Update listings

Co-authored-by: ivuorinen <11024+ivuorinen@users.noreply.github.com>

.claude/settings.json

@@ -0,0 +1,71 @@
+{
+  "hooks": {
+    "PreToolUse": [
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == *README.md || \"$file\" == *index.html ]]; then echo 'BLOCKED: README.md and index.html are generated artifacts. Edit create_listing.py instead, then run /regen-listings.' >&2; exit 2; fi"
+          }
+        ]
+      },
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == */uv.lock || \"$file\" == uv.lock ]]; then echo 'BLOCKED: uv.lock is auto-generated. Modify pyproject.toml and run uv sync instead.' >&2; exit 2; fi"
+          }
+        ]
+      },
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == *.github/workflows/*.yml || \"$file\" == .github/workflows/*.yml ]]; then echo 'BLOCKED: CI workflows use pinned action SHAs with # version comments for security. Edit workflow files carefully and maintain the SHA-pinning convention.' >&2; exit 2; fi"
+          }
+        ]
+      },
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == */renovate.json || \"$file\" == renovate.json ]]; then echo 'BLOCKED: renovate.json is rarely edited. Make changes deliberately and confirm with the user first.' >&2; exit 2; fi"
+          }
+        ]
+      },
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == emoji/* || \"$file\" == */emoji/* ]]; then echo 'BLOCKED: Emoji image files should not be written by Claude. Manage images manually or use /dedup-check.' >&2; exit 2; fi"
+          }
+        ]
+      }
+    ],
+    "PostToolUse": [
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == *.py ]]; then uv run ruff check --fix \"$file\" 2>&1 | tail -5 && uv run ruff format \"$file\" 2>&1 | tail -3; fi"
+          }
+        ]
+      },
+      {
+        "matcher": "Edit|Write",
+        "hooks": [
+          {
+            "type": "command",
+            "command": "file=$(jq -r '.tool_input.file_path // empty'); if [[ \"$file\" == *.py ]]; then uv run pytest --tb=short -q 2>&1 | tail -5; fi"
+          }
+        ]
+      }
+    ]
+  }
+}

.claude/skills/dedup-check/SKILL.md

@@ -0,0 +1,10 @@
+---
+name: dedup-check
+description: Run dedup in dry-run mode and report duplicate groups found
+disable-model-invocation: true
+---
+
+## Steps
+
+1. Run `uv run dedup --dry-run` from the project root
+2. Summarize the output — report how many duplicate groups were found and which files are involved

.claude/skills/regen-listings/SKILL.md

@@ -0,0 +1,11 @@
+---
+name: regen-listings
+description: Regenerate README.md and index.html from emoji/ contents and verify output
+disable-model-invocation: true
+---
+
+## Steps
+
+1. Run `uv run python3 create_listing.py` from the project root
+2. Confirm both `README.md` and `index.html` exist and are non-empty
+3. Report the file sizes of both generated files

.github/workflows/compress-images.yml

@@ -11,7 +11,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout Repo
-        uses: actions/checkout@v2
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
       - name: Compress Images
         id: calibre
         uses: calibreapp/image-actions@main
@@ -20,7 +20,7 @@ jobs:
           compressOnly: true
       - name: Create New Pull Request If Needed
         if: steps.calibre.outputs.markdown != ''
-        uses: peter-evans/create-pull-request@v3
+        uses: peter-evans/create-pull-request@c0f553fe549906ede9cf27b5156039d195d2ece0 # v8
         with:
           title: Compressed Images Nightly
           branch-suffix: timestamp

.github/workflows/generate-listings.yml

@@ -0,0 +1,30 @@
+name: Generate Listings
+on:
+  push:
+    paths:
+      - 'emoji/**'
+      - 'create_listing.py'
+    branches:
+      - master
+jobs:
+  generate:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout Repo
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
+
+      - name: Set up Python
+        uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
+        with:
+          python-version: '3.14'
+
+      - name: Generate listings
+        run: python3 create_listing.py
+
+      - name: Commit changes
+        run: |
+          git config --local user.email "github-actions[bot]@users.noreply.github.com"
+          git config --local user.name "github-actions[bot]"
+          git add README.md index.html
+          git diff --staged --quiet || git commit -m "Update listings"
+          git push

