Fix caption cache timing, commit path tracking, and clean up start script

- Move caption cache refresh before UI render so indicators show on load
- Return actual dest paths from commit_batch/commit_global to fix
  caption-on-apply silently failing when files are renamed on collision
- Simplify start.sh to only run NiceGUI (remove Streamlit)
- Add requests to requirements.txt

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CLAUDE.md

@@ -0,0 +1,61 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+Turbo Sorter Pro v12.5 - A dual-interface image organization tool combining Streamlit (admin dashboard) and NiceGUI (gallery interface) for managing large image collections through time-sync matching, ID collision resolution, category-based sorting, and gallery tagging with pairing capabilities.
+
+## Commands
+
+```bash
+# Install dependencies
+pip install -r requirements.txt
+
+# Run Streamlit dashboard (port 8501)
+streamlit run app.py --server.port=8501 --server.address=0.0.0.0
+
+# Run NiceGUI gallery (port 8080)
+python3 gallery_app.py
+
+# Both services (container startup)
+./start.sh
+
+# Syntax check all Python files
+python3 -m py_compile *.py
+```
+
+## Architecture
+
+### Dual-Framework Design
+- **Streamlit (app.py, port 8501)**: Administrative dashboard with 5 modular tabs for management workflows
+- **NiceGUI (gallery_app.py, port 8080)**: Modern gallery interface for image tagging and pairing operations
+- **Shared Backend**: Both UIs use `SorterEngine` (engine.py) and the same SQLite database
+
+### Core Components
+
+| File | Purpose |
+|------|---------|
+| `engine.py` | Static `SorterEngine` class - all DB operations, file handling, image compression |
+| `gallery_app.py` | NiceGUI gallery with `AppState` class for centralized state management |
+| `app.py` | Streamlit entry point, loads tab modules |
+| `tab_*.py` | Independent tab modules for each workflow |
+
+### Database
+SQLite at `/app/sorter_database.db` with tables: profiles, folder_ids, categories, staging_area, processed_log, folder_tags, profile_categories, pairing_settings.
+
+### Tab Workflows
+1. **Time-Sync Discovery** - Match images by timestamp across folders
+2. **ID Review** - Resolve ID collisions between target/control folders
+3. **Unused Archive** - Manage rejected image pairs
+4. **Category Sorter** - One-to-many categorization
+5. **Gallery Staged** - Grid-based tagging with Gallery/Pairing dual modes
+
+## Key Patterns
+
+- **ID Format**: `id001_`, `id002_` (zero-padded 3-digit prefix)
+- **Staging Pattern**: Two-phase commit (stage → commit) with undo support
+- **Image Formats**: .jpg, .jpeg, .png, .webp, .bmp, .tiff
+- **Compression**: WebP with ThreadPoolExecutor (8 workers)
+- **Permissions**: chmod 0o777 applied to committed files
+- **Default Paths**: `/storage` when not configured

engine.py

@@ -1,12 +1,31 @@
 import os
 import shutil
 import sqlite3
+import base64
+import requests
+from datetime import datetime
+from contextlib import contextmanager
 from PIL import Image
 from io import BytesIO
 
 class SorterEngine:
     DB_PATH = "/app/sorter_database.db"
 
+    @staticmethod
+    @contextmanager
+    def get_db():
+        """Context manager for database connections.
+        Ensures proper commit/rollback and always closes connection."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        try:
+            yield conn
+            conn.commit()
+        except Exception:
+            conn.rollback()
+            raise
+        finally:
+            conn.close()
+
     # --- 1. DATABASE INITIALIZATION ---
     @staticmethod
     def init_db():
@@ -27,21 +46,85 @@ class SorterEngine:
         cursor.execute('''CREATE TABLE IF NOT EXISTS processed_log 
             (source_path TEXT PRIMARY KEY, category TEXT, action_type TEXT)''')
         
-        # Seed categories if empty
+        # --- NEW: FOLDER TAGS TABLE (persists tags by folder) ---
+        # Check if old schema exists (without profile column) and migrate
+        cursor.execute("SELECT name FROM sqlite_master WHERE type='table' AND name='folder_tags'")
+        if cursor.fetchone():
+            cursor.execute("PRAGMA table_info(folder_tags)")
+            columns = [row[1] for row in cursor.fetchall()]
+            if 'profile' not in columns:
+                # Migrate: drop old table and recreate with profile column
+                cursor.execute("DROP TABLE folder_tags")
+                conn.commit()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS folder_tags
+            (profile TEXT, folder_path TEXT, filename TEXT, category TEXT, tag_index INTEGER,
+             PRIMARY KEY (profile, folder_path, filename))''')
+        
+        # --- NEW: PROFILE CATEGORIES TABLE (each profile has its own categories) ---
+        cursor.execute('''CREATE TABLE IF NOT EXISTS profile_categories 
+            (profile TEXT, category TEXT, PRIMARY KEY (profile, category))''')
+        
+        # --- NEW: PAIRING SETTINGS TABLE ---
+        cursor.execute('''CREATE TABLE IF NOT EXISTS pairing_settings 
+            (profile TEXT PRIMARY KEY, 
+             adjacent_folder TEXT, 
+             main_category TEXT, 
+             adj_category TEXT, 
+             main_output TEXT, 
+             adj_output TEXT, 
+             time_window INTEGER)''')
+        
+        # Seed categories if empty (legacy table)
         cursor.execute("SELECT COUNT(*) FROM categories")
         if cursor.fetchone()[0] == 0:
-            for cat in ["_TRASH", "Default", "Action", "Solo"]:
+            for cat in ["_TRASH", "control", "Default", "Action", "Solo"]:
                 cursor.execute("INSERT OR IGNORE INTO categories VALUES (?)", (cat,))
 
+        # --- CAPTION TABLES ---
+        # Per-category prompt templates
+        cursor.execute('''CREATE TABLE IF NOT EXISTS category_prompts
+            (profile TEXT, category TEXT, prompt_template TEXT,
+             PRIMARY KEY (profile, category))''')
+
+        # Stored captions
+        cursor.execute('''CREATE TABLE IF NOT EXISTS image_captions
+            (image_path TEXT PRIMARY KEY, caption TEXT, model TEXT,
+             generated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)''')
+
+        # Caption API settings per profile
+        cursor.execute('''CREATE TABLE IF NOT EXISTS caption_settings
+            (profile TEXT PRIMARY KEY,
+             api_endpoint TEXT DEFAULT 'http://localhost:8080/v1/chat/completions',
+             model_name TEXT DEFAULT 'local-model',
+             max_tokens INTEGER DEFAULT 300,
+             temperature REAL DEFAULT 0.7,
+             timeout_seconds INTEGER DEFAULT 60,
+             batch_size INTEGER DEFAULT 4)''')
+
+        # --- PERFORMANCE INDEXES ---
+        # Index for staging_area queries filtered by category
+        cursor.execute("CREATE INDEX IF NOT EXISTS idx_staging_category ON staging_area(target_category)")
+        # Index for folder_tags queries filtered by profile and folder_path
+        cursor.execute("CREATE INDEX IF NOT EXISTS idx_folder_tags_profile ON folder_tags(profile, folder_path)")
+        # Index for profile_categories lookups
+        cursor.execute("CREATE INDEX IF NOT EXISTS idx_profile_categories ON profile_categories(profile)")
+        # Index for caption lookups by image path
+        cursor.execute("CREATE INDEX IF NOT EXISTS idx_image_captions ON image_captions(image_path)")
+
         conn.commit()
         conn.close()
 
     # --- 2. PROFILE & PATH MANAGEMENT ---
     @staticmethod
-    def save_tab_paths(profile_name, t1_t=None, t2_t=None, t2_c=None, t4_s=None, t4_o=None, mode=None, t5_s=None, t5_o=None):
+    def save_tab_paths(profile_name, t1_t=None, t2_t=None, t2_c=None, t4_s=None, t4_o=None, mode=None, t5_s=None, t5_o=None,
+                       pair_adjacent_folder=None, pair_main_category=None, pair_adj_category=None, 
+                       pair_main_output=None, pair_adj_output=None, pair_time_window=None):
         """Updates specific tab paths in the database while preserving others."""
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
+        
+        # Save main profile settings
         cursor.execute("SELECT * FROM profiles WHERE name = ?", (profile_name,))
         row = cursor.fetchone()
         
@@ -61,6 +144,38 @@ class SorterEngine:
             t5_o if t5_o is not None else row[8]
         )
         cursor.execute("INSERT OR REPLACE INTO profiles VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)", new_values)
+        
+        # Save pairing settings if any are provided
+        if any(x is not None for x in [pair_adjacent_folder, pair_main_category, pair_adj_category, 
+                                        pair_main_output, pair_adj_output, pair_time_window]):
+            # Ensure table exists
+            cursor.execute('''CREATE TABLE IF NOT EXISTS pairing_settings 
+                (profile TEXT PRIMARY KEY, 
+                 adjacent_folder TEXT, 
+                 main_category TEXT, 
+                 adj_category TEXT, 
+                 main_output TEXT, 
+                 adj_output TEXT, 
+                 time_window INTEGER)''')
+            
+            # Get existing values
+            cursor.execute("SELECT * FROM pairing_settings WHERE profile = ?", (profile_name,))
+            pair_row = cursor.fetchone()
+            
+            if not pair_row:
+                pair_row = (profile_name, "", "control", "control", "/storage", "/storage", 60)
+            
+            pair_values = (
+                profile_name,
+                pair_adjacent_folder if pair_adjacent_folder is not None else pair_row[1],
+                pair_main_category if pair_main_category is not None else pair_row[2],
+                pair_adj_category if pair_adj_category is not None else pair_row[3],
+                pair_main_output if pair_main_output is not None else pair_row[4],
+                pair_adj_output if pair_adj_output is not None else pair_row[5],
+                pair_time_window if pair_time_window is not None else pair_row[6]
+            )
+            cursor.execute("INSERT OR REPLACE INTO pairing_settings VALUES (?, ?, ?, ?, ?, ?, ?)", pair_values)
+        
         conn.commit()
         conn.close()
     @staticmethod
@@ -91,33 +206,90 @@ class SorterEngine:
 
