR accuracy
+    if image is None:
+        return {'status': 'error', 'message': 'thumbnail not found'}
+
+    from app.services.vision.registry import registry
+    ocr_engine = registry.get_ocr()
+    results = ocr_engine.run(image)
+
+    if not results:
+        logger.info("No OCR text found for photo %s", photo_id)
+        return {'status': 'success', 'photo_id': photo_id, 'regions': 0}
+
+    from app.models.ocr_text import OCRText
+
+    session = _get_sync_session()
+    try:
+        session.execute(delete(OCRText).where(OCRText.photo_id == photo_id))
+        for r in results:
+            session.add(OCRText(
+                photo_id=photo_id,
+                text=r.text,
+                language=r.language,
+                confidence=r.confidence,
+                bbox=r.bbox,
+            ))
+        session.commit()
+    finally:
+        session.close()
+
+    logger.info("OCR: %d text regions for photo %s", len(results), photo_id)
+    return {'status': 'success', 'photo_id': photo_id, 'regions': len(results)}
+
+
+@shared_task(name='detect_objects', queue='vision')
+def detect_objects(photo_id: str):
+    """Detect objects in a photo, create Tag(kind=object) rows, and
+    link via photo_tags with confidence/bbox/source."""
+    if not settings.vision.enabled or not settings.vision.detector.enabled:
+        return {'status': 'skipped', 'reason': 'detection disabled'}
+
+    image = _load_thumb(photo_id, "medium")  # 640px
+    if image is None:
+        return {'status': 'error', 'message': 'thumbnail not found'}
+
+    from app.services.vision.registry import registry
+    detector = registry.get_detector()
+    detections = detector.detect(image)
+
+    if not detections:
+        logger.info("No objects detected for photo %s", photo_id)
+        return {'status': 'success', 'photo_id': photo_id, 'objects': 0}
+
+    from app.models.tags import Tag, photo_tags
+
+    source_name = "vision:yolov8n"
+
+    session = _get_sync_session()
+    try:
+        # Wipe previous detection results for this photo from this model
+        session.execute(
+            delete(photo_tags).where(
+                photo_tags.c.photo_id == photo_id,
+                photo_tags.c.source == source_name,
+            )
+        )
+
+        # Group detections by label, keep highest confidence per label
+        best_per_label: dict[str, tuple[float, list]] = {}
+        for det in detections:
+            if det.label not in best_per_label or det.confidence > best_per_label[det.label][0]:
+                best_per_label[det.label] = (det.confidence, det.bbox)
+
+        for label, (confidence, bbox) in best_per_label.items():
+            # Find or create the object tag
+            tag = session.execute(
+                select(Tag).where(Tag.name == label, Tag.kind == 'object')
+            ).scalar_one_or_none()
+
+            if not tag:
+                tag = Tag(name=label, kind='object', source=source_name)
+                session.add(tag)
+                session.flush()  # get tag.id
+
+            # Insert photo_tags association with ML metadata
+            session.execute(
+                photo_tags.insert().values(
+                    photo_id=photo_id,
+                    tag_id=tag.id,
+                    confidence=confidence,
+                    bbox=bbox,
+                    source=source_name,
+                )
+            )
+
+        session.commit()
+    finally:
+        session.close()
+
+    labels = [d.label for d in detections]
+    logger.info("Detected %d objects in photo %s: %s", len(detections), photo_id, labels)
+    return {'status': 'success', 'photo_id': photo_id, 'objects': len(detections)}
+
+
+@shared_task(name='classify_content', queue='vision')
+def classify_content(photo_id: str):
+    """Classify image content type (screenshot, document, artwork, etc.)
+    using CLIP zero-shot classification. Writes Tag(kind=content_type)."""
+    if not settings.vision.enabled or not settings.vision.classifier.enabled:
+        return {'status': 'skipped', 'reason': 'classifier disabled'}
+
+    image = _load_thumb(photo_id, "medium")
+    if image is None:
+        return {'status': 'error', 'message': 'thumbnail not found'}
+
+    from app.services.vision.registry import registry
+    classifier = registry.get_classifier()
+    results = classifier.classify(image)
+
+    if not results:
+        logger.info("No confident classification for photo %s", photo_id)
+        return {'status': 'success', 'photo_id': photo_id, 'content_type': None}
+
+    from app.models.tags import Tag, photo_tags
+
+    source_name = "vision:clip_classifier"
+    best = results[0]
+
+    session = _get_sync_session()
+    try:
+        # Wipe previous classification for this photo
+        session.execute(
+            delete(photo_tags).where(
+                photo_tags.c.photo_id == photo_id,
+                photo_tags.c.source == source_name,
+            )
+        )
+
+        # Find or create content_type tag
+        tag = session.execute(
+            select(Tag).where(Tag.name == best.label, Tag.kind == 'content_type')
+        ).scalar_one_or_none()
+
+        if not tag:
+            tag = Tag(name=best.label, kind='content_type', source=source_name)
+            session.add(tag)
+            session.flush()
+
+        session.execute(
+            photo_tags.insert().values(
+                photo_id=photo_id,
+                tag_id=tag.id,
+                confidence=best.confidence,
+                source=source_name,
+            )
+        )
+        session.commit()
+    finally:
+        session.close()
+
+    logger.info("Classified photo %s as '%s' (%.2f)", photo_id, best.label, best.confidence)
+    return {'status': 'success', 'photo_id': photo_id, 'content_type': best.label}
+
+
+def _load_original(photo_id: str) -> np.ndarray | None:
+    """Load the original photo file as an RGB numpy array, resized to
+    max 1280px on the longest edge for face detection."""
+    from sqlalchemy import create_engine, select as sa_select, text as sa_text
+    from app.models import Photo
+
+    session = _get_sync_session()
+    try:
+        photo = session.execute(
+            sa_select(Photo).where(Photo.id == photo_id)
+        ).scalar_one_or_none()
+        if not photo or not photo.filepath:
+            return None
+        filepath = photo.filepath
+    finally:
+        session.close()
+
+    if not Path(filepath).exists():
+        logger.warning("Original file not found: %s", filepath)
+        return None
+
+    try:
+        img = Image.open(filepath).convert("RGB")
+        # Cap at 4000px on longest edge to avoid OOM, but keep as large
+        # as possible for face detection accuracy
+        max_dim = 4000
+        w, h = img.size
+        if max(w, h) > max_dim:
+            scale = max_dim / max(w, h)
+            img = img.resize((int(w * scale), int(h * scale)), Image.BICUBIC)
+        return np.array(img)
+    except Exception as e:
+        logger.warning("Failed to load original %s: %s", filepath, e)
+        return None
+
+
+@shared_task(name='extract_faces', queue='vision')
+def extract_faces(photo_id: str):
+    """Detect faces and store recognition embeddings using InsightFace
+    (RetinaFace + ArcFace). No YOLO workaround needed — RetinaFace has
+    strong human-vs-non-human precision on its own."""
+    if not settings.vision.enabled or not settings.vision.faces.enabled:
+        return {'status': 'skipped', 'reason': 'faces disabled'}
+
+    image = _load_original(photo_id)
+    if image is None:
+        image = _load_thumb(photo_id, "large")
+    if image is None:
+        return {'status': 'error', 'message': 'no image available'}
+
+    from app.services.vision.registry import registry
+    face_proc = registry.get_face_processor()
+    faces = face_proc.process(image)
+
+    if not faces:
+        logger.info("No faces detected for photo %s", photo_id)
+
+    return _save_faces(photo_id, faces)
+
+
+def _save_faces(photo_id: str, faces) -> dict:
+    from app.models.face_embedding import FaceEmbedding
+
+    session = _get_sync_session()
+    try:
+        session.execute(delete(FaceEmbedding).where(FaceEmbedding.photo_id == photo_id))
+        for face in faces:
+            session.add(FaceEmbedding(
+                photo_id=photo_id,
+                bbox=face.bbox,
+                vector=face.embedding.tolist(),
+                quality=face.quality,
+                cluster_id=None,
+            ))
+        session.commit()
+    finally:
+        session.close()
+
+    if faces:
+        logger.info("Extracted %d verified face(s) from photo %s", len(faces), photo_id)
+    return {'status': 'success', 'photo_id': photo_id, 'faces': len(faces)}
+
+
+@shared_task(name='recluster_faces', queue='vision')
+def recluster_faces():
+    """Run DBSCAN clustering over all face embeddings and assign/create
+    Tag(kind=face_cluster) entries."""
+    if not settings.vision.enabled or not settings.vision.faces.enabled:
+        return {'status': 'skipped', 'reason': 'faces disabled'}
+
+    from app.models.face_embedding import FaceEmbedding
+    from app.models.tags import Tag, photo_tags
+    from app.services.vision.clustering import cluster_faces
+
+    source_name = "vision:sface"
+
+    session = _get_sync_session()
+    try:
+        face_rows = session.execute(
+            select(FaceEmbedding).order_by(FaceEmbedding.created_at)
+        ).scalars().all()
+
+        if len(face_rows) < 2:
+            logger.info("Not enough faces for clustering (%d)", len(face_rows))
+            return {'status': 'success', 'clusters': 0}
+
+        embeddings = np.array([f.vector for f in face_rows], dtype=np.float32)
+        labels = cluster_faces(embeddings, eps=settings.vision.faces.cluster_eps)
+
+        # Clean up old face_cluster tags and their photo_tags
+        old_cluster_tags = session.execute(
+            select(Tag).where(Tag.kind == 'face_cluster', Tag.source == source_name)
+        ).scalars().all()
+        for old_tag in old_cluster_tags:
+            session.execute(
+                delete(photo_tags).where(
+                    photo_tags.c.tag_id == old_tag.id,
+                    photo_tags.c.source == source_name,
+                )
+            )
+            session.delete(old_tag)
+        session.flush()
+
+        # Build new clusters
+        cluster_tag_map: dict[int, str] = {}
+        # Track which photos belong to which cluster
+        cluster_photos: dict[int, set[str]] = {}
+
+        for i, label in enumerate(labels):
+            if label == -1:
+                face_rows[i].cluster_id = None
+                continue
+
+            if label not in cluster_photos:
+                cluster_photos[label] = set()
+            cluster_photos[label].add(face_rows[i].photo_id)
+
+            if label not in cluster_tag_map:
+                cluster_name = f"Person {label + 1}"
+                tag = Tag(
+                    name=cluster_name,
+                    kind='face_cluster',
+                    source=source_name,
+                    representative_photo_id=face_rows[i].photo_id,
+                )
+                session.add(tag)
+                session.flush()
+                cluster_tag_map[label] = tag.id
+
+            face_rows[i].cluster_id = cluster_tag_map[label]
+
+        # Write photo_tags associations so the tag count and tag_ids
+        # filter work for face clusters
+        for label, photo_ids in cluster_photos.items():
+            tag_id = cluster_tag_map[label]
+            for pid in photo_ids:
+                session.execute(
+                    photo_tags.insert().values(
+                        photo_id=pid,
+                        tag_id=tag_id,
+                        source=source_name,
+                    )
+                )
+
+        session.commit()
+    finally:
+        session.close()
+
+    n_clusters = len(cluster_tag_map)
+    logger.info("Face clustering: %d clusters from %d faces", n_clusters, len(face_rows))
+    return {'status': 'success', 'clusters': n_clusters, 'faces': len(face_rows)}
+
+
+@shared_task(name='backfill_vision')
+def backfill_vision(task: str | None = None, limit: int | None = None):
+    """Queue vision tasks for photos that haven't been processed yet.
+    Uses a sync DB connection to avoid asyncpg conflicts in Celery."""
+    model_name = settings.vision.embedder.name
+    sql = """
+        SELECT p.id FROM photos p
+        LEFT JOIN embeddings e ON e.photo_id = p.id AND e.model = :model
+        WHERE e.photo_id IS NULL
+        AND p.processing_status = 'completed'
+        ORDER BY p.added_at DESC
+    """
+    if limit:
+        sql += f" LIMIT {limit}"
+
+    session = _get_sync_session()
+    try:
+        result = session.execute(sa_text(sql), {"model": model_name})
+        photo_ids = [row[0] for row in result.fetchall()]
+    finally:
+        session.close()
+
+    count = 0
+    for pid in photo_ids:
+        if task == 'embed' or task is None:
+            embed_photo.delay(pid)
+        if task == 'ocr' or task is None:
+            ocr_photo.delay(pid)
+        if task == 'detect' or task is None:
+            detect_objects.delay(pid)
+        if task == 'faces' or task is None:
+            extract_faces.delay(pid)
+        count += 1
+
+    logger.info("Backfill queued %d photos for vision processing", count)
+    return {'status': 'queued', 'count': count}