.github/workflows/pr-compress-images.yml

@@ -16,7 +16,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - name: Checkout Repo
-        uses: actions/checkout@v2
+        uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
 
       - name: Compress Images
         uses: calibreapp/image-actions@main

.github/workflows/test.yml

@@ -0,0 +1,25 @@
+name: Tests
+
+on:
+  push:
+    paths:
+      - '**.py'
+      - 'pyproject.toml'
+      - 'uv.lock'
+  pull_request:
+    paths:
+      - '**.py'
+      - 'pyproject.toml'
+      - 'uv.lock'
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6
+      - uses: astral-sh/setup-uv@5a095e7a2014a4212f075830d4f7277575a9d098 # v7
+      - uses: actions/setup-python@a309ff8b426b58ec0e2a45f0f869d46889d02405 # v6
+        with:
+          python-version: '3.14'
+      - run: uv sync --dev
+      - run: uv run pytest -v

.gitignore

@@ -0,0 +1,2 @@
+__pycache__/
+.claude/settings.local.json

CLAUDE.md

@@ -0,0 +1,52 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+Personal emoji/emote collection for chat apps (Slack, Discord, etc.). Contains 3000+ custom emoji images in `emoji/` with Python tooling for maintenance: listing generation and perceptual deduplication.
+
+## Commands
+
+```bash
+# Install dependencies (uses uv package manager)
+uv sync
+
+# Regenerate README.md and index.html from emoji/ contents
+uv run python3 create_listing.py
+
+# Find duplicate emojis (dry run)
+uv run python3 dedup.py --dry-run
+
+# Find duplicates with custom threshold (0=exact match, default)
+uv run python3 dedup.py --threshold 5 --dry-run
+
+# Actually remove duplicates
+uv run python3 dedup.py --dir emoji/
+
+# Or via uv entry point
+uv run dedup --dry-run
+
+# Run tests
+uv run pytest
+
+# Run tests with verbose output
+uv run pytest -v
+```
+
+## Architecture
+
+Two standalone Python scripts, no shared modules:
+
+- **`create_listing.py`** — Generates `README.md` (HTML tables) and `index.html` (searchable dark-theme SPA) from all images in `emoji/`. No dependencies beyond stdlib. Both output files are auto-generated and committed by CI on push.
+
+- **`dedup.py`** — Finds and removes duplicate images using multi-algorithm perceptual hashing (pHash, aHash, dHash, colorHash). Uses Union-Find clustering. Animated GIFs get extra frame-by-frame verification including timing. Keeps alphabetically-first filename per duplicate group.
+
+## Key Conventions
+
+- Python >=3.11 required; dependencies managed via `uv` with `uv.lock`
+- Image formats: `.png`, `.gif`, `.jpg`, `.jpeg`
+- `README.md` and `index.html` are generated artifacts — edit the scripts, not the outputs
+- CI uses pinned action SHAs (not tags) for security
+- Dependency updates managed by Renovate bot
+- Always use `uv run` to execute Python commands (e.g. `uv run pytest`, `uv run ruff`, `uv run python3 script.py`) to ensure the correct virtualenv and dependencies are used

_create-listing.php

@@ -1,64 +0,0 @@
-<?php
-
-$output        = 'README.md';
-$per_row       = 5;
-$files         = glob( 'emoji/*.{png,gif,jpg,jpeg}', GLOB_BRACE );
-$listing       = [];
-$per_row_width = floor( 100 / $per_row ) . '%';
-
-sort( $files );
-
-if ( count( $files ) < 1 ) {
-    die( 'No images to continue with.' );
-}
-
-function get_basename( string $file ) {
-    $parts = explode( DIRECTORY_SEPARATOR, $file );
-    return end( $parts );
-}
-
-foreach ( $files as $file ) {
-    $first = get_basename( $file );
-    $first = str_replace( 'emoji/', '', $first );
-    $first = trim( $first[0] );
-
-    if ( preg_match( '/([^a-zA-Z:])/', $first ) ) {
-        $first = '\[^a-zA-Z:\]';
-    }
-
-    if ( ! array_key_exists( $first, $listing ) ) {
-        $listing[ $first ] = [];
-    }
-
-    $listing[ $first ][] = $file;
-}
-
-$contents = "# Emotes\n\n";
-
-foreach ( $listing as $header => $icons ) {
-    $contents .= sprintf( "## %s\n\n", $header );
-
-    $chunks = array_chunk( $icons, $per_row );
-
-    $contents .= '<table style="text-align: center;width: 100%">' . "\n";
-
-    foreach ( $chunks as $chunk_icons ) {
-        $contents .= "<tr>\n";
-
-        foreach ( $chunk_icons as $icon ) {
-            $file = $icon;
-            [ $name, $ext ] = explode( '.', get_basename($icon), 2 );
-
-            $format   = '<td style=\'width: %s\'><img width=\'30\' src="%2$s" alt="%2$s"><br><kbd>:%3$s:</kbd></td>';
-            $contents .= sprintf( $format, $per_row_width, $file, $name ) . "\n";
-        }
-
-        $contents .= "</tr>\n";
-    }
-
-    $contents .= "</table>\n\n";
-}
-
-$contents .= "\n\n Generated: " . date('c');
-
-file_put_contents( $output, $contents );