     @staticmethod
     def load_profiles():
-        """Loads all workspace presets."""
+        """Loads all workspace presets including pairing settings.
+        Uses LEFT JOIN to fetch all data in a single query (fixes N+1 problem)."""
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
-        cursor.execute("SELECT * FROM profiles")
+
+        # Ensure pairing_settings table exists before JOIN
+        cursor.execute('''CREATE TABLE IF NOT EXISTS pairing_settings
+            (profile TEXT PRIMARY KEY,
+             adjacent_folder TEXT,
+             main_category TEXT,
+             adj_category TEXT,
+             main_output TEXT,
+             adj_output TEXT,
+             time_window INTEGER)''')
+
+        # Single query with LEFT JOIN - eliminates N+1 queries
+        cursor.execute('''
+            SELECT p.name, p.tab1_target, p.tab2_target, p.tab2_control,
+                   p.tab4_source, p.tab4_out, p.mode, p.tab5_source, p.tab5_out,
+                   ps.adjacent_folder, ps.main_category, ps.adj_category,
+                   ps.main_output, ps.adj_output, ps.time_window
+            FROM profiles p
+            LEFT JOIN pairing_settings ps ON p.name = ps.profile
+        ''')
         rows = cursor.fetchall()
-        conn.close()
-        return {r[0]: {
+
+        profiles = {}
+        for r in rows:
+            profile_name = r[0]
+            profiles[profile_name] = {
                 "tab1_target": r[1], "tab2_target": r[2], "tab2_control": r[3],
                 "tab4_source": r[4], "tab4_out": r[5], "mode": r[6],
-            "tab5_source": r[7], "tab5_out": r[8]
-        } for r in rows}
+                "tab5_source": r[7], "tab5_out": r[8],
+                # Pairing settings from JOIN (with defaults for NULL)
+                "pair_adjacent_folder": r[9] or "",
+                "pair_main_category": r[10] or "control",
+                "pair_adj_category": r[11] or "control",
+                "pair_main_output": r[12] or "/storage",
+                "pair_adj_output": r[13] or "/storage",
+                "pair_time_window": r[14] or 60
+            }
 
-    # --- 3. CATEGORY MANAGEMENT (Sorted A-Z) ---
+        conn.close()
+        return profiles
+
+    # --- 3. CATEGORY MANAGEMENT (Profile-based) ---
     @staticmethod
-    def get_categories():
+    def get_categories(profile=None):
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
+        
+        # Ensure table exists
+        cursor.execute('''CREATE TABLE IF NOT EXISTS profile_categories 
+            (profile TEXT, category TEXT, PRIMARY KEY (profile, category))''')
+        
+        if profile:
+            cursor.execute("SELECT category FROM profile_categories WHERE profile = ? ORDER BY category COLLATE NOCASE ASC", (profile,))
+            cats = [r[0] for r in cursor.fetchall()]
+            # If no categories for this profile, seed with defaults
+            if not cats:
+                for cat in ["_TRASH", "control"]:
+                    cursor.execute("INSERT OR IGNORE INTO profile_categories VALUES (?, ?)", (profile, cat))
+                conn.commit()
+                cats = ["_TRASH", "control"]
+        else:
+            # Fallback to legacy table
             cursor.execute("SELECT name FROM categories ORDER BY name COLLATE NOCASE ASC")
             cats = [r[0] for r in cursor.fetchall()]
+        
         conn.close()
         return cats
 
     @staticmethod
-    def add_category(name):
+    def add_category(name, profile=None):
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
+        
+        if profile:
+            cursor.execute('''CREATE TABLE IF NOT EXISTS profile_categories 
+                (profile TEXT, category TEXT, PRIMARY KEY (profile, category))''')
+            cursor.execute("INSERT OR IGNORE INTO profile_categories VALUES (?, ?)", (profile, name))
+        else:
             cursor.execute("INSERT OR IGNORE INTO categories VALUES (?)", (name,))
+        
         conn.commit()
         conn.close()
 
@@ -153,20 +325,42 @@ class SorterEngine:
 
     # --- 4. IMAGE & ID OPERATIONS ---
     @staticmethod
-    def get_images(path, recursive=False):
-        """Image scanner with optional recursive subfolder support."""
+    def get_images(path, recursive=False, exclude_paths=None):
+        """Image scanner with optional recursive subfolder support.
+        
+        Args:
+            path: Directory to scan
+            recursive: Whether to scan subdirectories
+            exclude_paths: List of paths to exclude from scanning
+        """
         exts = ('.jpg', '.jpeg', '.png', '.webp', '.bmp', '.tiff')
         if not path or not os.path.exists(path): return []
+        
+        exclude_paths = exclude_paths or []
+        # Normalize exclude paths
+        exclude_normalized = [os.path.normpath(os.path.abspath(p)) for p in exclude_paths]
+        
         image_list = []
         if recursive:
-            for root, _, files in os.walk(path):
+            for root, dirs, files in os.walk(path):
                 # Skip the trash folder from scanning
                 if "_DELETED" in root: continue
+                
+                # Skip excluded paths
+                root_normalized = os.path.normpath(os.path.abspath(root))
+                if any(root_normalized.startswith(exc) or exc.startswith(root_normalized) for exc in exclude_normalized):
+                    # Remove excluded dirs from dirs to prevent descending into them
+                    dirs[:] = [d for d in dirs if os.path.normpath(os.path.abspath(os.path.join(root, d))) not in exclude_normalized]
+                    if root_normalized in exclude_normalized:
+                        continue
+                
                 for f in files:
-                    if f.lower().endswith(exts): image_list.append(os.path.join(root, f))
+                    if f.lower().endswith(exts): 
+                        image_list.append(os.path.join(root, f))
         else:
             for f in os.listdir(path):
-                if f.lower().endswith(exts): image_list.append(os.path.join(path, f))
+                if f.lower().endswith(exts): 
+                    image_list.append(os.path.join(path, f))
         return sorted(image_list)
 
     @staticmethod
@@ -226,36 +420,45 @@ class SorterEngine:
     @staticmethod
     def stage_image(original_path, category, new_name):
         """Records a pending rename/move in the database."""
-        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        with SorterEngine.get_db() as conn:
             cursor = conn.cursor()
             cursor.execute("INSERT OR REPLACE INTO staging_area VALUES (?, ?, ?, 1)", (original_path, category, new_name))
-        conn.commit()
-        conn.close()
 
     @staticmethod
     def clear_staged_item(original_path):
         """Removes an item from the pending staging area."""
-        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        with SorterEngine.get_db() as conn:
             cursor = conn.cursor()
             cursor.execute("DELETE FROM staging_area WHERE original_path = ?", (original_path,))
-        conn.commit()
-        conn.close()
+
+    @staticmethod
+    def clear_staging_area():
+        """Clears all items from the staging area."""
+        with SorterEngine.get_db() as conn:
+            cursor = conn.cursor()
+            cursor.execute("DELETE FROM staging_area")
 
     @staticmethod
     def get_staged_data():
         """Retrieves current tagged/staged images."""
-        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        with SorterEngine.get_db() as conn:
             cursor = conn.cursor()
             cursor.execute("SELECT * FROM staging_area")
             rows = cursor.fetchall()
-        conn.close()
             # FIXED: Added "marked": r[3] to the dictionary
             return {r[0]: {"cat": r[1], "name": r[2], "marked": r[3]} for r in rows}
         
     @staticmethod
-    def commit_global(output_root, cleanup_mode, operation="Copy", source_root=None):
-        """Commits ALL staged files and fixes permissions."""
+    def commit_global(output_root, cleanup_mode, operation="Copy", source_root=None, profile=None):
+        """Commits ALL staged files and fixes permissions.
+        Returns dict mapping original_path -> {dest, cat} for committed files."""
         data = SorterEngine.get_staged_data()
+        committed = {}
+
+        # Save folder tags BEFORE processing (so we can restore them later)
+        if source_root:
+            SorterEngine.save_folder_tags(source_root, profile)
+
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
 
@@ -281,6 +484,9 @@ class SorterEngine:
                 # --- FIX PERMISSIONS ---
                 SorterEngine.fix_permissions(final_dst)
 
+                # Track actual destination
+                committed[old_p] = {"dest": final_dst, "cat": info['cat']}
+
                 # Log History
                 cursor.execute("INSERT OR REPLACE INTO processed_log VALUES (?, ?, ?)",
                                (old_p, info['cat'], operation))
@@ -304,6 +510,7 @@ class SorterEngine:
         cursor.execute("DELETE FROM staging_area")
         conn.commit()
         conn.close()
+        return committed
 