backend/requirements.txt

@@ -5,7 +5,10 @@ python-multipart==0.0.6
 
 # Database
 sqlalchemy[asyncio]==2.0.25
-aiosqlite==0.19.0
+aiosqlite==0.19.0           # SQLite escape hatch (docker-compose.sqlite.yml override)
+asyncpg==0.29.0             # async Postgres driver (default)
+psycopg2-binary==2.9.9      # sync Postgres driver, used by Alembic CLI
+pgvector==0.2.5             # pgvector SQLAlchemy types
 alembic==1.13.1
 
 # Redis and Celery
@@ -31,6 +34,14 @@ pyexiftool==0.5.6
 # File watching
 watchfiles==0.21.0
 
+# Vision pipeline (ONNX Runtime CPU inference)
+onnxruntime==1.18.1
+open-clip-torch==2.24.0     # tokenizer + export helper; inference via ONNX
+rapidocr-onnxruntime==1.3.22
+scikit-learn==1.4.0          # DBSCAN for face clustering
+insightface>=0.7.3           # RetinaFace + ArcFace face detection/recognition
+numpy>=1.26.0,<2.0
+
 # Utilities
 pyyaml==6.0.1
 pydantic==2.5.3

docker-compose.sqlite.yml

@@ -0,0 +1,47 @@
+# SQLite escape hatch override.
+#
+# Usage (omit the `db` service from the up command):
+#
+#   docker compose -f docker-compose.yml -f docker-compose.sqlite.yml \
+#       up frontend backend worker redis
+#
+# This pins the backend and worker to the legacy SQLite database file at
+# /data/db/mulita.db (in the existing db_data volume), drops the dependency
+# on Postgres, and skips Alembic — the SQLite schema is still managed by
+# the inline ALTERs in app/database.py:init_db.
+#
+# Vision features that depend on pgvector (PR4 onward) will refuse to enable
+# in this mode; the search/embedding endpoints will return 503 with a clear
+# error pointing back at the default Postgres setup.
+
+services:
+  backend:
+    command: sh -c "uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload"
+    environment:
+      - DATABASE_URL=sqlite+aiosqlite:////data/db/mulita.db
+      - REDIS_URL=redis://redis:6379
+      - CELERY_BROKER_URL=redis://redis:6379
+      - CELERY_RESULT_BACKEND=redis://redis:6379
+      - PHOTO_DIRS=${PHOTO_DIRS:-/photos}
+      - ALLOWED_ORIGINS=${ALLOWED_ORIGINS:-*}
+      - LOG_LEVEL=${LOG_LEVEL:-INFO}
+      - TZ=${TZ:-UTC}
+    depends_on:
+      redis:
+        condition: service_started
+
+  worker:
+    environment:
+      - DATABASE_URL=sqlite+aiosqlite:////data/db/mulita.db
+      - REDIS_URL=redis://redis:6379
+      - CELERY_BROKER_URL=redis://redis:6379
+      - CELERY_RESULT_BACKEND=redis://redis:6379
+      - PHOTO_DIRS=${PHOTO_DIRS:-/photos}
+      - CELERYD_CONCURRENCY=${CELERYD_CONCURRENCY:-4}
+      - LOG_LEVEL=${LOG_LEVEL:-INFO}
+      - TZ=${TZ:-UTC}
+    depends_on:
+      redis:
+        condition: service_started
+      backend:
+        condition: service_started

docker-compose.yml

@@ -1,5 +1,3 @@
-version: '3.8'
-
 services:
   frontend:
     build:
@@ -36,9 +34,13 @@ services:
       - ${PHOTO_DIRS:-./photos}:/photos:rw
       - thumbs_data:/data/thumbs
       - proxies_data:/data/proxies
-      - db_data:/data/db
+      - db_data:/data/db   # retained so the docker-compose.sqlite.yml override has somewhere to put mulita.db
+    # Run Alembic migrations before starting uvicorn. On a fresh Postgres
+    # the empty 0001 baseline is a no-op stamp; create_all in init_db then
+    # builds the schema.
+    command: sh -c "alembic upgrade head && uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload"
     environment:
-      - DATABASE_URL=sqlite+aiosqlite:////data/db/mulita.db
+      - DATABASE_URL=postgresql+asyncpg://mulita:mulita@db:5432/mulita
       - REDIS_URL=redis://redis:6379
       - CELERY_BROKER_URL=redis://redis:6379
       - CELERY_RESULT_BACKEND=redis://redis:6379
@@ -47,7 +49,10 @@ services:
       - LOG_LEVEL=${LOG_LEVEL:-INFO}
       - TZ=${TZ:-UTC}
     depends_on:
-      - redis
+      redis:
+        condition: service_started
+      db:
+        condition: service_healthy
     networks:
       - mulita-network
     restart: unless-stopped
@@ -57,15 +62,16 @@ services:
       context: ./backend
       dockerfile: Dockerfile
     container_name: mulita-worker
-    command: celery -A app.tasks.celery worker --loglevel=${LOG_LEVEL:-info} --concurrency=${CELERYD_CONCURRENCY:-4}
+    command: sh -c "python -m app.services.vision.bootstrap_models && celery -A app.tasks.celery worker --loglevel=${LOG_LEVEL:-info} --concurrency=${CELERYD_CONCURRENCY:-4} -Q default,high,low,vision"
     volumes:
       - ./mulita.yml:/app/config/mulita.yml:ro
       - ${PHOTO_DIRS:-./photos}:/photos:rw
       - thumbs_data:/data/thumbs
       - proxies_data:/data/proxies
       - db_data:/data/db
+      - models_data:/data/models
     environment:
-      - DATABASE_URL=sqlite+aiosqlite:////data/db/mulita.db
+      - DATABASE_URL=postgresql+asyncpg://mulita:mulita@db:5432/mulita
       - REDIS_URL=redis://redis:6379
       - CELERY_BROKER_URL=redis://redis:6379
       - CELERY_RESULT_BACKEND=redis://redis:6379
@@ -74,12 +80,34 @@ services:
       - LOG_LEVEL=${LOG_LEVEL:-INFO}
       - TZ=${TZ:-UTC}
     depends_on:
-      - redis
+      redis:
-      - backend
+        condition: service_started
+      backend:
+        condition: service_started
+      db:
+        condition: service_healthy
     networks:
       - mulita-network
     restart: unless-stopped
 