create_listing.py

@@ -0,0 +1,228 @@
+#!/usr/bin/env python3
+"""Generate README.md and index.html with emoji listings."""
+
+import html
+import re
+from collections import defaultdict
+from datetime import datetime, timezone
+from pathlib import Path
+from urllib.parse import quote
+
+PER_ROW = 10
+EMOJI_DIR = Path("emoji")
+EXTENSIONS = (".png", ".gif", ".jpg", ".jpeg")
+
+
+def generate_readme(files: list[Path]) -> None:
+    """Generate README.md with HTML tables of all emoji images."""
+    listing = defaultdict(list)
+    for file in files:
+        first_char = file.name[0].lower()
+        if not re.match(r"[a-z]", first_char):
+            first_char = r"\[^a-zA-Z:\]"
+        listing[first_char].append(file)
+
+    per_row_width = f"{100 // PER_ROW}%"
+    contents = "# Emotes\n\n"
+
+    for header in sorted(listing.keys(), key=lambda x: (not x.startswith("\\"), x)):
+        icons = listing[header]
+        contents += f"## {header}\n\n"
+        contents += '<table style="text-align: center;width: 100%">\n'
+
+        for i in range(0, len(icons), PER_ROW):
+            chunk = icons[i : i + PER_ROW]
+            contents += "<tr>\n"
+
+            for icon in chunk:
+                name = icon.stem
+                encoded_path = f"emoji/{quote(icon.name)}"
+                display_path = f"emoji/{icon.name}"
+
+                contents += (
+                    f"<td style='width: {per_row_width}'>"
+                    f"<img width='30' src=\"{encoded_path}\" "
+                    f'alt="{display_path}" title=":{name}:"></td>\n'
+                )
+
+            contents += "</tr>\n"
+
+        contents += "</table>\n\n"
+
+    contents += f"\n\n Generated: {datetime.now(timezone.utc).isoformat()}"
+
+    Path("README.md").write_text(contents, encoding="utf-8")
+    print(f"Generated README.md with {len(files)} emojis")
+
+
+def generate_html(files: list[Path]) -> None:
+    """Generate index.html with searchable emoji grid grouped alphabetically."""
+    # Group files by first character
+    listing = defaultdict(list)
+    for file in files:
+        first_char = file.name[0].lower()
+        if not re.match(r"[a-z]", first_char):
+            first_char = "#"
+        listing[first_char].append(file)
+
+    # Build grouped HTML
+    sections = []
+    for header in sorted(listing.keys(), key=lambda x: (x != "#", x)):
+        display_header = "0-9 / Special" if header == "#" else header.upper()
+        emoji_items = []
+        for file in listing[header]:
+            name = file.stem
+            encoded_path = f"emoji/{quote(file.name)}"
+            escaped_name = html.escape(name)
+            emoji_items.append(
+                f'      <div class="emoji" data-keyword="{escaped_name}">'
+                f'<img src="{encoded_path}" alt="{escaped_name}" title=":{escaped_name}:"></div>'
+            )
+        sections.append(
+            f'    <section data-group="{html.escape(header)}">\n'
+            f"      <h2>{display_header}</h2>\n"
+            f'      <div class="grid">\n{chr(10).join(emoji_items)}\n      </div>\n'
+            f"    </section>"
+        )
+
+    contents = f"""<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8">
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  <title>Emotes</title>
+  <style>
+    * {{ box-sizing: border-box; }}