     # --- 6. CORE UTILITIES (SYNC & UNDO) ---
     @staticmethod
@@ -415,10 +622,12 @@ class SorterEngine:
 
     @staticmethod
     def commit_batch(file_list, output_root, cleanup_mode, operation="Copy"):
-        """Commits files and fixes permissions."""
+        """Commits files and fixes permissions.
+        Returns dict mapping original_path -> actual_dest_path for committed files."""
         data = SorterEngine.get_staged_data()
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
+        committed = {}
 
         if not os.path.exists(output_root): os.makedirs(output_root, exist_ok=True)
 
@@ -447,6 +656,9 @@ class SorterEngine:
                 # --- FIX PERMISSIONS ---
                 SorterEngine.fix_permissions(final_dst)
 
+                # Track actual destination
+                committed[file_path] = {"dest": final_dst, "cat": info['cat']}
+
                 # Update DB
                 cursor.execute("DELETE FROM staging_area WHERE original_path = ?", (file_path,))
                 cursor.execute("INSERT OR REPLACE INTO processed_log VALUES (?, ?, ?)",
@@ -467,6 +679,7 @@ class SorterEngine:
 
         conn.commit()
         conn.close()
+        return committed
 
     @staticmethod
     def rename_category(old_name, new_name):
@@ -494,11 +707,553 @@ class SorterEngine:
             conn.close()
 