+  db:
+    image: pgvector/pgvector:pg16
+    container_name: mulita-db
+    environment:
+      POSTGRES_USER: mulita
+      POSTGRES_PASSWORD: mulita
+      POSTGRES_DB: mulita
+    volumes:
+      - pg_data:/var/lib/postgresql/data
+    networks:
+      - mulita-network
+    restart: unless-stopped
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U mulita -d mulita"]
+      interval: 5s
+      timeout: 5s
+      retries: 10
+
   redis:
     image: redis:7-alpine
     container_name: mulita-redis
@@ -102,4 +130,6 @@ volumes:
   thumbs_data:
   proxies_data:
   db_data:
-  redis_data:
+  redis_data:
+  pg_data:
+  models_data:

frontend/src/App.tsx

@@ -2,6 +2,10 @@ import { useState } from 'react'
 import { Timeline } from './components/timeline/Timeline'
 import { DuplicatesView } from './components/duplicates/DuplicatesView'
 import { MapView } from './components/map/MapView'
+import { PeopleView } from './components/people/PeopleView'
+import { TagsView } from './components/tags/TagsView'
+import { ColorsView } from './components/colors/ColorsView'
+import { RatedView } from './components/rated/RatedView'
 import { LeftSidebar } from './components/layout/LeftSidebar'
 import { RightSidebar } from './components/layout/RightSidebar'
 import { TopBar } from './components/layout/TopBar'
@@ -86,6 +90,14 @@ function App() {
               <MapView />
             ) : currentSection === 'duplicates' ? (
               <DuplicatesView />
+            ) : currentSection === 'people' ? (
+              <PeopleView />
+            ) : currentSection === 'tags' ? (
+              <TagsView />
+            ) : currentSection === 'colors' ? (
+              <ColorsView />
+            ) : currentSection === 'rated' ? (
+              <RatedView />
             ) : (
               <Timeline />
             )}

frontend/src/components/colors/ColorsView.tsx

@@ -0,0 +1,183 @@
+import { useState, useMemo, useCallback } from 'react'
+import { Palette, ArrowLeft, Loader2 } from 'lucide-react'
+import clsx from 'clsx'
+import { photos as photosApi } from '../../services/api'
+import { useFilterStore } from '../../store/filterStore'
+import { usePhotosQuery } from '../../hooks/usePhotosQuery'
+import { COLOR_LABEL_OPTIONS, type ColorLabel } from '../../constants/colorLabels'
+import { useCardGridNav } from '../../hooks/useCardGridNav'
+import { Timeline } from '../timeline/Timeline'
+import type { Photo } from '../../types/photo'
+
+interface ColorGroup {
+  label: string
+  value: ColorLabel | null
+  className: string
+  count: number
+  representative: Photo | null
+}
+
+/**
+ * Colors view — two states:
+ * 1. Grid of color label cards (default) — arrow keys + Enter to browse
+ * 2. Detail view showing a color's photos in the full Timeline — Esc to go back
+ */
+export function ColorsView() {
+  const { data: allPhotos = [], isLoading } = usePhotosQuery()
+  const setColorLabel = useFilterStore((s) => s.setColorLabel)
+  const [selectedGroup, setSelectedGroup] = useState<ColorGroup | null>(null)
+
+  const groups = useMemo(() => {
+    const buckets = new Map<string, Photo[]>()
+    const uncolored: Photo[] = []
+
+    for (const photo of allPhotos) {
+      if (photo.color_label) {
+        const arr = buckets.get(photo.color_label) ?? []
+        arr.push(photo)
+        buckets.set(photo.color_label, arr)
+      } else {
+        uncolored.push(photo)
+      }
+    }
+
+    const result: ColorGroup[] = []
+    for (const { value, className } of COLOR_LABEL_OPTIONS) {
+      const photos = buckets.get(value) ?? []
+      if (photos.length === 0) continue
+      result.push({
+        label: value.charAt(0).toUpperCase() + value.slice(1),
+        value,
+        className,
+        count: photos.length,
+        representative: photos[0],
+      })
+    }
+    if (uncolored.length > 0) {
+      result.push({
+        label: 'Uncolored',
+        value: null,
+        className: 'bg-neutral-400',
+        count: uncolored.length,
+        representative: uncolored[0],
+      })
+    }
+    return result
+  }, [allPhotos])
+
+  const enterDetail = useCallback(
+    (group: ColorGroup) => {
+      setColorLabel((group.value ?? 'none') as ColorLabel)
+      setSelectedGroup(group)
+    },
+    [setColorLabel]
+  )
+
+  const exitDetail = useCallback(() => {
+    setColorLabel(null)
+    setSelectedGroup(null)
+  }, [setColorLabel])
+
+  const { activeIndex, gridRef } = useCardGridNav({
+    items: groups,
+    inDetail: selectedGroup !== null,
+    onEnter: enterDetail,
+    onExit: exitDetail,
+  })
+
+  if (selectedGroup) {
+    return (
+      <div className="flex h-full flex-col overflow-hidden">
+        <div className="flex items-center gap-3 border-b border-border px-4 py-3">
+          <button
+            onClick={exitDetail}
+            className="rounded p-1 text-text-muted transition-colors hover:bg-surface-2 hover:text-text"
+            title="Back to colors"
+          >
+            <ArrowLeft className="h-4 w-4" />
+          </button>
+          <div className="flex items-center gap-2">
+            <span className={`inline-block h-3 w-3 rounded-full ${selectedGroup.className}`} />
+            <h2 className="text-sm font-semibold text-text">{selectedGroup.label}</h2>
+          </div>
+        </div>
+        <div className="flex-1 overflow-hidden">
+          <Timeline />
+        </div>
+      </div>
+    )
+  }
+
+  if (isLoading) {
+    return (
+      <div className="flex h-full items-center justify-center text-text-muted">
+        <Loader2 className="mr-2 h-5 w-5 animate-spin" />
+        Loading colors...
+      </div>
+    )
+  }
+
+  if (groups.length === 0) {
+    return (
+      <div className="flex h-full flex-col items-center justify-center gap-3 text-text-muted">
+        <Palette className="h-12 w-12 opacity-40" />
+        <p className="text-sm">No color labels assigned yet</p>
+        <p className="max-w-xs text-center text-xs opacity-70">
+          Color labels will appear here once you assign them to photos.
+        </p>
+      </div>
+    )
+  }
+
+  return (
+    <div className="h-full overflow-auto p-4">
+      <div className="mb-4 flex items-center gap-2 text-text-muted">
+        <Palette className="h-4 w-4" />
+        <span className="text-sm font-medium">
+          {groups.length} {groups.length === 1 ? 'color' : 'colors'}
+        </span>
+      </div>
+
+      <div
+        ref={gridRef}
+        className="grid grid-cols-[repeat(auto-fill,minmax(140px,1fr))] gap-3"
+      >
+        {groups.map((group, i) => (
+          <div
+            key={group.label}
+            className={clsx(
+              'group cursor-pointer overflow-hidden rounded-lg border bg-surface transition-all hover:border-primary/50 hover:shadow-md',
+              i === activeIndex
+                ? 'border-primary ring-1 ring-primary/30'
+                : 'border-border'
+            )}
+            onClick={() => enterDetail(group)}
+          >
+            <div className="relative aspect-square overflow-hidden bg-surface-2">
+              {group.representative ? (
+                <img
+                  src={photosApi.getThumbnailUrl(group.representative.id, 'small')}
+                  alt={group.label}
+                  className="h-full w-full object-cover"
+                />
+              ) : (
+                <div className="flex h-full w-full items-center justify-center">
+                  <Palette className="h-10 w-10 text-text-muted/30" />
+                </div>
+              )}
+
+              <span className="absolute bottom-1.5 right-1.5 rounded-full bg-black/60 px-2 py-0.5 text-[11px] font-medium text-white backdrop-blur-sm">
+                {group.count}
+              </span>
+            </div>
+
+            <div className="flex items-center gap-1.5 px-2 py-1.5">
+              <span className={`inline-block h-2.5 w-2.5 rounded-full ${group.className}`} />
+              <p className="truncate text-xs font-medium text-text">{group.label}</p>
+            </div>
+          </div>
+        ))}
+      </div>
+    </div>
+  )
+}

frontend/src/components/dialogs/SettingsDialog.tsx

@@ -195,7 +195,7 @@ export function SettingsDialog({ isOpen, onClose }: SettingsDialogProps) {
   if (!isOpen) return null
 
   return (
-    <div className="fixed inset-0 z-50">
+    <div className="fixed inset-0 z-[2000]">
       <div
         className="absolute inset-0 bg-black/60 backdrop-blur-sm"
         onClick={onClose}

frontend/src/components/layout/LeftSidebar.tsx

@@ -17,6 +17,7 @@ import {
   Pencil,
   PanelLeftClose,
   Settings,
+  Users,
 } from 'lucide-react'
 import clsx from 'clsx'
 import { sourceFolders, photos as photosApi, type FolderTreeNode } from '../../services/api'
@@ -63,6 +64,7 @@ export function LeftSidebar({ onCollapse, onOpenSettings }: LeftSidebarProps) {
   const navigateToSection = useFilterStore((s) => s.navigateToSection)
   const currentSection = useFilterStore((s) => s.currentSection)
   const { data: allTags = [] } = useTagsQuery()
+  const { data: faceClusters = [] } = useTagsQuery('face_cluster')
   const { data: stats } = useLibraryStatsQuery()
   const [dropTargetId, setDropTargetId] = useState<string | null>(null)
 