+    body {{
+      font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif;
+      margin: 0;
+      padding: 20px;
+      background: #1a1a1a;
+      color: #fff;
+    }}
+    #search {{
+      width: 100%;
+      max-width: 400px;
+      padding: 12px 16px;
+      font-size: 16px;
+      border: 2px solid #333;
+      border-radius: 8px;
+      background: #2a2a2a;
+      color: #fff;
+      margin-bottom: 20px;
+    }}
+    #search:focus {{
+      outline: none;
+      border-color: #666;
+    }}
+    #search::placeholder {{
+      color: #888;
+    }}
+    section {{
+      margin-bottom: 24px;
+    }}
+    section.hidden {{
+      display: none;
+    }}
+    h2 {{
+      font-size: 18px;
+      font-weight: 600;
+      margin: 0 0 12px 0;
+      color: #ccc;
+    }}
+    .grid {{
+      display: grid;
+      grid-template-columns: repeat(auto-fill, minmax(50px, 1fr));
+      gap: 8px;
+    }}
+    .emoji {{
+      display: flex;
+      align-items: center;
+      justify-content: center;
+      padding: 8px;
+      background: #2a2a2a;
+      border-radius: 6px;
+      transition: background 0.15s;
+    }}
+    .emoji:hover {{
+      background: #3a3a3a;
+    }}
+    .emoji img {{
+      width: 32px;
+      height: 32px;
+      object-fit: contain;
+    }}
+    .emoji.hidden {{
+      display: none;
+    }}
+    #count {{
+      color: #888;
+      font-size: 14px;
+      margin-bottom: 12px;
+    }}
+    h1 {{
+      margin: 0 0 20px 0;
+      font-size: 24px;
+    }}
+    h1 a {{
+      color: #fff;
+      text-decoration: none;
+    }}
+    h1 a:hover {{
+      text-decoration: underline;
+    }}
+  </style>
+</head>
+<body>
+  <h1><a href="https://github.com/ivuorinen/emoji">ivuorinen/emoji</a></h1>
+  <input type="text" id="search" placeholder="Search emojis..." autofocus>
+  <div id="count">{len(files)} emojis</div>
+  <div id="content">
+{chr(10).join(sections)}
+  </div>
+  <script>
+    let timeout;
+    const search = document.getElementById('search');
+    const emojis = document.querySelectorAll('.emoji');
+    const sections = document.querySelectorAll('section');
+    const count = document.getElementById('count');
+    const total = emojis.length;
+
+    search.addEventListener('input', function(e) {{
+      clearTimeout(timeout);
+      timeout = setTimeout(() => {{
+        const query = e.target.value.toLowerCase();
+        let visible = 0;
+        emojis.forEach(el => {{
+          const match = el.dataset.keyword.toLowerCase().includes(query);
+          el.classList.toggle('hidden', !match);
+          if (match) visible++;
+        }});
+        sections.forEach(sec => {{
+          const hasVisible = sec.querySelector('.emoji:not(.hidden)');
+          sec.classList.toggle('hidden', !hasVisible);
+        }});
+        count.textContent = query ? visible + ' of ' + total + ' emojis' : total + ' emojis';
+      }}, 150);
+    }});
+  </script>
+</body>
+</html>
+"""
+
+    Path("index.html").write_text(contents, encoding="utf-8")
+    print(f"Generated index.html with {len(files)} emojis")
+
+
+def main():
+    files = sorted(f for f in EMOJI_DIR.iterdir() if f.suffix.lower() in EXTENSIONS)
+
+    if not files:
+        raise SystemExit("No images to continue with.")
+
+    generate_readme(files)
+    generate_html(files)
+
+
+if __name__ == "__main__":
+    main()