     @staticmethod
-    def delete_category(name):
+    def delete_category(name, profile=None):
         """Deletes a category and clears any staged tags associated with it."""
         conn = sqlite3.connect(SorterEngine.DB_PATH)
         cursor = conn.cursor()
+        
+        if profile:
+            cursor.execute("DELETE FROM profile_categories WHERE profile = ? AND category = ?", (profile, name))
+        else:
             cursor.execute("DELETE FROM categories WHERE name = ?", (name,))
+        
         cursor.execute("DELETE FROM staging_area WHERE target_category = ?", (name,))
         conn.commit()
         conn.close()
+
+    # In engine.py / SorterEngine class
+    @staticmethod
+    def get_tagged_page_indices(all_images, page_size):
+        staged = SorterEngine.get_staged_data()
+        if not staged: return set()
+        tagged_pages = set()
+        staged_keys = set(staged.keys())
+        for idx, img_path in enumerate(all_images):
+            if img_path in staged_keys:
+                tagged_pages.add(idx // page_size)
+        return tagged_pages
+
+    # --- 7. FOLDER TAG PERSISTENCE ---
+    @staticmethod
+    def save_folder_tags(folder_path, profile=None):
+        """
+        Saves current staging data associated with a folder for later restoration.
+        Call this BEFORE clearing the staging area.
+        """
+        import re
+        staged = SorterEngine.get_staged_data()
+        if not staged:
+            return 0
+        
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+        
+        # Ensure table exists with profile column
+        cursor.execute('''CREATE TABLE IF NOT EXISTS folder_tags 
+            (profile TEXT, folder_path TEXT, filename TEXT, category TEXT, tag_index INTEGER,
+             PRIMARY KEY (profile, folder_path, filename))''')
+        
+        # Check if old schema (without profile) - migrate if needed
+        cursor.execute("PRAGMA table_info(folder_tags)")
+        columns = [row[1] for row in cursor.fetchall()]
+        if 'profile' not in columns:
+            cursor.execute("DROP TABLE folder_tags")
+            cursor.execute('''CREATE TABLE folder_tags 
+                (profile TEXT, folder_path TEXT, filename TEXT, category TEXT, tag_index INTEGER,
+                 PRIMARY KEY (profile, folder_path, filename))''')
+            conn.commit()
+        
+        profile = profile or "Default"
+        saved_count = 0
+        for orig_path, info in staged.items():
+            # Only save tags for files that are in this folder (or subfolders)
+            if orig_path.startswith(folder_path):
+                filename = os.path.basename(orig_path)
+                category = info['cat']
+                
+                # Extract index from the new_name (e.g., "Action_042.jpg" -> 42)
+                new_name = info['name']
+                match = re.search(r'_(\d+)', new_name)
+                tag_index = int(match.group(1)) if match else 0
+                
+                cursor.execute(
+                    "INSERT OR REPLACE INTO folder_tags VALUES (?, ?, ?, ?, ?)",
+                    (profile, folder_path, filename, category, tag_index)
+                )
+                saved_count += 1
+        
+        conn.commit()
+        conn.close()
+        return saved_count
+
+    @staticmethod
+    def restore_folder_tags(folder_path, all_images, profile=None):
+        """
+        Restores previously saved tags for a folder back into the staging area.
+        Call this when loading/reloading a folder.
+        Returns the number of tags restored.
+        """
+        try:
+            conn = sqlite3.connect(SorterEngine.DB_PATH)
+            cursor = conn.cursor()
+            
+            # Ensure table exists with profile column
+            cursor.execute('''CREATE TABLE IF NOT EXISTS folder_tags 
+                (profile TEXT, folder_path TEXT, filename TEXT, category TEXT, tag_index INTEGER,
+                 PRIMARY KEY (profile, folder_path, filename))''')
+            
+            # Check if old schema (without profile) - migrate if needed
+            cursor.execute("PRAGMA table_info(folder_tags)")
+            columns = [row[1] for row in cursor.fetchall()]
+            if 'profile' not in columns:
+                cursor.execute("DROP TABLE folder_tags")
+                cursor.execute('''CREATE TABLE folder_tags 
+                    (profile TEXT, folder_path TEXT, filename TEXT, category TEXT, tag_index INTEGER,
+                     PRIMARY KEY (profile, folder_path, filename))''')
+                conn.commit()
+            
+            profile = profile or "Default"
+            
+            # Get saved tags for this folder and profile
+            cursor.execute(
+                "SELECT filename, category, tag_index FROM folder_tags WHERE profile = ? AND folder_path = ?",
+                (profile, folder_path)
+            )
+            saved_tags = {row[0]: {"cat": row[1], "index": row[2]} for row in cursor.fetchall()}
+            
+            if not saved_tags:
+                conn.close()
+                return 0
+            
+            # Build a map of filename -> full path from current images
+            filename_to_path = {}
+            for img_path in all_images:
+                fname = os.path.basename(img_path)
+                if fname not in filename_to_path:
+                    filename_to_path[fname] = img_path
+            
+            # Restore tags to staging area
+            restored = 0
+            for filename, tag_info in saved_tags.items():
+                if filename in filename_to_path:
+                    full_path = filename_to_path[filename]
+                    cursor.execute("SELECT 1 FROM staging_area WHERE original_path = ?", (full_path,))
+                    if not cursor.fetchone():
+                        ext = os.path.splitext(filename)[1]
+                        new_name = f"{tag_info['cat']}_{tag_info['index']:03d}{ext}"
+                        cursor.execute(
+                            "INSERT OR REPLACE INTO staging_area VALUES (?, ?, ?, 1)",
+                            (full_path, tag_info['cat'], new_name)
+                        )
+                        restored += 1
+            
+            conn.commit()
+            conn.close()
+            return restored
+        except Exception as e:
+            print(f"Error restoring folder tags: {e}")
+            return 0
+
+    @staticmethod
+    def clear_folder_tags(folder_path):
+        """Clears saved tags for a specific folder."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+        cursor.execute("DELETE FROM folder_tags WHERE folder_path = ?", (folder_path,))
+        conn.commit()
+        conn.close()
+
+    @staticmethod
+    def get_saved_folder_tags(folder_path):
+        """Returns saved tags for a folder (for debugging/display)."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+        cursor.execute(
+            "SELECT filename, category, tag_index FROM folder_tags WHERE folder_path = ?",
+            (folder_path,)
+        )
+        result = {row[0]: {"cat": row[1], "index": row[2]} for row in cursor.fetchall()}
+        conn.close()
+        return result
+
+    # --- 8. CAPTION SETTINGS & PROMPTS ---
+    @staticmethod
+    def get_caption_settings(profile):
+        """Get caption API settings for a profile."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        # Ensure table exists
+        cursor.execute('''CREATE TABLE IF NOT EXISTS caption_settings
+            (profile TEXT PRIMARY KEY,
+             api_endpoint TEXT DEFAULT 'http://localhost:8080/v1/chat/completions',
+             model_name TEXT DEFAULT 'local-model',
+             max_tokens INTEGER DEFAULT 300,
+             temperature REAL DEFAULT 0.7,
+             timeout_seconds INTEGER DEFAULT 60,
+             batch_size INTEGER DEFAULT 4)''')
+
+        cursor.execute("SELECT * FROM caption_settings WHERE profile = ?", (profile,))
+        row = cursor.fetchone()
+        conn.close()
+
+        if row:
+            return {
+                "profile": row[0],
+                "api_endpoint": row[1],
+                "model_name": row[2],
+                "max_tokens": row[3],
+                "temperature": row[4],
+                "timeout_seconds": row[5],
+                "batch_size": row[6]
+            }
+        else:
+            # Return defaults
+            return {
+                "profile": profile,
+                "api_endpoint": "http://localhost:8080/v1/chat/completions",
+                "model_name": "local-model",
+                "max_tokens": 300,
+                "temperature": 0.7,
+                "timeout_seconds": 60,
+                "batch_size": 4
+            }
+
+    @staticmethod
+    def save_caption_settings(profile, api_endpoint=None, model_name=None, max_tokens=None,
+                              temperature=None, timeout_seconds=None, batch_size=None):
+        """Save caption API settings for a profile."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        # Ensure table exists
+        cursor.execute('''CREATE TABLE IF NOT EXISTS caption_settings
+            (profile TEXT PRIMARY KEY,
+             api_endpoint TEXT DEFAULT 'http://localhost:8080/v1/chat/completions',
+             model_name TEXT DEFAULT 'local-model',
+             max_tokens INTEGER DEFAULT 300,
+             temperature REAL DEFAULT 0.7,
+             timeout_seconds INTEGER DEFAULT 60,
+             batch_size INTEGER DEFAULT 4)''')
+
+        # Get existing values
+        cursor.execute("SELECT * FROM caption_settings WHERE profile = ?", (profile,))
+        row = cursor.fetchone()
+
+        if not row:
+            row = (profile, "http://localhost:8080/v1/chat/completions", "local-model", 300, 0.7, 60, 4)
+
+        new_values = (
+            profile,
+            api_endpoint if api_endpoint is not None else row[1],
+            model_name if model_name is not None else row[2],
+            max_tokens if max_tokens is not None else row[3],
+            temperature if temperature is not None else row[4],
+            timeout_seconds if timeout_seconds is not None else row[5],
+            batch_size if batch_size is not None else row[6]
+        )
+
+        cursor.execute("INSERT OR REPLACE INTO caption_settings VALUES (?, ?, ?, ?, ?, ?, ?)", new_values)
+        conn.commit()
+        conn.close()
+
+    @staticmethod
+    def get_category_prompt(profile, category):
+        """Get prompt template for a category."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS category_prompts
+            (profile TEXT, category TEXT, prompt_template TEXT,
+             PRIMARY KEY (profile, category))''')
+
+        cursor.execute(
+            "SELECT prompt_template FROM category_prompts WHERE profile = ? AND category = ?",
+            (profile, category)
+        )
+        row = cursor.fetchone()
+        conn.close()
+
+        if row and row[0]:
+            return row[0]
+        else:
+            # Default prompt
+            return "Describe this image in detail for training purposes. Include subjects, actions, setting, colors, and composition."
+
+    @staticmethod
+    def save_category_prompt(profile, category, prompt):
+        """Save prompt template for a category."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS category_prompts
+            (profile TEXT, category TEXT, prompt_template TEXT,
+             PRIMARY KEY (profile, category))''')
+
+        cursor.execute(
+            "INSERT OR REPLACE INTO category_prompts VALUES (?, ?, ?)",
+            (profile, category, prompt)
+        )
+        conn.commit()
+        conn.close()
+
+    @staticmethod
+    def get_all_category_prompts(profile):
+        """Get all prompt templates for a profile."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS category_prompts
+            (profile TEXT, category TEXT, prompt_template TEXT,
+             PRIMARY KEY (profile, category))''')
+
+        cursor.execute(
+            "SELECT category, prompt_template FROM category_prompts WHERE profile = ?",
+            (profile,)
+        )
+        result = {row[0]: row[1] for row in cursor.fetchall()}
+        conn.close()
+        return result
+
+    # --- 9. CAPTION STORAGE ---
+    @staticmethod
+    def save_caption(image_path, caption, model):
+        """Save a generated caption to the database."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS image_captions
+            (image_path TEXT PRIMARY KEY, caption TEXT, model TEXT,
+             generated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)''')
+
+        cursor.execute(
+            "INSERT OR REPLACE INTO image_captions VALUES (?, ?, ?, ?)",
+            (image_path, caption, model, datetime.now().isoformat())
+        )
+        conn.commit()
+        conn.close()
+
+    @staticmethod
+    def get_caption(image_path):
+        """Get caption for an image."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS image_captions
+            (image_path TEXT PRIMARY KEY, caption TEXT, model TEXT,
+             generated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)''')
+
+        cursor.execute(
+            "SELECT caption, model, generated_at FROM image_captions WHERE image_path = ?",
+            (image_path,)
+        )
+        row = cursor.fetchone()
+        conn.close()
+
+        if row:
+            return {"caption": row[0], "model": row[1], "generated_at": row[2]}
+        return None
+
+    @staticmethod
+    def get_captions_batch(image_paths):
+        """Get captions for multiple images."""
+        if not image_paths:
+            return {}
+
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS image_captions
+            (image_path TEXT PRIMARY KEY, caption TEXT, model TEXT,
+             generated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)''')
+
+        placeholders = ','.join('?' * len(image_paths))
+        cursor.execute(
+            f"SELECT image_path, caption, model, generated_at FROM image_captions WHERE image_path IN ({placeholders})",
+            image_paths
+        )
+        result = {row[0]: {"caption": row[1], "model": row[2], "generated_at": row[3]} for row in cursor.fetchall()}
+        conn.close()
+        return result
+
+    @staticmethod
+    def delete_caption(image_path):
+        """Delete caption for an image."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+        cursor.execute("DELETE FROM image_captions WHERE image_path = ?", (image_path,))
+        conn.commit()
+        conn.close()
+
+    @staticmethod
+    def get_all_caption_paths():
+        """Get set of all image paths that have captions."""
+        conn = sqlite3.connect(SorterEngine.DB_PATH)
+        cursor = conn.cursor()
+
+        cursor.execute('''CREATE TABLE IF NOT EXISTS image_captions
+            (image_path TEXT PRIMARY KEY, caption TEXT, model TEXT,
+             generated_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP)''')
+
+        cursor.execute("SELECT image_path FROM image_captions")
+        result = {row[0] for row in cursor.fetchall()}
+        conn.close()
+        return result
+
+    # --- 10. VLLM API CAPTIONING ---
+    @staticmethod
+    def caption_image_vllm(image_path, prompt, settings):
+        """
+        Generate caption for an image using VLLM API.
+
+        Args:
+            image_path: Path to the image file
+            prompt: Text prompt for captioning
+            settings: Dict with api_endpoint, model_name, max_tokens, temperature, timeout_seconds
+
+        Returns:
+            Tuple of (caption_text, error_message). If successful, error is None.
+        """
+        try:
+            # Read and encode image
+            with open(image_path, 'rb') as f:
+                img_bytes = f.read()
+            b64_image = base64.b64encode(img_bytes).decode('utf-8')
+
+            # Determine MIME type
+            ext = os.path.splitext(image_path)[1].lower()
+            mime_types = {
+                '.jpg': 'image/jpeg',
+                '.jpeg': 'image/jpeg',
+                '.png': 'image/png',
+                '.webp': 'image/webp',
+                '.bmp': 'image/bmp',
+                '.tiff': 'image/tiff'
+            }
+            mime_type = mime_types.get(ext, 'image/jpeg')
+
+            # Build request payload (OpenAI-compatible format)
+            payload = {
+                "model": settings.get('model_name', 'local-model'),
+                "messages": [{
+                    "role": "user",
+                    "content": [
+                        {"type": "text", "text": prompt},
+                        {"type": "image_url", "image_url": {"url": f"data:{mime_type};base64,{b64_image}"}}
+                    ]
+                }],
+                "max_tokens": settings.get('max_tokens', 300),
+                "temperature": settings.get('temperature', 0.7)
+            }
+
+            # Make API request
+            response = requests.post(
+                settings.get('api_endpoint', 'http://localhost:8080/v1/chat/completions'),
+                json=payload,
+                timeout=settings.get('timeout_seconds', 60)
+            )
+            response.raise_for_status()
+
+            result = response.json()
+            caption = result['choices'][0]['message']['content']
+            return caption.strip(), None
+
+        except requests.Timeout:
+            return None, f"API timeout after {settings.get('timeout_seconds', 60)}s"
+        except requests.RequestException as e:
+            return None, f"API error: {str(e)}"
+        except KeyError as e:
+            return None, f"Invalid API response: missing {str(e)}"
+        except Exception as e:
+            return None, f"Error: {str(e)}"
+
+    @staticmethod
+    def caption_batch_vllm(image_paths, get_prompt_fn, settings, progress_cb=None):
+        """
+        Caption multiple images using VLLM API.
+
+        Args:
+            image_paths: List of (image_path, category) tuples
+            get_prompt_fn: Function(category) -> prompt string
+            settings: Caption settings dict
+            progress_cb: Optional callback(current, total, status_msg) for progress updates
+
+        Returns:
+            Dict with results: {"success": count, "failed": count, "captions": {path: caption}}
+        """
+        results = {"success": 0, "failed": 0, "captions": {}, "errors": {}}
+        total = len(image_paths)
+
+        for i, (image_path, category) in enumerate(image_paths):
+            if progress_cb:
+                progress_cb(i, total, f"Captioning {os.path.basename(image_path)}...")
+
+            prompt = get_prompt_fn(category)
+            caption, error = SorterEngine.caption_image_vllm(image_path, prompt, settings)
+
+            if caption:
+                # Save to database
+                SorterEngine.save_caption(image_path, caption, settings.get('model_name', 'local-model'))
+                results["captions"][image_path] = caption
+                results["success"] += 1
+            else:
+                # Store error
+                error_caption = f"[ERROR] {error}"
+                SorterEngine.save_caption(image_path, error_caption, settings.get('model_name', 'local-model'))
+                results["errors"][image_path] = error
+                results["failed"] += 1
+
+        if progress_cb:
+            progress_cb(total, total, "Complete!")
+
+        return results
+
+    @staticmethod
+    def write_caption_sidecar(image_path, caption):
+        """
+        Write caption to a .txt sidecar file next to the image.
+
+        Args:
+            image_path: Path to the image file
+            caption: Caption text to write
+
+        Returns:
+            Path to sidecar file, or None on error
+        """
+        try:
+            # Create sidecar path (same name, .txt extension)
+            base_path = os.path.splitext(image_path)[0]
+            sidecar_path = f"{base_path}.txt"
+
+            with open(sidecar_path, 'w', encoding='utf-8') as f:
+                f.write(caption)
+
+            # Fix permissions
+            SorterEngine.fix_permissions(sidecar_path)
+
+            return sidecar_path
+        except Exception as e:
+            print(f"Warning: Could not write sidecar for {image_path}: {e}")
+            return None
+
+    @staticmethod
+    def read_caption_sidecar(image_path):
+        """
+        Read caption from a .txt sidecar file if it exists.
+
+        Args:
+            image_path: Path to the image file
+
+        Returns:
+            Caption text or None if no sidecar exists
+        """
+        try:
+            base_path = os.path.splitext(image_path)[0]
+            sidecar_path = f"{base_path}.txt"
+
+            if os.path.exists(sidecar_path):
+                with open(sidecar_path, 'r', encoding='utf-8') as f:
+                    return f.read().strip()
+        except Exception:
+            pass
+        return None