@@ -255,6 +257,9 @@ export function LeftSidebar({ onCollapse, onOpenSettings }: LeftSidebarProps) {
       case 'tags':
         navigateToSection('tags', { groupBy: 'tag' })
         break
+      case 'people':
+        navigateToSection('people', { groupBy: 'tag' })
+        break
       case 'colors':
         navigateToSection('colors', { groupBy: 'color' })
         break
@@ -349,8 +354,10 @@ export function LeftSidebar({ onCollapse, onOpenSettings }: LeftSidebarProps) {
       : undefined,
   })
 
-  // Total tag count for the badge on the Tags entry.
+  // Total tag count for the badge on the Tags entry (user tags only).
-  const tagsTotalCount = allTags.reduce((sum, t) => sum + (t.photo_count || 0), 0)
+  const userTags = allTags.filter((t) => t.kind === 'user')
+  const tagsTotalCount = userTags.reduce((sum, t) => sum + (t.photo_count || 0), 0)
+  const peopleTotalCount = faceClusters.reduce((sum, t) => sum + (t.photo_count || 0), 0)
 
   const libraryTree: TreeItem[] = [
     {
@@ -361,6 +368,7 @@ export function LeftSidebar({ onCollapse, onOpenSettings }: LeftSidebarProps) {
         { id: 'all-photos', label: 'All Photos', icon: <Image className="h-4 w-4" />, count: stats?.all_photos ?? 0 },
         { id: 'rated', label: 'Rated', icon: <Star className="h-4 w-4" />, count: stats?.rated ?? 0 },
         { id: 'tags', label: 'Tags', icon: <TagIcon className="h-4 w-4" />, count: tagsTotalCount },
+        { id: 'people', label: 'People', icon: <Users className="h-4 w-4" />, count: peopleTotalCount },
         { id: 'colors', label: 'Colors', icon: <Palette className="h-4 w-4" />, count: stats?.colored ?? 0 },
         { id: 'map', label: 'Map', icon: <MapPin className="h-4 w-4" />, count: stats?.with_gps ?? 0 },
         { id: 'duplicates', label: 'Duplicates', icon: <Copy className="h-4 w-4" />, count: stats?.duplicates ?? 0 },

frontend/src/components/people/PeopleView.tsx

@@ -0,0 +1,237 @@
+import { useState, useCallback } from 'react'
+import { Users, Pencil, Check, X, Loader2, ArrowLeft } from 'lucide-react'
+import clsx from 'clsx'
+import { useMutation, useQueryClient } from '@tanstack/react-query'
+import { useTagsQuery } from '../../hooks/useTagsQuery'
+import {
+  tags as tagsApi,
+  photos as photosApi,
+  type Tag,
+} from '../../services/api'
+import { useFilterStore } from '../../store/filterStore'
+import { useCardGridNav } from '../../hooks/useCardGridNav'
+import { Timeline } from '../timeline/Timeline'
+import { toast } from '../ToastContainer'
+
+/**
+ * People view — two states:
+ * 1. Grid of face cluster cards (default) — arrow keys + Enter to browse
+ * 2. Detail view showing a person's photos in the full Timeline — Esc to go back
+ */
+export function PeopleView() {
+  const { data: clusters = [], isLoading } = useTagsQuery('face_cluster')
+  const queryClient = useQueryClient()
+  const setTagIds = useFilterStore((s) => s.setTagIds)
+
+  const [selectedPerson, setSelectedPerson] = useState<Tag | null>(null)
+  const [editingId, setEditingId] = useState<string | null>(null)
+  const [editName, setEditName] = useState('')
+
+  const renameMutation = useMutation({
+    mutationFn: ({ id, name }: { id: string; name: string }) =>
+      tagsApi.update(id, { name }),
+    onSuccess: () => {
+      queryClient.invalidateQueries({ queryKey: ['tags'] })
+      setEditingId(null)
+      if (selectedPerson && editingId === selectedPerson.id) {
+        setSelectedPerson({ ...selectedPerson, name: editName.trim() })
+      }
+      toast.success('Renamed')
+    },
+    onError: (e: any) =>
+      toast.error('Rename failed', e?.response?.data?.detail || e.message),
+  })
+
+  const startEditing = (tag: Tag) => {
+    setEditingId(tag.id)
+    setEditName(tag.name)
+  }
+
+  const submitRename = () => {
+    if (!editingId || !editName.trim()) return
+    renameMutation.mutate({ id: editingId, name: editName.trim() })
+  }
+
+  const enterDetail = useCallback(
+    (person: Tag) => {
+      setTagIds([person.id])
+      setSelectedPerson(person)
+    },
+    [setTagIds]
+  )
+
+  const exitDetail = useCallback(() => {
+    setTagIds([])
+    setSelectedPerson(null)
+  }, [setTagIds])
+
+  const { activeIndex, gridRef } = useCardGridNav({
+    items: clusters,
+    inDetail: selectedPerson !== null,
+    onEnter: enterDetail,
+    onExit: exitDetail,
+  })
+
+  // ── Detail view: a person's photos ─────────────────────────────────
+  if (selectedPerson) {
+    const isEditing = editingId === selectedPerson.id
+    return (
+      <div className="flex h-full flex-col overflow-hidden">
+        <div className="flex items-center gap-3 border-b border-border px-4 py-3">
+          <button
+            onClick={exitDetail}
+            className="rounded p-1 text-text-muted transition-colors hover:bg-surface-2 hover:text-text"
+            title="Back to people"
+          >
+            <ArrowLeft className="h-4 w-4" />
+          </button>
+
+          {isEditing ? (
+            <div className="flex items-center gap-1.5">
+              <input
+                autoFocus
+                value={editName}
+                onChange={(e) => setEditName(e.target.value)}
+                onKeyDown={(e) => {
+                  if (e.key === 'Enter') submitRename()
+                  if (e.key === 'Escape') setEditingId(null)
+                }}
+                className="rounded border border-border bg-bg px-2 py-1 text-sm text-text focus:border-primary focus:outline-none"
+              />
+              <button onClick={submitRename} className="rounded p-1 text-green-500 hover:bg-green-500/10">
+                <Check className="h-4 w-4" />
+              </button>
+              <button onClick={() => setEditingId(null)} className="rounded p-1 text-text-muted hover:bg-surface-2">
+                <X className="h-4 w-4" />
+              </button>
+            </div>
+          ) : (
+            <div className="flex items-center gap-2">
+              <h2 className="text-sm font-semibold text-text">{selectedPerson.name}</h2>
+              <button
+                onClick={() => startEditing(selectedPerson)}
+                className="rounded p-0.5 text-text-muted transition-colors hover:text-text"
+                title="Rename"
+              >
+                <Pencil className="h-3.5 w-3.5" />
+              </button>
+            </div>
+          )}
+        </div>
+        <div className="flex-1 overflow-hidden">
+          <Timeline />
+        </div>
+      </div>
+    )
+  }
+
+  // ── Card grid ──────────────────────────────────────────────────────
+  if (isLoading) {
+    return (
+      <div className="flex h-full items-center justify-center text-text-muted">
+        <Loader2 className="mr-2 h-5 w-5 animate-spin" />
+        Loading people...
+      </div>
+    )
+  }
+
+  if (clusters.length === 0) {
+    return (
+      <div className="flex h-full flex-col items-center justify-center gap-3 text-text-muted">
+        <Users className="h-12 w-12 opacity-40" />
+        <p className="text-sm">No people identified yet</p>
+        <p className="max-w-xs text-center text-xs opacity-70">
+          Face detection runs automatically when photos are scanned.
+          People will appear here once faces are found and clustered.
+        </p>
+      </div>
+    )
+  }
+
+  return (
+    <div className="h-full overflow-auto p-4">
+      <div className="mb-4 flex items-center gap-2 text-text-muted">
+        <Users className="h-4 w-4" />
+        <span className="text-sm font-medium">
+          {clusters.length} {clusters.length === 1 ? 'person' : 'people'} identified
+        </span>
+      </div>
+
+      <div
+        ref={gridRef}
+        className="grid grid-cols-[repeat(auto-fill,minmax(140px,1fr))] gap-3"
+      >
+        {clusters.map((tag, i) => (
+          <div
+            key={tag.id}
+            className={clsx(
+              'group cursor-pointer overflow-hidden rounded-lg border bg-surface transition-all hover:border-primary/50 hover:shadow-md',
+              i === activeIndex
+                ? 'border-primary ring-1 ring-primary/30'
+                : editingId === tag.id
+                  ? 'border-primary ring-1 ring-primary/30'
+                  : 'border-border'
+            )}
+            onClick={() => {
+              if (editingId !== tag.id) enterDetail(tag)
+            }}
+          >
+            <div className="relative aspect-square overflow-hidden bg-surface-2">
+              {tag.representative_photo_id ? (
+                <img
+                  src={photosApi.getThumbnailUrl(tag.representative_photo_id, 'small')}
+                  alt={tag.name}
+                  className="h-full w-full object-cover"
+                />
+              ) : (
+                <div className="flex h-full w-full items-center justify-center">
+                  <Users className="h-10 w-10 text-text-muted/30" />
+                </div>
+              )}
+
+              <span className="absolute bottom-1.5 right-1.5 rounded-full bg-black/60 px-2 py-0.5 text-[11px] font-medium text-white backdrop-blur-sm">
+                {tag.photo_count}
+              </span>
+
+              <button
+                className="absolute right-1.5 top-1.5 rounded-full bg-black/50 p-1 text-white opacity-0 transition-opacity group-hover:opacity-100"
+                onClick={(e) => {
+                  e.stopPropagation()
+                  startEditing(tag)
+                }}
+                title="Rename"
+              >
+                <Pencil className="h-3 w-3" />
+              </button>
+            </div>
+
+            <div className="px-2 py-1.5">
+              {editingId === tag.id ? (
+                <div className="flex items-center gap-1" onClick={(e) => e.stopPropagation()}>
+                  <input
+                    autoFocus
+                    value={editName}
+                    onChange={(e) => setEditName(e.target.value)}
+                    onKeyDown={(e) => {
+                      if (e.key === 'Enter') submitRename()
+                      if (e.key === 'Escape') setEditingId(null)
+                    }}
+                    className="min-w-0 flex-1 rounded border border-border bg-bg px-1.5 py-0.5 text-xs text-text focus:border-primary focus:outline-none"
+                  />
+                  <button onClick={submitRename} className="rounded p-0.5 text-green-500 hover:bg-green-500/10">
+                    <Check className="h-3 w-3" />
+                  </button>
+                  <button onClick={() => setEditingId(null)} className="rounded p-0.5 text-text-muted hover:bg-surface-2">
+                    <X className="h-3 w-3" />
+                  </button>
+                </div>
+              ) : (
+                <p className="truncate text-xs font-medium text-text">{tag.name}</p>
+              )}
+            </div>
+          </div>
+        ))}
+      </div>
+    </div>
+  )
+}