dedup.py

@@ -0,0 +1,374 @@
+#!/usr/bin/env python3
+"""Find and remove duplicate emoji files using perceptual hashing."""
+
+import argparse
+import hashlib
+from pathlib import Path
+from dataclasses import dataclass
+
+import imagehash
+from PIL import Image
+
+EXTENSIONS = (".png", ".gif", ".jpg", ".jpeg")
+
+# Number of hash algorithms that must agree for images to be considered similar
+MIN_HASH_AGREEMENT = 4
+
+# Maximum file size difference ratio for duplicates (e.g., 0.05 = 5% difference allowed)
+MAX_SIZE_DIFF_RATIO = 0.02
+
+
+@dataclass
+class ImageInfo:
+    """Container for image metadata and hashes."""
+
+    phash: imagehash.ImageHash
+    ahash: imagehash.ImageHash
+    dhash: imagehash.ImageHash
+    colorhash: imagehash.ImageHash
+    width: int
+    height: int
+    n_frames: int  # 1 for static images
+    md5: str  # File content hash for exact duplicate detection
+
+    def _has_degenerate_hash(self) -> bool:
+        """Check if this image has degenerate (all-zero) hashes, indicating mostly transparent content."""
+        zero_hash = "0000000000000000"
+        # If 3+ hashes are all zeros, the image is likely mostly transparent
+        zero_count = sum(1 for h in [str(self.phash), str(self.ahash), str(self.dhash)] if h == zero_hash)
+        return zero_count >= 3
+
+    def is_candidate(self, other: "ImageInfo", threshold: int) -> tuple[bool, int, int]:
+        """
+        Check if two images are candidate duplicates based on metadata and hashes.
+        Returns (is_candidate, agreements, total_distance).
+
+        This is a fast pre-filter. GIFs require additional frame verification.
+        """
+        # Dimensions must match exactly
+        if self.width != other.width or self.height != other.height:
+            return False, 0, 999
+
+        # Frame count must match for animated images
+        if self.n_frames != other.n_frames:
+            return False, 0, 999
+
+        # Calculate perceptual hash distances
+        distances = [
+            self.phash - other.phash,
+            self.ahash - other.ahash,
+            self.dhash - other.dhash,
+            self.colorhash - other.colorhash,
+        ]
+        total_distance = sum(distances)
+        agreements = sum(1 for d in distances if d <= threshold)
+
+        # For static images: detect re-compressed/re-exported duplicates
+        # Require identical structure AND color, with small perceptual variance:
+        # - aHash=0 AND dHash=0 AND colorHash=0 AND pHash <= 10
+        # - OR all 4 hashes match exactly (total_distance = 0)
+        if self.n_frames == 1:
+            phash_dist = self.phash - other.phash
+            ahash_dist = self.ahash - other.ahash
+            dhash_dist = self.dhash - other.dhash
+            chash_dist = self.colorhash - other.colorhash
+            # Identical structure + color, small perceptual variance = re-compressed image
+            if ahash_dist == 0 and dhash_dist == 0 and chash_dist == 0 and phash_dist <= 10:
+                return True, agreements, total_distance
+            # All hashes match exactly
+            if total_distance == 0:
+                return True, agreements, total_distance
+            return False, agreements, total_distance
+
+        # For animated images: require all 4 hashes to agree (will be verified by frame check)
+        return agreements >= MIN_HASH_AGREEMENT, agreements, total_distance
+
+    def is_animated(self) -> bool:
+        """Check if this is an animated image (multiple frames)."""
+        return self.n_frames > 1
+
+
+class UnionFind:
+    """Union-Find data structure for clustering similar images."""
+
+    def __init__(self):
+        self.parent = {}
+
+    def find(self, x):
+        if x not in self.parent:
+            self.parent[x] = x
+        if self.parent[x] != x:
+            self.parent[x] = self.find(self.parent[x])
+        return self.parent[x]
+
+    def union(self, x, y):
+        px, py = self.find(x), self.find(y)
+        if px != py:
+            self.parent[px] = py
+
+
+def _compute_hashes(img: Image.Image) -> tuple[imagehash.ImageHash, ...]:
+    """Compute all hash types for a single image/frame."""
+    # Convert to RGBA to handle transparency consistently
+    if img.mode != "RGBA":
+        img = img.convert("RGBA")
+    return (
+        imagehash.phash(img),
+        imagehash.average_hash(img),
+        imagehash.dhash(img),
+        imagehash.colorhash(img),
+    )
+
+