requirements.txt

@@ -1,2 +1,4 @@
 streamlit
 Pillow
+nicegui
+requests

start.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+
+# NiceSorter - Start Script
+# Runs the NiceGUI gallery interface on port 8080
+
+set -e
+
+# Navigate to app directory if running in container
+if [ -d "/app" ]; then
+    cd /app
+fi
+
+# Install dependencies if requirements.txt exists
+if [ -f "requirements.txt" ]; then
+    echo "📦 Installing dependencies..."
+    pip install --no-cache-dir -q -r requirements.txt
+fi
+
+# Initialize database
+echo "🗄️ Initializing database..."
+python3 -c "from engine import SorterEngine; SorterEngine.init_db()"
+
+# Start NiceGUI
+echo "🚀 Starting NiceSorter on http://0.0.0.0:8080"
+exec python3 gallery_app.py

tab_gallery_sorter.py

@@ -1,34 +1,134 @@
 import streamlit as st
 import os
 import math
+import concurrent.futures
+from typing import Dict, Set, List, Optional, Tuple
 from engine import SorterEngine
 
 # ==========================================
-# 1. CACHED DATA LOADER (The Fix)
+# STATE MANAGEMENT
 # ==========================================
-@st.cache_data(show_spinner=False)
-def get_cached_images(path, mutation_id):
-    """
-    Scans the folder ONLY when 'path' or 'mutation_id' changes.
-    Navigating pages does NOT change these, so it remains instant.
-    """
-    return SorterEngine.get_images(path, recursive=True)
 
+class StreamlitState:
+    """Centralized state management with type hints."""
     
-# ==========================================
-# 2. CALLBACKS (Updated with Refresh Logic)
-# ==========================================
-def trigger_refresh():
-    """Increments the mutation counter to force a file re-scan."""
-    if 't5_file_id' not in st.session_state: st.session_state.t5_file_id = 0
+    @staticmethod
+    def init():
+        """Initialize all session state variables."""
+        defaults = {
+            't5_file_id': 0,
+            't5_page': 0,
+            't5_active_cat': 'Default',
+            't5_next_index': 1,
+            't5_op_mode': 'Copy',
+            't5_cleanup_mode': 'Keep',
+            't5_page_size': 24,
+            't5_grid_cols': 4,
+            't5_quality': 50,
+        }
+        for key, value in defaults.items():
+            if key not in st.session_state:
+                st.session_state[key] = value
+    
+    @staticmethod
+    def trigger_refresh():
+        """Force file cache invalidation."""
         st.session_state.t5_file_id += 1
     
-def cb_tag_image(img_path, selected_cat, index_val, path_o):
+    @staticmethod
+    def change_page(delta: int):
+        """Navigate pages by delta."""
+        st.session_state.t5_page = max(0, st.session_state.t5_page + delta)
+    
+    @staticmethod
+    def set_page(page_idx: int):
+        """Jump to specific page."""
+        st.session_state.t5_page = page_idx
+    
+    @staticmethod
+    def slider_change(key: str):
+        """Handle slider-based page navigation (1-based to 0-based)."""
+        st.session_state.t5_page = st.session_state[key] - 1
+
+# ==========================================
+# CACHING & DATA LOADING
+# ==========================================
+
+@st.cache_data(show_spinner=False)
+def get_cached_images(path: str, mutation_id: int) -> List[str]:
+    """Scan folder for images. mutation_id forces refresh."""
+    return SorterEngine.get_images(path, recursive=True)
+
+@st.cache_data(show_spinner=False, max_entries=2000)
+def get_cached_thumbnail(path: str, quality: int, target_size: int, mtime: float) -> Optional[bytes]:
+    """Load and compress thumbnail with caching."""
+    try:
+        return SorterEngine.compress_for_web(path, quality, target_size)
+    except Exception:
+        return None
+
+@st.cache_data(show_spinner=False)
+def get_cached_green_dots(all_images: List[str], page_size: int, staged_keys: frozenset) -> Set[int]:
     """
-    Tags image with manual number. 
-    Handles collisions by creating variants (e.g. _001_1) and warning the user.
+    Calculate which pages have tagged images (cached).
+    Returns set of page indices with staged images.
     """
-    if selected_cat.startswith("---") or selected_cat == "":
+    staged_set = set(staged_keys)
+    tagged_pages = set()
+    
+    for idx, img_path in enumerate(all_images):
+        if img_path in staged_set:
+            tagged_pages.add(idx // page_size)
+    
+    return tagged_pages
+
+@st.cache_data(show_spinner=False)
+def build_index_map(active_cat: str, path_o: str, staged_data_frozen: frozenset) -> Dict[int, str]:
+    """
+    Build mapping of index numbers to file paths for active category.
+    Returns: {1: '/path/to/Cat_001.jpg', 2: '/path/to/Cat_002.jpg', ...}
+    """
+    index_map = {}
+    
+    # Convert frozenset back to dict for processing
+    staged_dict = {k: v for k, v in staged_data_frozen}
+    
+    # Check staging area
+    for orig_path, info in staged_dict.items():
+        if info['cat'] == active_cat:
+            idx = _extract_index(info['name'])
+            if idx is not None:
+                index_map[idx] = orig_path
+    
+    # Check disk
+    cat_path = os.path.join(path_o, active_cat)
+    if os.path.exists(cat_path):
+        for filename in os.listdir(cat_path):
+            if filename.startswith(active_cat):
+                idx = _extract_index(filename)
+                if idx is not None and idx not in index_map:
+                    index_map[idx] = os.path.join(cat_path, filename)
+    
+    return index_map
+
+def _extract_index(filename: str) -> Optional[int]:
+    """Extract numeric index from filename (e.g., 'Cat_042.jpg' -> 42)."""
+    try:
+        parts = filename.rsplit('_', 1)
+        if len(parts) > 1:
+            num_str = parts[1].split('.')[0]
+            return int(num_str)
+    except (ValueError, IndexError):
+        pass
+    return None
+
+# ==========================================
+# ACTIONS
+# ==========================================
+
+def action_tag(img_path: str, selected_cat: str, index_val: int, path_o: str):
+    """Tag image with category and index, handling collisions."""
+    if selected_cat.startswith("---") or not selected_cat:
         st.toast("⚠️ Select a valid category first!", icon="🚫")
         return
     
@@ -36,316 +136,608 @@ def cb_tag_image(img_path, selected_cat, index_val, path_o):
     base_name = f"{selected_cat}_{index_val:03d}"
     new_name = f"{base_name}{ext}"
     
-    # --- COLLISION DETECTION ---
-    # 1. Check Staging DB
+    # Collision detection
     staged = SorterEngine.get_staged_data()
-    # Get all names currently staged for this category
     staged_names = {v['name'] for v in staged.values() if v['cat'] == selected_cat}
-    
-    # 2. Check Hard Drive
     dest_path = os.path.join(path_o, selected_cat, new_name)
     
     collision = False
     suffix = 1
-    
-    # Loop until we find a free name
     while new_name in staged_names or os.path.exists(dest_path):
         collision = True
         new_name = f"{base_name}_{suffix}{ext}"
         dest_path = os.path.join(path_o, selected_cat, new_name)
         suffix += 1
     
-    # --- SAVE ---
     SorterEngine.stage_image(img_path, selected_cat, new_name)
     
     if collision:
-        st.toast(f"⚠️ Conflict! Saved as variant: {new_name}", icon="🔀")
+        st.toast(f"⚠️ Conflict! Saved as: {new_name}", icon="🔀")
     
-    # REMOVED: st.session_state.t5_next_index += 1
-    # The numbers in the input boxes will now stay static.
+    # Auto-increment index
+    st.session_state.t5_next_index = index_val + 1
 
-def cb_untag_image(img_path):
+def action_untag(img_path: str):
+    """Remove staging from image."""
     SorterEngine.clear_staged_item(img_path)
 