frontend/src/components/rated/RatedView.tsx

@@ -0,0 +1,170 @@
+import { useState, useMemo, useCallback } from 'react'
+import { Star, ArrowLeft, Loader2 } from 'lucide-react'
+import clsx from 'clsx'
+import { photos as photosApi } from '../../services/api'
+import { useFilterStore } from '../../store/filterStore'
+import { usePhotosQuery } from '../../hooks/usePhotosQuery'
+import { useCardGridNav } from '../../hooks/useCardGridNav'
+import { Timeline } from '../timeline/Timeline'
+import type { Photo } from '../../types/photo'
+
+interface RatingGroup {
+  rating: number
+  label: string
+  count: number
+  representative: Photo | null
+}
+
+/**
+ * Rated view — two states:
+ * 1. Grid of rating-level cards (default) — arrow keys + Enter to browse
+ * 2. Detail view showing a rating level's photos in the full Timeline — Esc to go back
+ */
+export function RatedView() {
+  const { data: allPhotos = [], isLoading } = usePhotosQuery()
+  const setRatingMin = useFilterStore((s) => s.setRatingMin)
+  const setRatingMax = useFilterStore((s) => s.setRatingMax)
+  const [selectedGroup, setSelectedGroup] = useState<RatingGroup | null>(null)
+
+  const groups = useMemo(() => {
+    const buckets = new Map<number, Photo[]>()
+
+    for (const photo of allPhotos) {
+      if (photo.rating > 0) {
+        const arr = buckets.get(photo.rating) ?? []
+        arr.push(photo)
+        buckets.set(photo.rating, arr)
+      }
+    }
+
+    // Highest rating first
+    const result: RatingGroup[] = []
+    for (let r = 5; r >= 1; r--) {
+      const photos = buckets.get(r) ?? []
+      if (photos.length === 0) continue
+      result.push({
+        rating: r,
+        label: '★'.repeat(r),
+        count: photos.length,
+        representative: photos[0],
+      })
+    }
+    return result
+  }, [allPhotos])
+
+  const enterDetail = useCallback(
+    (group: RatingGroup) => {
+      setRatingMin(group.rating)
+      setRatingMax(group.rating)
+      setSelectedGroup(group)
+    },
+    [setRatingMin, setRatingMax]
+  )
+
+  const exitDetail = useCallback(() => {
+    // Restore the section preset: ratingMin=1 (all rated), no max
+    setRatingMin(1)
+    setRatingMax(0)
+    setSelectedGroup(null)
+  }, [setRatingMin, setRatingMax])
+
+  const { activeIndex, gridRef } = useCardGridNav({
+    items: groups,
+    inDetail: selectedGroup !== null,
+    onEnter: enterDetail,
+    onExit: exitDetail,
+  })
+
+  if (selectedGroup) {
+    return (
+      <div className="flex h-full flex-col overflow-hidden">
+        <div className="flex items-center gap-3 border-b border-border px-4 py-3">
+          <button
+            onClick={exitDetail}
+            className="rounded p-1 text-text-muted transition-colors hover:bg-surface-2 hover:text-text"
+            title="Back to ratings"
+          >
+            <ArrowLeft className="h-4 w-4" />
+          </button>
+          <h2 className="text-sm font-semibold text-amber-400">{selectedGroup.label}</h2>
+        </div>
+        <div className="flex-1 overflow-hidden">
+          <Timeline />
+        </div>
+      </div>
+    )
+  }
+
+  if (isLoading) {
+    return (
+      <div className="flex h-full items-center justify-center text-text-muted">
+        <Loader2 className="mr-2 h-5 w-5 animate-spin" />
+        Loading ratings...
+      </div>
+    )
+  }
+
+  if (groups.length === 0) {
+    return (
+      <div className="flex h-full flex-col items-center justify-center gap-3 text-text-muted">
+        <Star className="h-12 w-12 opacity-40" />
+        <p className="text-sm">No rated photos yet</p>
+        <p className="max-w-xs text-center text-xs opacity-70">
+          Rate photos with 1–5 stars and they will appear here grouped by
+          rating.
+        </p>
+      </div>
+    )
+  }
+
+  return (
+    <div className="h-full overflow-auto p-4">
+      <div className="mb-4 flex items-center gap-2 text-text-muted">
+        <Star className="h-4 w-4" />
+        <span className="text-sm font-medium">
+          {groups.length} rating {groups.length === 1 ? 'level' : 'levels'}
+        </span>
+      </div>
+
+      <div
+        ref={gridRef}
+        className="grid grid-cols-[repeat(auto-fill,minmax(140px,1fr))] gap-3"
+      >
+        {groups.map((group, i) => (
+          <div
+            key={group.rating}
+            className={clsx(
+              'group cursor-pointer overflow-hidden rounded-lg border bg-surface transition-all hover:border-primary/50 hover:shadow-md',
+              i === activeIndex
+                ? 'border-primary ring-1 ring-primary/30'
+                : 'border-border'
+            )}
+            onClick={() => enterDetail(group)}
+          >
+            <div className="relative aspect-square overflow-hidden bg-surface-2">
+              {group.representative ? (
+                <img
+                  src={photosApi.getThumbnailUrl(group.representative.id, 'small')}
+                  alt={group.label}
+                  className="h-full w-full object-cover"
+                />
+              ) : (
+                <div className="flex h-full w-full items-center justify-center">
+                  <Star className="h-10 w-10 text-text-muted/30" />
+                </div>
+              )}
+
+              <span className="absolute bottom-1.5 right-1.5 rounded-full bg-black/60 px-2 py-0.5 text-[11px] font-medium text-white backdrop-blur-sm">
+                {group.count}
+              </span>
+            </div>
+
+            <div className="px-2 py-1.5">
+              <p className="truncate text-xs font-medium text-amber-400">{group.label}</p>
+            </div>
+          </div>
+        ))}
+      </div>
+    </div>
+  )
+}