+def _compute_md5(path: Path) -> str:
+    """Compute MD5 hash of file contents."""
+    md5 = hashlib.md5()
+    with open(path, "rb") as f:
+        for chunk in iter(lambda: f.read(8192), b""):
+            md5.update(chunk)
+    return md5.hexdigest()
+
+
+def _get_gif_frame_info(path: Path) -> list[tuple[str, int]] | None:
+    """
+    Get perceptual hash and duration for each frame of a GIF.
+    Returns list of (hash_string, duration_ms) tuples, or None if not a multi-frame image.
+    """
+    try:
+        with Image.open(path) as img:
+            n_frames = getattr(img, "n_frames", 1)
+            if n_frames <= 1:
+                return None
+
+            frame_info = []
+            for i in range(n_frames):
+                img.seek(i)
+                frame = img.copy()
+                if frame.mode != "RGBA":
+                    frame = frame.convert("RGBA")
+                duration = img.info.get("duration", 0)
+                frame_info.append((str(imagehash.phash(frame)), duration))
+            return frame_info
+    except Exception:
+        return None
+
+
+def _gifs_are_identical(path1: Path, path2: Path) -> bool:
+    """
+    Compare two GIFs frame-by-frame to check if they have identical content AND timing.
+    Returns True only if all frames and durations match.
+    """
+    info1 = _get_gif_frame_info(path1)
+    info2 = _get_gif_frame_info(path2)
+
+    # If either isn't a multi-frame GIF, fall back to MD5 comparison
+    if info1 is None or info2 is None:
+        return _compute_md5(path1) == _compute_md5(path2)
+
+    # Frame counts must match
+    if len(info1) != len(info2):
+        return False
+
+    # All frames AND durations must match
+    return info1 == info2
+
+
+def compute_image_info(path: Path) -> ImageInfo | None:
+    """
+    Compute image metadata and perceptual hashes.
+    For animated GIFs, samples middle frame to avoid blank first-frame issues.
+    Returns None if image can't be processed.
+    """
+    try:
+        md5 = _compute_md5(path)
+
+        with Image.open(path) as img:
+            width, height = img.size
+            n_frames = getattr(img, "n_frames", 1)
+            is_animated = getattr(img, "is_animated", False)
+
+            if not is_animated:
+                hashes = _compute_hashes(img)
+            else:
+                # For animated images, use middle frame for hashing
+                middle_frame = n_frames // 2
+                try:
+                    img.seek(middle_frame)
+                    hashes = _compute_hashes(img.copy())
+                except EOFError:
+                    img.seek(0)
+                    hashes = _compute_hashes(img)
+
+            return ImageInfo(
+                phash=hashes[0],
+                ahash=hashes[1],
+                dhash=hashes[2],
+                colorhash=hashes[3],
+                width=width,
+                height=height,
+                n_frames=n_frames,
+                md5=md5,
+            )
+
+    except Exception as e:
+        print(f"  Warning: Could not process {path.name}: {e}")
+        return None
+
+
+def _files_size_similar(path1: Path, path2: Path) -> bool:
+    """Check if two files have similar sizes (within MAX_SIZE_DIFF_RATIO)."""
+    size1 = path1.stat().st_size
+    size2 = path2.stat().st_size
+    if size1 == 0 or size2 == 0:
+        return size1 == size2
+    ratio = abs(size1 - size2) / max(size1, size2)
+    return ratio <= MAX_SIZE_DIFF_RATIO
+
+
+def _verify_duplicate_pair(path_i: Path, info_i: ImageInfo, path_j: Path, info_j: ImageInfo, threshold: int) -> bool:
+    """
+    Verify if two candidate images are true duplicates.
+    For animated GIFs, compares frames and timing. For static images, perceptual match is sufficient.
+    """
+    # For animated images, verify frame-by-frame including timing
+    if info_i.is_animated() and info_j.is_animated():
+        return _gifs_are_identical(path_i, path_j)
+
+    # For static images, perceptual hash agreement is sufficient
+    # (handles re-compressed/re-exported duplicates with different file sizes)
+    return True
+
+
+def find_similar_groups(files: list[Path], threshold: int) -> list[list[tuple[Path, ImageInfo]]]:
+    """Find groups of similar images using multi-hash consensus and union-find."""
+    # Compute image info for all files
+    images: list[tuple[Path, ImageInfo]] = []
+    for file in files:
+        info = compute_image_info(file)
+        if info is not None:
+            # Skip images with degenerate (all-zero) hashes - they can't be meaningfully compared
+            if not info._has_degenerate_hash():
+                images.append((file, info))
+
+    if not images:
+        return []
+
+    # Use union-find to cluster similar images
+    # First pass: find candidates based on hashes and metadata
+    # Second pass: verify GIFs with frame comparison
+    uf = UnionFind()
+    for i, (path_i, info_i) in enumerate(images):
+        uf.find(i)  # Initialize
+        for j in range(i + 1, len(images)):
+            path_j, info_j = images[j]
+
+            # Check if candidates based on hashes/metadata
+            is_candidate, _, _ = info_i.is_candidate(info_j, threshold)
+            if not is_candidate:
+                continue
+
+            # For animated images, also check file size similarity
+            # (static images may have different compression, so skip size check)
+            if info_i.is_animated() and not _files_size_similar(path_i, path_j):
+                continue
+
+            # Verify: for GIFs, compare frames; for static, already verified by hashes
+            if _verify_duplicate_pair(path_i, info_i, path_j, info_j, threshold):
+                uf.union(i, j)
+
+    # Group by cluster
+    clusters: dict[int, list[tuple[Path, ImageInfo]]] = {}
+    for i, (path, info) in enumerate(images):
+        root = uf.find(i)
+        if root not in clusters:
+            clusters[root] = []
+        clusters[root].append((path, info))
+
+    # Return only groups with duplicates
+    return [group for group in clusters.values() if len(group) > 1]
+
+
+def deduplicate(groups: list[list[tuple[Path, ImageInfo]]], dry_run: bool, threshold: int) -> tuple[int, int]:
+    """Remove duplicates, keeping first alphabetically. Returns (groups, removed)."""
+    total_removed = 0
+
+    for group in groups:
+        # Sort by filename alphabetically
+        sorted_group = sorted(group, key=lambda x: x[0].name.lower())
+        keep_path, keep_info = sorted_group[0]
+        remove = sorted_group[1:]
+
+        # Calculate agreement info for display
+        agreements_info = [keep_info.is_candidate(info, threshold) for _, info in remove]
+
+        frames_str = f", {keep_info.n_frames} frames" if keep_info.is_animated() else ""
+        print(f"\nSimilar group ({len(group)} files, {keep_info.width}x{keep_info.height}{frames_str}):")
+        print(f"  KEEP: {keep_path.name}")
+
+        for (path, info), (_, agreements, total_dist) in zip(remove, agreements_info):
+            action = "WOULD DELETE" if dry_run else "DELETE"
+            print(f"  {action}: {path.name} (agreements: {agreements}/4, dist: {total_dist})")
+            if not dry_run:
+                path.unlink()
+                total_removed += 1
+
+    if dry_run:
+        return len(groups), sum(len(g) - 1 for g in groups)
+    return len(groups), total_removed
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Find and remove duplicate emoji files using perceptual hashing.")
+    parser.add_argument(
+        "--threshold",
+        type=int,
+        default=0,
+        help="Similarity threshold (0=exact, default=0)",
+    )
+    parser.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="Show duplicates without deleting",
+    )
+    parser.add_argument(
+        "--dir",
+        type=Path,
+        default=Path("emoji"),
+        help="Directory to scan (default: emoji/)",
+    )
+    args = parser.parse_args()
+
+    emoji_dir = args.dir
+    if not emoji_dir.exists():
+        print(f"Error: Directory '{emoji_dir}' does not exist.")
+        return
+
+    files = [f for f in emoji_dir.iterdir() if f.suffix.lower() in EXTENSIONS]
+
+    if not files:
+        print(f"No image files found in {emoji_dir}/ folder.")
+        return
+
+    print(f"Scanning {len(files)} files (threshold: {args.threshold})...")
+    if args.dry_run:
+        print("(dry-run mode - no files will be deleted)")
+
+    groups = find_similar_groups(files, args.threshold)
+
+    if not groups:
+        print("\nNo similar images found.")
+        return
+
+    group_count, removed = deduplicate(groups, args.dry_run, args.threshold)
+
+    print("\n--- Summary ---")
+    print(f"Files scanned: {len(files)}")
+    print(f"Similar groups: {group_count}")
+    if args.dry_run:
+        print(f"Files to remove: {removed}")
+    else:
+        print(f"Files removed: {removed}")
+
+
+if __name__ == "__main__":
+    main()