-def cb_delete_image(img_path):
+def action_delete(img_path: str):
+    """Delete image to trash."""
     SorterEngine.delete_to_trash(img_path)
-    trigger_refresh() # Force re-scan so the image disappears from the list
+    StreamlitState.trigger_refresh()
 
-def cb_apply_batch(current_batch, path_o, cleanup_mode, operation):
+def action_apply_batch(current_batch: List[str], path_o: str, cleanup_mode: str, operation: str):
+    """Apply staged changes for current page."""
     SorterEngine.commit_batch(current_batch, path_o, cleanup_mode, operation)
-    trigger_refresh() # Force re-scan to remove moved files
+    StreamlitState.trigger_refresh()
 
-def cb_apply_global(path_o, cleanup_mode, operation, path_s):
+def action_apply_global(path_o: str, cleanup_mode: str, operation: str, path_s: str):
+    """Apply all staged changes globally."""
     SorterEngine.commit_global(path_o, cleanup_mode, operation, source_root=path_s)
-    trigger_refresh() # Force re-scan
+    StreamlitState.trigger_refresh()
 
-def cb_change_page(delta):
-    if 't5_page' not in st.session_state: st.session_state.t5_page = 0
-    st.session_state.t5_page += delta
-    # No trigger_refresh() here -> This is why page turning is now instant!
+def action_add_category(name: str):
+    """Add new category."""
+    if name:
+        SorterEngine.add_category(name)
+        st.session_state.t5_active_cat = name
 
-def cb_jump_page(k):
-    val = st.session_state[k]
-    st.session_state.t5_page = val - 1
+def action_rename_category(old_name: str, new_name: str):
+    """Rename category."""
+    if new_name and new_name != old_name:
+        SorterEngine.rename_category(old_name, new_name)
+        st.session_state.t5_active_cat = new_name
 
+def action_delete_category(cat_name: str):
+    """Delete category."""
+    SorterEngine.delete_category(cat_name)
+    # Reset to first available category
+    cats = SorterEngine.get_categories() or ["Default"]
+    st.session_state.t5_active_cat = cats[0]
 
 # ==========================================
-# 3. FRAGMENTS (Sidebar, Grid, Batch)
+# DIALOGS
 # ==========================================
-# ... (Sidebar code remains exactly the same) ...
+
+@st.dialog("🔍 Full Resolution", width="large")
+def view_high_res(img_path: str):
+    """Modal for full resolution inspection."""
+    img_data = SorterEngine.compress_for_web(img_path, quality=90, target_size=None)
+    if img_data:
+        st.image(img_data, use_container_width=True)
+        st.caption(f"📁 {img_path}")
+    else:
+        st.error(f"Could not load: {img_path}")
+
+@st.dialog("🖼️ Tag Preview", width="large")
+def view_tag_preview(img_path: str, title: str):
+    """Show image associated with a numbered tag."""
+    st.subheader(title)
+    
+    img_data = SorterEngine.compress_for_web(img_path, quality=80, target_size=800)
+    if img_data:
+        st.image(img_data, use_container_width=True)
+        st.caption(f"📁 {img_path}")
+    else:
+        st.error(f"Could not load: {img_path}")
+
+# ==========================================
+# UI COMPONENTS
+# ==========================================
+
 @st.fragment
-def render_sidebar_content():
+def render_sidebar_content(path_o: str):
+    """Render category management sidebar."""
     st.divider()
     st.subheader("🏷️ Category Manager")
-    cats = SorterEngine.get_categories()
-    processed_cats = []
+    
+    # Get and process categories with separators
+    cats = SorterEngine.get_categories() or ["Default"]
+    processed_cats = _add_category_separators(cats)
+    
+    # Sync radio selection immediately
+    if "t5_radio_select" in st.session_state:
+        new_selection = st.session_state.t5_radio_select
+        if not new_selection.startswith("---"):
+            st.session_state.t5_active_cat = new_selection
+    
+    if "t5_active_cat" not in st.session_state:
+        st.session_state.t5_active_cat = cats[0]
+    
+    current_cat = st.session_state.t5_active_cat
+    
+    # NUMBER GRID (1-25) with previews
+    if current_cat and not current_cat.startswith("---"):
+        st.caption(f"**{current_cat}** Index Map")
+        
+        # Build index map (cached)
+        staged = SorterEngine.get_staged_data()
+        staged_frozen = frozenset(staged.items())
+        index_map = build_index_map(current_cat, path_o, staged_frozen)
+        
+        # Render 5x5 grid
+        grid_cols = st.columns(5, gap="small")
+        for i in range(1, 26):
+            is_used = i in index_map
+            btn_type = "primary" if is_used else "secondary"
+            
+            with grid_cols[(i - 1) % 5]:
+                if st.button(f"{i}", key=f"grid_{i}", type=btn_type, use_container_width=True):
+                    st.session_state.t5_next_index = i
+                    if is_used:
+                        view_tag_preview(index_map[i], f"{current_cat} #{i}")
+                    else:
+                        st.toast(f"Next index set to #{i}")
+        
+        st.divider()
+    
+    # CATEGORY SELECTOR
+    st.radio("Active Category", processed_cats, key="t5_radio_select")
+    
+    # INDEX CONTROLS
+    st.caption("Tagging Settings")
+    c_num1, c_num2 = st.columns([3, 1], vertical_alignment="bottom")
+    
+    c_num1.number_input("Next Index #", min_value=1, step=1, key="t5_next_index")
+    
+    if c_num2.button("🔄", help="Auto-detect next index"):
+        used_indices = list(index_map.keys()) if index_map else []
+        st.session_state.t5_next_index = max(used_indices) + 1 if used_indices else 1
+        st.rerun()
+    
+    st.divider()
+    
+    # CATEGORY MANAGEMENT TABS
+    tab_add, tab_edit = st.tabs(["➕ Add", "✏️ Edit"])
+    
+    with tab_add:
+        c1, c2 = st.columns([3, 1])
+        new_cat = c1.text_input(
+            "New Category", 
+            label_visibility="collapsed", 
+            placeholder="Enter name...",
+            key="t5_new_cat"
+        )
+        if c2.button("Add", key="btn_add_cat"):
+            action_add_category(new_cat)
+            st.rerun()
+    
+    with tab_edit:
+        if current_cat and not current_cat.startswith("---") and current_cat in cats:
+            st.caption(f"Editing: **{current_cat}**")
+            
+            rename_val = st.text_input(
+                "Rename to:",
+                value=current_cat,
+                key=f"ren_{current_cat}"
+            )
+            
+            if st.button("💾 Save", key=f"save_{current_cat}", use_container_width=True):
+                action_rename_category(current_cat, rename_val)
+                st.rerun()
+            
+            st.markdown("---")
+            
+            if st.button(
+                "🗑️ Delete Category",
+                key=f"del_cat_{current_cat}",
+                type="primary",
+                use_container_width=True
+            ):
+                action_delete_category(current_cat)
+                st.rerun()
+
+def _add_category_separators(cats: List[str]) -> List[str]:
+    """Add alphabetical separators between categories."""
+    processed = []
     last_char = ""
-    if cats:
+    
     for cat in cats:
         current_char = cat[0].upper()
         if last_char and current_char != last_char:
-                processed_cats.append(f"--- {current_char} ---")
-            processed_cats.append(cat)
+            processed.append(f"--- {current_char} ---")
+        processed.append(cat)
         last_char = current_char
     
-    if "t5_active_cat" not in st.session_state: st.session_state.t5_active_cat = cats[0] if cats else "Default"
-    current_selection = st.session_state.t5_active_cat
-    if not current_selection.startswith("---") and current_selection not in cats:
-        st.session_state.t5_active_cat = cats[0] if cats else "Default"
+    return processed
 
-    selection = st.radio("Active Tag", processed_cats, key="t5_radio_select")
-    if not selection.startswith("---"): st.session_state.t5_active_cat = selection
+def render_pagination_carousel(key_suffix: str, total_pages: int, current_page: int, tagged_pages: Set[int]):
+    """Render pagination controls with green dot indicators."""
+    if total_pages <= 1:
+        return
     
-    st.divider()
-    tab_add, tab_edit = st.tabs(["➕ Add", "✏️ Edit"])
-    with tab_add:
-        c1, c2 = st.columns([3, 1])
-        new_cat = c1.text_input("New Name", label_visibility="collapsed", placeholder="New...", key="t5_new_cat")
-        if c2.button("Add", key="btn_add_cat"):
-            if new_cat:
-                SorterEngine.add_category(new_cat)
-                st.rerun()
-    with tab_edit:
-        target_cat = st.session_state.t5_active_cat
-        if target_cat and not target_cat.startswith("---") and target_cat in cats:
-            st.caption(f"Editing: **{target_cat}**")
-            rename_val = st.text_input("Rename to:", value=target_cat, key=f"ren_{target_cat}")
-            if st.button("💾 Save", key=f"save_{target_cat}", use_container_width=True):
-                if rename_val and rename_val != target_cat:
-                    SorterEngine.rename_category(target_cat, rename_val)
-                    st.session_state.t5_active_cat = rename_val
-                    st.rerun()
-            st.markdown("---")
-            if st.button("🗑️ Delete", key=f"del_cat_{target_cat}", type="primary", use_container_width=True):
-                SorterEngine.delete_category(target_cat)
-                st.rerun()
-        else:
-            st.info("Select a valid category to edit.")
+    # Rapid navigation slider (1-based)
+    st.slider(
+        "Page Navigator",
+        min_value=1,
+        max_value=total_pages,
+        value=current_page + 1,
+        step=1,
+        key=f"slider_{key_suffix}",
+        label_visibility="collapsed",
+        on_change=StreamlitState.slider_change,
+        args=(f"slider_{key_suffix}",)
+    )
     