frontend/src/components/tags/TagsView.tsx

@@ -0,0 +1,138 @@
+import { useState, useMemo, useCallback } from 'react'
+import { Tag as TagIcon, ArrowLeft, Loader2 } from 'lucide-react'
+import clsx from 'clsx'
+import { useTagsQuery } from '../../hooks/useTagsQuery'
+import { photos as photosApi, type Tag } from '../../services/api'
+import { useFilterStore } from '../../store/filterStore'
+import { useCardGridNav } from '../../hooks/useCardGridNav'
+import { Timeline } from '../timeline/Timeline'
+
+/**
+ * Tags view — two states:
+ * 1. Grid of tag cards (default) — arrow keys + Enter to browse
+ * 2. Detail view showing a tag's photos in the full Timeline — Esc to go back
+ */
+export function TagsView() {
+  const { data: allTags = [], isLoading } = useTagsQuery()
+  const setTagIds = useFilterStore((s) => s.setTagIds)
+  const [selectedTag, setSelectedTag] = useState<Tag | null>(null)
+
+  // Exclude face_cluster tags (those live in PeopleView)
+  const tags = useMemo(
+    () => allTags.filter((t) => t.kind !== 'face_cluster'),
+    [allTags]
+  )
+
+  const enterDetail = useCallback(
+    (tag: Tag) => {
+      setTagIds([tag.id])
+      setSelectedTag(tag)
+    },
+    [setTagIds]
+  )
+
+  const exitDetail = useCallback(() => {
+    setTagIds([])
+    setSelectedTag(null)
+  }, [setTagIds])
+
+  const { activeIndex, gridRef } = useCardGridNav({
+    items: tags,
+    inDetail: selectedTag !== null,
+    onEnter: enterDetail,
+    onExit: exitDetail,
+  })
+
+  if (selectedTag) {
+    return (
+      <div className="flex h-full flex-col overflow-hidden">
+        <div className="flex items-center gap-3 border-b border-border px-4 py-3">
+          <button
+            onClick={exitDetail}
+            className="rounded p-1 text-text-muted transition-colors hover:bg-surface-2 hover:text-text"
+            title="Back to tags"
+          >
+            <ArrowLeft className="h-4 w-4" />
+          </button>
+          <h2 className="text-sm font-semibold text-text">{selectedTag.name}</h2>
+        </div>
+        <div className="flex-1 overflow-hidden">
+          <Timeline />
+        </div>
+      </div>
+    )
+  }
+
+  if (isLoading) {
+    return (
+      <div className="flex h-full items-center justify-center text-text-muted">
+        <Loader2 className="mr-2 h-5 w-5 animate-spin" />
+        Loading tags...
+      </div>
+    )
+  }
+
+  if (tags.length === 0) {
+    return (
+      <div className="flex h-full flex-col items-center justify-center gap-3 text-text-muted">
+        <TagIcon className="h-12 w-12 opacity-40" />
+        <p className="text-sm">No tags yet</p>
+        <p className="max-w-xs text-center text-xs opacity-70">
+          Tags will appear here once photos are tagged — either manually or by
+          the auto-tagger.
+        </p>
+      </div>
+    )
+  }
+
+  return (
+    <div className="h-full overflow-auto p-4">
+      <div className="mb-4 flex items-center gap-2 text-text-muted">
+        <TagIcon className="h-4 w-4" />
+        <span className="text-sm font-medium">
+          {tags.length} {tags.length === 1 ? 'tag' : 'tags'}
+        </span>
+      </div>
+
+      <div
+        ref={gridRef}
+        className="grid grid-cols-[repeat(auto-fill,minmax(140px,1fr))] gap-3"
+      >
+        {tags.map((tag, i) => (
+          <div
+            key={tag.id}
+            className={clsx(
+              'group cursor-pointer overflow-hidden rounded-lg border bg-surface transition-all hover:border-primary/50 hover:shadow-md',
+              i === activeIndex
+                ? 'border-primary ring-1 ring-primary/30'
+                : 'border-border'
+            )}
+            onClick={() => enterDetail(tag)}
+          >
+            <div className="relative aspect-square overflow-hidden bg-surface-2">
+              {tag.representative_photo_id ? (
+                <img
+                  src={photosApi.getThumbnailUrl(tag.representative_photo_id, 'small')}
+                  alt={tag.name}
+                  className="h-full w-full object-cover"
+                />
+              ) : (
+                <div className="flex h-full w-full items-center justify-center">
+                  <TagIcon className="h-10 w-10 text-text-muted/30" />
+                </div>
+              )}
+
+              <span className="absolute bottom-1.5 right-1.5 rounded-full bg-black/60 px-2 py-0.5 text-[11px] font-medium text-white backdrop-blur-sm">
+                {tag.photo_count}
+              </span>
+            </div>
+
+            <div className="px-2 py-1.5">
+              <p className="truncate text-xs font-medium text-text">{tag.name}</p>
+            </div>
+          </div>
+        ))}
+      </div>
+    </div>
+  )
+}

frontend/src/components/timeline/Timeline.tsx

@@ -6,7 +6,6 @@ import { useFilterStore } from '../../store/filterStore'
 import { PhotoThumbnail } from './PhotoThumbnail'
 import { usePhotosQuery } from '../../hooks/usePhotosQuery'
 import { useActiveHeapMembers } from '../../hooks/useActiveHeapMembersQuery'
-import { COLOR_LABEL_OPTIONS } from '../../constants/colorLabels'
 import type { Photo } from '../../types/photo'
 
 // Layout constants for the grid + grouped headers.
@@ -27,22 +26,19 @@ type TimelineItem =
 /**
  * Build the flat header|row item array the virtualizer renders.
  *
- * Five modes:
+ * Two modes:
- *  - groupBy='tag': one bucket per unique tag (plus an "Untagged" bucket
+ *  - sortBy is a date field: month buckets.
- *    for photos with no tags). A photo with N tags appears in N buckets.
- *  - groupBy='rating': one bucket per star rating 5..1 (plus "Unrated"
- *    for rating 0). Each photo lands in exactly one bucket.
- *  - groupBy='color': one bucket per color label, in canonical order
- *    (plus an "Uncolored" bucket for photos with no label).
- *  - groupBy='date' AND sortBy is a date field: month buckets (existing).
  *  - otherwise: one un-headered stream.
+ *
+ * Tag, rating, and color grouping now live in their own dedicated views
+ * (TagsView, RatedView, ColorsView) instead of being handled here.
  */
 function buildItems(
   photos: Photo[],
   columns: number,
   rowHeight: number,
   sortBy: string,
-  groupBy: 'date' | 'tag' | 'rating' | 'color'
+  groupBy: string,
 ): TimelineItem[] {
   if (photos.length === 0) return []
 
@@ -61,155 +57,13 @@ function buildItems(
     }
   }
 
-  // ── Tag grouping ──────────────────────────────────────────────────────
+  // Date grouping only applies when groupBy is explicitly 'date' and
-  if (groupBy === 'tag') {
+  // the sort field is a date column. Other sections (tags, colors,
-    // Bucket by tag name. A photo with multiple tags lands in multiple
+  // rated, people) reuse Timeline for their detail views and should
-    // buckets. Photos with no tags go into "Untagged".
+  // render a flat grid without month headers.
-    const tagBuckets = new Map<string, PhotoCell[]>()
-    const untagged: PhotoCell[] = []
-
-    photos.forEach((photo, globalIndex) => {
-      const cell: PhotoCell = { photo, globalIndex }
-      const tags = photo.tags ?? []
-      if (tags.length === 0) {
-        untagged.push(cell)
-      } else {
-        for (const t of tags) {
-          const arr = tagBuckets.get(t.name) ?? []
-          arr.push(cell)
-          tagBuckets.set(t.name, arr)
-        }
-      }
-    })
-
-    // Sort tag groups alphabetically; Untagged goes at the end.
-    const sortedTagNames = Array.from(tagBuckets.keys()).sort((a, b) =>
-      a.localeCompare(b)
-    )
-
-    let bucketIndex = 0
-    for (const name of sortedTagNames) {
-      items.push({
-        type: 'header',
-        key: `tag::${bucketIndex}::${name}`,
-        label: name,
-        height: HEADER_HEIGHT,
-      })
-      pushRowsForGroup(`tag::${bucketIndex}::${name}`, tagBuckets.get(name)!)
-      bucketIndex++
-    }
-    if (untagged.length > 0) {
-      items.push({
-        type: 'header',
-        key: `tag::${bucketIndex}::__untagged`,
-        label: 'Untagged',
-        height: HEADER_HEIGHT,
-      })
-      pushRowsForGroup(`tag::${bucketIndex}::untagged`, untagged)
-    }
-    return items
-  }
-
-  // ── Rating grouping ───────────────────────────────────────────────────
-  if (groupBy === 'rating') {
-    // Bucket by star rating. Each photo lands in exactly one bucket;
-    // rating 0 goes into "Unrated".
-    const ratingBuckets = new Map<number, PhotoCell[]>()
-    const unrated: PhotoCell[] = []
-
-    photos.forEach((photo, globalIndex) => {
-      const cell: PhotoCell = { photo, globalIndex }
-      if (photo.rating > 0) {
-        const arr = ratingBuckets.get(photo.rating) ?? []
-        arr.push(cell)
-        ratingBuckets.set(photo.rating, arr)
-      } else {
-        unrated.push(cell)
-      }
-    })
-
-    // Highest rating first; Unrated goes at the end.
-    const sortedRatings = Array.from(ratingBuckets.keys()).sort((a, b) => b - a)
-
-    let bucketIndex = 0
-    for (const rating of sortedRatings) {
-      items.push({
-        type: 'header',
-        key: `rating::${bucketIndex}::${rating}`,
-        label: '★'.repeat(rating),
-        height: HEADER_HEIGHT,
-      })
-      pushRowsForGroup(
-        `rating::${bucketIndex}::${rating}`,
-        ratingBuckets.get(rating)!
-      )
-      bucketIndex++
-    }
-    if (unrated.length > 0) {
-      items.push({
-        type: 'header',
-        key: `rating::${bucketIndex}::__unrated`,
-        label: 'Unrated',
-        height: HEADER_HEIGHT,
-      })
-      pushRowsForGroup(`rating::${bucketIndex}::unrated`, unrated)
-    }
-    return items
-  }
-
-  // ── Color label grouping ──────────────────────────────────────────────
-  if (groupBy === 'color') {
-    // Bucket by color_label. Each photo lands in exactly one bucket;
-    // photos with no label go into "Uncolored".
-    const colorBuckets = new Map<string, PhotoCell[]>()
-    const uncolored: PhotoCell[] = []
-
-    photos.forEach((photo, globalIndex) => {
-      const cell: PhotoCell = { photo, globalIndex }
-      const label = photo.color_label
-      if (label) {
-        const arr = colorBuckets.get(label) ?? []
-        arr.push(cell)
-        colorBuckets.set(label, arr)
-      } else {
-        uncolored.push(cell)
-      }
-    })
-
-    // Walk the canonical color order so headers always read R-O-Y-G-B-P,
-    // matching every other color UI in the app. Skip empty buckets and
-    // ignore any unexpected label values that aren't in the canonical
-    // list (they'd be invalid backend state).
-    let bucketIndex = 0
-    for (const { value } of COLOR_LABEL_OPTIONS) {
-      const cells = colorBuckets.get(value)
-      if (!cells || cells.length === 0) continue
-      const label = value.charAt(0).toUpperCase() + value.slice(1)
-      items.push({
-        type: 'header',
-        key: `color::${bucketIndex}::${value}`,
-        label,
-        height: HEADER_HEIGHT,
-      })
-      pushRowsForGroup(`color::${bucketIndex}::${value}`, cells)
-      bucketIndex++
-    }
-    if (uncolored.length > 0) {
-      items.push({
-        type: 'header',
-        key: `color::${bucketIndex}::__uncolored`,
-        label: 'Uncolored',
-        height: HEADER_HEIGHT,
-      })
-      pushRowsForGroup(`color::${bucketIndex}::uncolored`, uncolored)
-    }
-    return items
-  }
-
-  // ── Date grouping (existing) ──────────────────────────────────────────
   const isDateSort = sortBy === 'taken_at' || sortBy === 'added_at'
 