-@st.dialog("🔍 High-Res Inspection", width="large")
-def view_high_res(img_path):
-    """
-    Opens a modal and loads the ORIGINAL size image on demand.
-    We still compress to WebP (q=90) to ensure it sends fast, 
-    but we do NOT resize the dimensions.
-    """
-    # Load with target_size=None to keep original dimensions
-    # Quality=90 for high fidelity
-    img_data = SorterEngine.compress_for_web(img_path, quality=90, target_size=None)
+    # Calculate button window (show current ±2 pages)
+    window_radius = 2
+    start_p = max(0, current_page - window_radius)
+    end_p = min(total_pages, current_page + window_radius + 1)
     
-    if img_data:
-        st.image(img_data, use_container_width=True)
-        st.caption(f"Filename: {os.path.basename(img_path)}")
-    else:
-        st.error("Could not load full resolution image.")
+    # Adjust near edges to maintain consistent width
+    if current_page < window_radius:
+        end_p = min(total_pages, 5)
+    elif current_page > total_pages - window_radius - 1:
+        start_p = max(0, total_pages - 5)
     
-# ... (Gallery Grid code remains exactly the same) ...
-# --- UPDATED CACHE FUNCTION ---
-@st.cache_data(show_spinner=False, max_entries=2000)
-def get_cached_thumbnail(path, quality, target_size, mtime):
-    # We pass the dynamic target_size here
-    return SorterEngine.compress_for_web(path, quality, target_size)
+    num_buttons = end_p - start_p
+    if num_buttons < 1:
+        start_p = 0
+        end_p = total_pages
+        num_buttons = total_pages
+    
+    # Render button row: [Prev] [1] [2] [3] ... [Next]
+    cols = st.columns([1] + [1] * num_buttons + [1])
+    
+    # Previous button
+    with cols[0]:
+        st.button(
+            "◀",
+            disabled=(current_page == 0),
+            on_click=StreamlitState.change_page,
+            args=(-1,),
+            key=f"prev_{key_suffix}",
+            use_container_width=True
+        )
+    
+    # Page number buttons
+    for i, p_idx in enumerate(range(start_p, end_p)):
+        with cols[i + 1]:
+            label = str(p_idx + 1)
+            if p_idx in tagged_pages:
+                label += " 🟢"
+            
+            btn_type = "primary" if p_idx == current_page else "secondary"
+            
+            st.button(
+                label,
+                type=btn_type,
+                key=f"btn_p{p_idx}_{key_suffix}",
+                use_container_width=True,
+                on_click=StreamlitState.set_page,
+                args=(p_idx,)
+            )
+    
+    # Next button
+    with cols[-1]:
+        st.button(
+            "▶",
+            disabled=(current_page >= total_pages - 1),
+            on_click=StreamlitState.change_page,
+            args=(1,),
+            key=f"next_{key_suffix}",
+            use_container_width=True
+        )
 
-# --- UPDATED GALLERY FRAGMENT ---
 @st.fragment
-def render_gallery_grid(current_batch, quality, grid_cols, path_o): # <--- 1. Added path_o
+def render_gallery_grid(
+    current_batch: List[str],
+    quality: int,
+    grid_cols: int,
+    path_o: str
+):
+    """Render image gallery grid with parallel loading."""
     staged = SorterEngine.get_staged_data()
     history = SorterEngine.get_processed_log()
-    selected_cat = st.session_state.get("t5_active_cat", "Default")
+    selected_cat = st.session_state.t5_active_cat
     tagging_disabled = selected_cat.startswith("---")
     
-    # 2. Ensure global counter exists (default to 1)
-    if "t5_next_index" not in st.session_state: st.session_state.t5_next_index = 1
-
-    # 3. Smart Resolution (Wide screen assumption)
     target_size = int(2400 / grid_cols)
     
-    # 4. Parallel Load (16 threads for WebP)
-    import concurrent.futures
+    # Parallel thumbnail loading
+    batch_cache = _load_thumbnails_parallel(current_batch, quality, target_size)
+    
+    # Render grid
+    cols = st.columns(grid_cols)
+    
+    for idx, img_path in enumerate(current_batch):
+        with cols[idx % grid_cols]:
+            _render_image_card(
+                img_path=img_path,
+                batch_cache=batch_cache,
+                staged=staged,
+                history=history,
+                selected_cat=selected_cat,
+                tagging_disabled=tagging_disabled,
+                path_o=path_o
+            )
+
+def _load_thumbnails_parallel(
+    batch: List[str],
+    quality: int,
+    target_size: int
+) -> Dict[str, Optional[bytes]]:
+    """Load thumbnails in parallel using ThreadPoolExecutor."""
     batch_cache = {}
     
-    def fetch_one(p):
+    def fetch_one(path: str) -> Tuple[str, Optional[bytes]]:
         try:
-            mtime = os.path.getmtime(p)
-            return p, get_cached_thumbnail(p, quality, target_size, mtime)
-        except:
-            return p, None
+            mtime = os.path.getmtime(path)
+            data = get_cached_thumbnail(path, quality, target_size, mtime)
+            return path, data
+        except Exception:
+            return path, None
     
     with concurrent.futures.ThreadPoolExecutor(max_workers=16) as executor:
-        future_to_path = {executor.submit(fetch_one, p): p for p in current_batch}
-        for future in concurrent.futures.as_completed(future_to_path):
-            p, data = future.result()
-            batch_cache[p] = data
+        futures = {executor.submit(fetch_one, p): p for p in batch}
+        for future in concurrent.futures.as_completed(futures):
+            path, data = future.result()
+            batch_cache[path] = data
     
-    # 5. Render Grid
-    cols = st.columns(grid_cols)
-    for idx, img_path in enumerate(current_batch):
+    return batch_cache
+
+def _render_image_card(
+    img_path: str,
+    batch_cache: Dict[str, Optional[bytes]],
+    staged: Dict,
+    history: Dict,
+    selected_cat: str,
+    tagging_disabled: bool,
+    path_o: str
+):
+    """Render individual image card."""
     unique_key = f"frag_{os.path.basename(img_path)}"
-        with cols[idx % grid_cols]:
     is_staged = img_path in staged
     is_processed = img_path in history
     
     with st.container(border=True):
-                # Header: [Name] [Zoom] [Delete]
+        # Header: filename + zoom + delete
         c_name, c_zoom, c_del = st.columns([4, 1, 1])
-                c_name.caption(os.path.basename(img_path)[:10])
+        c_name.caption(os.path.basename(img_path)[:15])
         
         if c_zoom.button("🔍", key=f"zoom_{unique_key}"):
             view_high_res(img_path)
         
-                c_del.button("❌", key=f"del_{unique_key}", on_click=cb_delete_image, args=(img_path,))
+        c_del.button(
+            "❌",
+            key=f"del_{unique_key}",
+            on_click=action_delete,
+            args=(img_path,)
+        )
         
-                # Status Banners
+        # Status indicator
         if is_staged:
-                    st.success(f"🏷️ {staged[img_path]['cat']}")
+            staged_info = staged[img_path]
+            idx = _extract_index(staged_info['name'])
+            idx_str = f" #{idx}" if idx else ""
+            st.success(f"🏷️ {staged_info['cat']}{idx_str}")
         elif is_processed:
             st.info(f"✅ {history[img_path]['action']}")
         
-                # Image (Cached)
+        # Thumbnail
         img_data = batch_cache.get(img_path)
         if img_data:
             st.image(img_data, use_container_width=True)
+        else:
+            st.error("Failed to load")
         
-                # Action Area
+        # Action buttons
         if not is_staged:
-                    # 6. Split Row: [Idx Input] [Tag Button]
             c_idx, c_tag = st.columns([1, 2], vertical_alignment="bottom")
             
-                    # Manual Override Box (Defaults to global session value)
             card_index = c_idx.number_input(
-                        "Idx", 
-                        min_value=1, step=1, 
+                "Index",
+                min_value=1,
+                step=1,
                 value=st.session_state.t5_next_index,
                 label_visibility="collapsed",
                 key=f"idx_{unique_key}"
             )
             
-                    # Tag Button (Passes path_o for conflict check)
             c_tag.button(
                 "Tag",
                 key=f"tag_{unique_key}",
                 disabled=tagging_disabled,
                 use_container_width=True,
-                        on_click=cb_tag_image, 
-                        # Passing card_index + path_o is vital here
+                on_click=action_tag,
                 args=(img_path, selected_cat, card_index, path_o)
             )
         else:
-                    st.button("Untag", key=f"untag_{unique_key}", use_container_width=True,
-                              on_click=cb_untag_image, args=(img_path,))
+            # Show untag with index number
+            staged_name = staged[img_path]['name']
+            idx = _extract_index(staged_name)
+            untag_label = f"Untag (#{idx})" if idx else "Untag"
+            
+            st.button(
+                untag_label,
+                key=f"untag_{unique_key}",
+                use_container_width=True,
+                on_click=action_untag,
+                args=(img_path,)
+            )
 
-# ... (Batch Actions code remains exactly the same) ...
 @st.fragment
-def render_batch_actions(current_batch, path_o, page_num, path_s):
-    st.write(f"### 🚀 Processing Actions")
-    st.caption("Settings apply to both Page and Global actions.")
+def render_batch_actions(
+    current_batch: List[str],
+    path_o: str,
+    page_num: int,
+    path_s: str
+):
+    """Render batch processing controls."""
+    st.write("### 🚀 Processing Actions")
+    st.caption("Settings apply to both Page and Global actions")
     
     c_set1, c_set2 = st.columns(2)
     
-    # CHANGED: "Copy" is now first, making it the default
-    op_mode = c_set1.radio("Tagged Files:", ["Copy", "Move"], horizontal=True, key="t5_op_mode")
+    c_set1.radio(
+        "Tagged Files:",
+        ["Copy", "Move"],
+        horizontal=True,
+        key="t5_op_mode"
+    )
     
-    cleanup = c_set2.radio("Untagged Files:", ["Keep", "Move to Unused", "Delete"], horizontal=True, key="t5_cleanup_mode")
+    c_set2.radio(
+        "Untagged Files:",
+        ["Keep", "Move to Unused", "Delete"],
+        horizontal=True,
+        key="t5_cleanup_mode"
+    )
     
     st.divider()
     
     c_btn1, c_btn2 = st.columns(2)
     
-    # BUTTON 1: APPLY PAGE
-    if c_btn1.button(f"APPLY PAGE {page_num}", type="secondary", use_container_width=True,
-                     on_click=cb_apply_batch, args=(current_batch, path_o, cleanup, op_mode)):
-        st.toast(f"Page {page_num} Applied!")
+    # Apply Page button
+    if c_btn1.button(
+        f"APPLY PAGE {page_num}",
+        type="secondary",
+        use_container_width=True,
+        on_click=action_apply_batch,
+        args=(
+            current_batch,
+            path_o,
+            st.session_state.t5_cleanup_mode,
+            st.session_state.t5_op_mode
+        )
+    ):
+        st.toast(f"Page {page_num} applied!")
         st.rerun()
     
-    # BUTTON 2: APPLY GLOBAL
-    if c_btn2.button("APPLY ALL (GLOBAL)", type="primary", use_container_width=True,
-                     help="Process ALL tagged files across all pages.",
-                     on_click=cb_apply_global, args=(path_o, cleanup, op_mode, path_s)):
-        st.toast("Global Apply Complete!")
+    # Apply Global button
+    if c_btn2.button(
+        "APPLY ALL (GLOBAL)",
+        type="primary",
+        use_container_width=True,
+        help="Process ALL tagged files",
+        on_click=action_apply_global,
+        args=(
+            path_o,
+            st.session_state.t5_cleanup_mode,
+            st.session_state.t5_op_mode,
+            path_s
+        )
+    ):
+        st.toast("Global apply complete!")
         st.rerun()
 
 # ==========================================
-# 4. MAIN RENDERER
+# MAIN RENDER FUNCTION
 # ==========================================
-def render(quality, profile_name):
+
+def render(quality: int, profile_name: str):
+    """Main render function for Streamlit app."""
     st.subheader("🖼️ Gallery Staging Sorter")
     
-    # Init Mutation ID (This triggers the scanner cache refresh)
-    if 't5_file_id' not in st.session_state: st.session_state.t5_file_id = 0
-    if 't5_page' not in st.session_state: st.session_state.t5_page = 0
+    # Initialize state
+    StreamlitState.init()
     
+    # Load profiles and paths
     profiles = SorterEngine.load_profiles()
     p_data = profiles.get(profile_name, {})
-    c1, c2 = st.columns(2)
-    path_s = c1.text_input("Source Folder", value=p_data.get("tab5_source", "/storage"), key="t5_s")
-    path_o = c2.text_input("Output Folder", value=p_data.get("tab5_out", "/storage"), key="t5_o")
     
-    if path_s != p_data.get("tab5_source") or path_o != p_data.get("tab5_out"):
-        if st.button("💾 Save Settings"):
+    c1, c2, c3 = st.columns([3, 3, 1])
+    
+    path_s = c1.text_input(
+        "Source Folder",
+        value=p_data.get("tab5_source", "/storage"),
+        key="t5_s"
+    )
+    
+    path_o = c2.text_input(
+        "Output Folder",
+        value=p_data.get("tab5_out", "/storage"),
+        key="t5_o"
+    )
+    
+    # Save settings button
+    if c3.button("💾 Save", use_container_width=True):
         SorterEngine.save_tab_paths(profile_name, t5_s=path_s, t5_o=path_o)
-            # Saving settings might mean new folder, so we trigger refresh
-            trigger_refresh()
+        StreamlitState.trigger_refresh()
+        st.toast("Settings saved!")
         st.rerun()
     
-    if not os.path.exists(path_s): return
+    # Validate source path
+    if not os.path.exists(path_s):
+        st.warning("⚠️ Source path does not exist")
+        return
     
+    # Render sidebar
     with st.sidebar:
-        render_sidebar_content()
+        render_sidebar_content(path_o)
     
-    with st.expander("👀 View Settings"):
-        c_v1, c_v2 = st.columns(2)
-        page_size = c_v1.slider("Images per Page", 12, 100, 24, 4)
-        grid_cols = c_v2.slider("Grid Columns", 2, 8, 4)
+    # View settings
+    with st.expander("👀 View Settings", expanded=False):
+        c_v1, c_v2, c_v3 = st.columns(3)
         
-    # --- USING CACHED LOADER ---
-    # We pass the mutation ID. If ID is same as last run, scan is SKIPPED.
+        st.session_state.t5_page_size = c_v1.slider(
+            "Images/Page",
+            12, 100,
+            st.session_state.t5_page_size,
+            4
+        )
+        
+        st.session_state.t5_grid_cols = c_v2.slider(
+            "Grid Columns",
+            2, 8,
+            st.session_state.t5_grid_cols
+        )
+        
+        st.session_state.t5_quality = c_v3.slider(
+            "Preview Quality",
+            10, 100,
+            st.session_state.t5_quality,
+            10
+        )
+    
+    # Load images (cached)
     all_images = get_cached_images(path_s, st.session_state.t5_file_id)
     
     if not all_images:
-        st.info("No images found.")
+        st.info("📂 No images found in source folder")
         return
     
-    total_items = len(all_images)
-    total_pages = math.ceil(total_items / page_size)
-    if st.session_state.t5_page >= total_pages: st.session_state.t5_page = max(0, total_pages - 1)
-    if st.session_state.t5_page < 0: st.session_state.t5_page = 0
+    # Pagination calculations
+    page_size = st.session_state.t5_page_size
+    total_pages = math.ceil(len(all_images) / page_size)
     
-    start_idx = st.session_state.t5_page * page_size
-    end_idx = start_idx + page_size
-    current_batch = all_images[start_idx:end_idx]
+    # Bounds checking
+    if st.session_state.t5_page >= total_pages:
+        st.session_state.t5_page = max(0, total_pages - 1)
+    if st.session_state.t5_page < 0:
+        st.session_state.t5_page = 0
     
-    def nav_controls(key_suffix):
-        c1, c2, c3, c4 = st.columns([1.5, 1, 0.5, 1.5], vertical_alignment="center")
-        c1.button("⬅️ Prev", disabled=(st.session_state.t5_page == 0), on_click=cb_change_page, args=(-1,), key=f"p_{key_suffix}", use_container_width=True)
-        c2.number_input("Page", min_value=1, max_value=total_pages, value=st.session_state.t5_page + 1, step=1, label_visibility="collapsed", key=f"jump_{key_suffix}", on_change=cb_jump_page, args=(f"jump_{key_suffix}",))
-        c3.markdown(f"<div style='text-align: left; font-weight: bold;'>/ {total_pages}</div>", unsafe_allow_html=True)
-        c4.button("Next ➡️", disabled=(st.session_state.t5_page >= total_pages - 1), on_click=cb_change_page, args=(1,), key=f"n_{key_suffix}", use_container_width=True)
+    current_page = st.session_state.t5_page
+    start_idx = current_page * page_size
+    current_batch = all_images[start_idx : start_idx + page_size]
     
-    st.divider()
-    nav_controls("top")
-    render_gallery_grid(current_batch, quality, grid_cols, path_o)
-    st.divider()
-    nav_controls("bottom")
+    # Calculate green dots (cached)
+    staged = SorterEngine.get_staged_data()
+    green_dots = get_cached_green_dots(
+        all_images,
+        page_size,
+        frozenset(staged.keys())
+    )
+    
+    # Render UI components
     st.divider()
     
-    render_batch_actions(current_batch, path_o, st.session_state.t5_page + 1, path_s)
+    # Top pagination
+    render_pagination_carousel("top", total_pages, current_page, green_dots)
+    
+    # Gallery grid
+    render_gallery_grid(
+        current_batch,
+        st.session_state.t5_quality,
+        st.session_state.t5_grid_cols,
+        path_o
+    )
+    
+    st.divider()
+    
+    # Bottom pagination
+    render_pagination_carousel("bot", total_pages, current_page, green_dots)
+    
+    st.divider()
+    
+    # Batch actions
+    render_batch_actions(current_batch, path_o, current_page + 1, path_s)