-  if (!isDateSort) {
+  if (!isDateSort || groupBy !== 'date') {
     // No grouping — one row stream.
     const cells: PhotoCell[] = photos.map((photo, globalIndex) => ({
       photo,
@@ -344,8 +198,7 @@ export function Timeline() {
   const activeHeapName = activeHeap?.name ?? null
 
   // Build the flat virtualizer items: a mix of group headers and rows of
-  // photos. Date headers appear when sorted by a date field; tag headers
+  // photos. Date headers appear only in the main timeline (groupBy='date').
-  // appear when groupBy === 'tag' (overrides date grouping).
   const items = useMemo(
     () => buildItems(photos, columns, cellSize, sortBy, groupBy),
     [photos, columns, cellSize, sortBy, groupBy]

frontend/src/hooks/useCardGridNav.ts

@@ -0,0 +1,115 @@
+import { useState, useEffect, useCallback, useRef } from 'react'
+import { usePhotoStore } from '../store/photoStore'
+
+/**
+ * Keyboard navigation for card grids (tags, colors, ratings, people).
+ *
+ * Arrow keys move the active index through the grid (wrapping at row
+ * boundaries based on the actual CSS column count), Enter opens the
+ * selected card, and Escape / Backspace exits the detail view.
+ *
+ * In detail mode, Escape only exits back to the card grid when the
+ * preview is closed and no photos are selected — otherwise it defers
+ * to Timeline's own Escape handler (clear selection / close preview).
+ *
+ * The grid container ref is used to measure the rendered column count
+ * so up/down navigation stays column-aligned.
+ */
+export function useCardGridNav<T>(opts: {
+  items: T[]
+  /** True when the detail view is showing (disables grid nav, enables Esc) */
+  inDetail: boolean
+  onEnter: (item: T, index: number) => void
+  onExit: () => void
+}) {
+  const { items, inDetail, onEnter, onExit } = opts
+  const [activeIndex, setActiveIndex] = useState(0)
+  const gridRef = useRef<HTMLDivElement>(null)
+  const viewMode = usePhotoStore((s) => s.viewMode)
+  const selectedPhotos = usePhotoStore((s) => s.selectedPhotos)
+
+  // Clamp active index when the item list shrinks
+  useEffect(() => {
+    if (items.length > 0 && activeIndex >= items.length) {
+      setActiveIndex(items.length - 1)
+    }
+  }, [items.length, activeIndex])
+
+  // Measure column count from the grid container
+  const getColumns = useCallback(() => {
+    const el = gridRef.current
+    if (!el) return 1
+    return getComputedStyle(el).gridTemplateColumns.split(' ').length
+  }, [])
+
+  // Scroll the active card into view
+  const scrollIntoView = useCallback((index: number) => {
+    const el = gridRef.current
+    if (!el) return
+    const card = el.children[index] as HTMLElement | undefined
+    card?.scrollIntoView({ block: 'nearest' })
+  }, [])
+
+  useEffect(() => {
+    if (items.length === 0) return
+
+    const handleKeyDown = (e: KeyboardEvent) => {
+      const target = e.target as HTMLElement | null
+      if (target && (target.tagName === 'INPUT' || target.tagName === 'TEXTAREA')) return
+
+      // Detail view: Escape or Backspace exits back to card grid, but
+      // only when the preview is closed and no photos are selected —
+      // otherwise defer to Timeline's own Escape handler.
+      if (inDetail) {
+        if (e.key === 'Backspace') {
+          e.preventDefault()
+          onExit()
+        } else if (e.key === 'Escape' && viewMode === 'grid' && selectedPhotos.length === 0) {
+          e.preventDefault()
+          onExit()
+        }
+        return
+      }
+
+      // Card grid navigation
+      const cols = getColumns()
+      const count = items.length
+      let next = activeIndex
+
+      switch (e.key) {
+        case 'ArrowRight':
+          e.preventDefault()
+          next = Math.min(activeIndex + 1, count - 1)
+          break
+        case 'ArrowLeft':
+          e.preventDefault()
+          next = Math.max(activeIndex - 1, 0)
+          break
+        case 'ArrowDown':
+          e.preventDefault()
+          next = Math.min(activeIndex + cols, count - 1)
+          break
+        case 'ArrowUp':
+          e.preventDefault()
+          next = Math.max(activeIndex - cols, 0)
+          break
+        case 'Enter':
+          e.preventDefault()
+          if (items[activeIndex]) onEnter(items[activeIndex], activeIndex)
+          return
+        default:
+          return
+      }
+
+      if (next !== activeIndex) {
+        setActiveIndex(next)
+        scrollIntoView(next)
+      }
+    }
+
+    window.addEventListener('keydown', handleKeyDown)
+    return () => window.removeEventListener('keydown', handleKeyDown)
+  }, [items, activeIndex, inDetail, onEnter, onExit, getColumns, scrollIntoView, viewMode, selectedPhotos])
+
+  return { activeIndex, setActiveIndex, gridRef }
+}

frontend/src/hooks/useFilterUrlSync.ts

@@ -60,6 +60,12 @@ function parseUrl(): HydratePayload {
     if (Number.isFinite(n) && n >= 0 && n <= 5) out.ratingMin = n
   }
 
+  const rx = sp.get('rating_max')
+  if (rx) {
+    const n = parseInt(rx, 10)
+    if (Number.isFinite(n) && n >= 0 && n <= 5) out.ratingMax = n
+  }
+
   const cl = sp.get('color_label')
   if (cl && ALLOWED_COLORS.includes(cl as ColorLabel)) {
     out.colorLabel = cl as ColorLabel
@@ -110,6 +116,7 @@ function writeUrl(f: FilterState & { currentSection?: string }) {
   if (f.dateTo) sp.set('date_to', f.dateTo)
   if (f.mediaTypes.length > 0) sp.set('media_type', f.mediaTypes.join(','))
   if (f.ratingMin > 0) sp.set('rating_min', String(f.ratingMin))
+  if (f.ratingMax > 0) sp.set('rating_max', String(f.ratingMax))
   if (f.colorLabel) sp.set('color_label', f.colorLabel)
   if (f.flag !== 'any') sp.set('flag', f.flag)
   if (f.heapId) sp.set('heap_id', f.heapId)

frontend/src/hooks/usePhotosQuery.ts

@@ -32,6 +32,7 @@ export function usePhotosQuery() {
   const dateTo = useFilterStore((s) => s.dateTo)
   const mediaTypes = useFilterStore((s) => s.mediaTypes)
   const ratingMin = useFilterStore((s) => s.ratingMin)
+  const ratingMax = useFilterStore((s) => s.ratingMax)
   const colorLabel = useFilterStore((s) => s.colorLabel)
   const flag = useFilterStore((s) => s.flag)
   const heapId = useFilterStore((s) => s.heapId)
@@ -50,6 +51,7 @@ export function usePhotosQuery() {
         dateTo,
         mediaTypes,
         ratingMin,
+        ratingMax,
         colorLabel,
         flag,
         heapId,
@@ -60,7 +62,7 @@ export function usePhotosQuery() {
         sortBy,
         sortOrder,
       }),
-    [q, dateFrom, dateTo, mediaTypes, ratingMin, colorLabel, flag, heapId, folderId, tagIds, duplicates, groupBy, sortBy, sortOrder]
+    [q, dateFrom, dateTo, mediaTypes, ratingMin, ratingMax, colorLabel, flag, heapId, folderId, tagIds, duplicates, groupBy, sortBy, sortOrder]
   )
 
   const queryClient = useQueryClient()

frontend/src/hooks/useSearchQuery.ts

@@ -0,0 +1,37 @@
+import { useQuery } from '@tanstack/react-query'
+import { search, type SearchResult } from '../services/api'
+import { useFilterStore } from '../store/filterStore'
+
+/**
+ * Hybrid search hook — fires POST /photos/search when the user has a
+ * non-empty search query. Returns results ranked by RRF (FTS + semantic).
+ *
+ * When `q` is empty, this hook is disabled and returns no data — the
+ * normal usePhotosQuery takes over for browse mode.
+ */
+export function useSearchQuery() {
+  const q = useFilterStore((s) => s.q)
+  const tagIds = useFilterStore((s) => s.tagIds)
+  const dateFrom = useFilterStore((s) => s.dateFrom)
+  const dateTo = useFilterStore((s) => s.dateTo)
+
+  const hasQuery = q.trim().length > 0
+
+  return useQuery<SearchResult[]>({
+    queryKey: ['search', q, tagIds, dateFrom, dateTo],
+    queryFn: async () => {
+      const resp = await search.query({
+        q: q.trim(),
+        filters: {
+          tag_ids: tagIds.length > 0 ? tagIds : undefined,
+          date_from: dateFrom ?? undefined,
+          date_to: dateTo ?? undefined,
+        },
+        limit: 200,
+      })
+      return resp.results
+    },
+    enabled: hasQuery,
+    staleTime: 30_000,
+  })
+}

frontend/src/hooks/useTagsQuery.ts

@@ -1,12 +1,12 @@
 import { useQuery } from '@tanstack/react-query'
-import { tags as tagsApi, type Tag } from '../services/api'
+import { tags as tagsApi, type Tag, type TagKind } from '../services/api'
 
 export const TAGS_QUERY_KEY = ['tags'] as const
 
-export function useTagsQuery() {
+export function useTagsQuery(kind?: TagKind) {
   return useQuery<Tag[]>({
-    queryKey: TAGS_QUERY_KEY,
+    queryKey: kind ? ['tags', kind] : TAGS_QUERY_KEY,
-    queryFn: tagsApi.list,
+    queryFn: () => tagsApi.list(kind),
     staleTime: 30_000,
   })
 }

frontend/src/services/api.ts

@@ -519,21 +519,27 @@ export const heaps = {
 }
 
 // Tags API
+export type TagKind = 'user' | 'object' | 'scene' | 'face_cluster'
+
 export interface Tag {
   id: string
   name: string
   color: string | null
+  kind: TagKind
+  source: string | null
+  representative_photo_id: string | null
   photo_count: number
 }
 
 export const tags = {
-  list: async (): Promise<Tag[]> => {
+  list: async (kind?: TagKind): Promise<Tag[]> => {
-    const response = await api.get('/tags')
+    const params = kind ? { kind } : undefined
+    const response = await api.get('/tags', { params })
     return response.data
   },
 
-  create: async (name: string, color?: string): Promise<Tag> => {
+  create: async (name: string, color?: string, kind: TagKind = 'user'): Promise<Tag> => {
-    const response = await api.post('/tags', { name, color })
+    const response = await api.post('/tags', { name, color, kind })
     return response.data
   },
 
@@ -546,6 +552,11 @@ export const tags = {
     await api.delete(`/tags/${tagId}`)
   },
 
+  merge: async (sourceId: string, targetId: string): Promise<{ merged_into: string; target_name: string }> => {
+    const response = await api.post(`/tags/${sourceId}/merge`, { target_id: targetId })
+    return response.data
+  },
+
   /** Add one or more tags to a photo. */
   addToPhoto: async (photoId: string, tagIds: string[]) => {
     const response = await api.post(`/photos/${photoId}/tags`, { tag_ids: tagIds })
@@ -558,6 +569,38 @@ export const tags = {
   },
 }
 
+// Search API — hybrid FTS + semantic search
+export interface SearchResult {
+  id: string
+  filename: string
+  filepath: string
+  media_type: string
+  width: number
+  height: number
+  taken_at: string | null
+  rating: number
+  color_label: string | null
+  thumb_small: string
+  thumb_medium: string
+  score: number
+}
+
+export const search = {
+  query: async (params: {
+    q?: string
+    filters?: {
+      tag_ids?: string[]
+      date_from?: string
+      date_to?: string
+    }
+    limit?: number
+    offset?: number
+  }): Promise<{ results: SearchResult[]; total: number }> => {
+    const response = await api.post('/photos/search', params)
+    return response.data
+  },
+}
+
 // Discard API
 export const discard = {
   list: async () => {

frontend/src/store/filterStore.ts

@@ -19,6 +19,7 @@ export interface FilterState {
   dateTo: string | null
   mediaTypes: MediaType[]
   ratingMin: number // 0-5; 0 means no filter
+  ratingMax: number // 0-5; 0 means no filter
   colorLabel: ColorLabel | null
   flag: FlagFilter
   /** When set, restrict to photos in this heap. Independent of `activeHeapId`
@@ -59,6 +60,7 @@ interface FilterStore extends FilterState {
   setDateTo: (date: string | null) => void
   toggleMediaType: (t: MediaType) => void
   setRatingMin: (rating: number) => void
+  setRatingMax: (rating: number) => void
   setColorLabel: (label: ColorLabel | null) => void
   setFlag: (flag: FlagFilter) => void
   setHeapId: (id: string | null) => void
@@ -93,6 +95,7 @@ export const INITIAL_FILTERS: FilterState = {
   dateTo: null,
   mediaTypes: [],
   ratingMin: 0,
+  ratingMax: 0,
   colorLabel: null,
   flag: 'any',
   heapId: null,
@@ -114,6 +117,7 @@ function snapshotFilters(s: FilterState): FilterState {
     dateTo: s.dateTo,
     mediaTypes: [...s.mediaTypes],
     ratingMin: s.ratingMin,
+    ratingMax: s.ratingMax,
     colorLabel: s.colorLabel,
     flag: s.flag,
     heapId: s.heapId,
@@ -142,6 +146,7 @@ export const useFilterStore = create<FilterStore>((set) => ({
         : [...s.mediaTypes, t],
     })),
   setRatingMin: (ratingMin) => set({ ratingMin }),
+  setRatingMax: (ratingMax) => set({ ratingMax }),
   setColorLabel: (colorLabel) => set({ colorLabel }),
   setFlag: (flag) => set({ flag }),
   setHeapId: (heapId) => set({ heapId }),
@@ -205,6 +210,7 @@ export function filtersToParams(f: FilterState): Record<string, string | number>
   if (f.dateTo) params.date_to = f.dateTo
   if (f.mediaTypes.length > 0) params.media_type = f.mediaTypes.join(',')
   if (f.ratingMin > 0) params.rating_min = f.ratingMin
+  if (f.ratingMax > 0) params.rating_max = f.ratingMax
   if (f.colorLabel) params.color_label = f.colorLabel
   if (f.flag === 'discarded') params.is_discarded = 'true'
   if (f.heapId) params.heap_id = f.heapId
@@ -224,6 +230,7 @@ export function hasActiveFilters(f: FilterState): boolean {
     f.dateTo !== null ||
     f.mediaTypes.length > 0 ||
     f.ratingMin > 0 ||
+    f.ratingMax > 0 ||
     f.colorLabel !== null ||
     f.flag !== 'any' ||
     f.heapId !== null ||

mulita.yml

@@ -22,3 +22,31 @@ performance:
   cache_ttl: 3600
   db_pool_size: 20
   db_pool_recycle: 3600
+
+# AI vision pipeline — embedding, OCR, object detection, face recognition.
+# Runs on the dedicated `vision` Celery queue (PR4+). Set enabled: false
+# to disable all vision processing.
+vision:
+  enabled: true
+  backend: onnx            # "onnx" (CPU) | "rocm" (future GPU)
+  models_dir: /data/models
+  embedder:
+    name: openclip_vitb32
+    batch_size: 8
+  ocr:
+    enabled: true
+    languages: [en]
+    min_confidence: 0.5
+  detector:
+    enabled: true
+    min_confidence: 0.35
+    max_detections: 50
+  faces:
+    enabled: true
+    min_face_size: 40
+    recognition_threshold: 0.65
+    cluster_eps: 0.5
+  classifier:
+    enabled: true
+    min_confidence: 0.3
+  worker_concurrency: 2