Googlarr

A chaotic and deeply unnecessary tool that applies googly eyes to your Plex posters on a schedule. You were warned.

---

WARNING — THIS PROJECT MAY RUIN YOUR POSTERS

Googlarr uses the Plex API to modify the posters for your Plex media items in place. It tries to save and restore the originals, but:

• It might miss a poster
• It might fail to restore properly
• Plex might cache something weird
• You might regret everything

Use at your own risk. This project is for prank purposes only. I cannot provide support in case of failure.

---

What It Does

Googlarr is a scheduled prank daemon that:

1. Scans your Plex libraries for movies and/or shows
2. Detects faces and eye positions in poster images
3. Applies googly eyes using image overlays
4. Swaps the posters on a cron-based schedule
5. Restores the originals after the prank window

---

Setup

1. Copy config/config.example.yml to config/config.yml and fill in your Plex URL, token, and libraries
2. Build and run via ./googlarr.sh build && ./googlarr.sh run (or docker compose up -d)
3. Pray

---

Example Config

plex:
  url: http://your.plex.server:32400
  token: your_plex_token
  libraries:
    - Movies
    - TV Shows

paths:
  originals_dir: data/originals
  prank_dir: data/prank

database: "data/googlarr.db"

schedule:
  start: "0 9 * * *"    # Apply prank at 9am daily (cron format)
  stop: "0 17 * * *"    # Restore at 5pm daily (cron format)

detection:
  face_detection_confidence: 0.5
  landmark_detection_confidence: 0.5
  max_faces: 10
  scale_by_face_size: true
  face_based_eye_scale: 0.35
  use_same_size_for_both_eyes: true
  debug_draw_faces: false

---

CLI Tools

Check Status

python -m googlarr.status              # Show summary + schedule + failed items
python -m googlarr.status <item_id>    # Show details for specific item

Manual Poster Control

python -m googlarr.apply               # Manually apply prank posters
python -m googlarr.restore             # Manually restore original posters

Retry Failed Items

python -m googlarr.reset               # Reset all failed items back to NEW
python -m googlarr.reset <item_id>     # Reset specific item only

Regenerate Prank (Single Item)

python -m googlarr.regenerate <item_id>  # Download + regenerate prank for one item

---

Web UI

Googlarr includes a lightweight web interface running on port 8721. Access it at:

http://your-host:8721

The web UI starts automatically when you run the daemon and provides:

Status Page

• Current Status: Shows if prank window is ACTIVE or INACTIVE
• Schedule Info: Next apply and restore times
• Statistics: Breakdown of items by status with human-readable labels
• Failed Items: List of failed items with retry counts
• Override Controls:
  • Apply Now - Immediately apply all ready pranks (bypass schedule)
  • Restore Now - Immediately restore all applied pranks (bypass schedule)
  • ↻ Reload Config - Signal daemon to reload config instantly
• Auto-Refresh: Updates every 5 seconds

Library Page

• Browse: Select which library to view from dropdown
• Status Filter: Filter posters by status (Prank Ready, Original, Failed, etc.)
• Poster Grid: Responsive grid with colour-coded status badges on each poster
• Hover Preview: Hover over a pranked poster to see the googly eye version — badge changes to "Viewing Prank"
• Infinite Scroll: Automatically loads more items as you scroll

Features

• ✅ Dark theme, easy on the eyes
• ✅ Fully responsive (desktop, tablet, mobile)
• ✅ Real-time status updates
• ✅ No external dependencies (vanilla JavaScript)
• ✅ Integrates with daemon for instant control

---

Config Validation

Googlarr validates your config on startup and will exit with a clear error message if anything is missing:

Required keys:

• plex.url, plex.token, plex.libraries
• schedule.start, schedule.stop (must be valid cron expressions)
• paths.originals_dir, paths.prank_dir
• database
• detection.face_detection_confidence, detection.landmark_detection_confidence, detection.max_faces

If any required key is missing or invalid, the daemon will not start.

Architecture

State Machine

Each poster goes through this state flow:

NEW → WORKING_DOWNLOAD → ORIGINAL_DOWNLOADED → WORKING_PRANKIFY → PRANK_GENERATED → PRANK_APPLIED
                                                                          ↑
                                                                   (restored to this)

Failed items retry up to 3 times automatically before staying in FAILED state.

Key Components

• main.py: Async daemon with three concurrent tasks (sync, poster_worker, update_posters_task)
• prank.py: Core prank logic (face detection, overlay generation, poster upload)
• db.py: SQLite state management
• config.py: YAML config loader + validation
• detect.py: Face/eye detection using MediaPipe
• overlay.py: Image manipulation (googly eye overlay application)

---

Responsive Daemon

The daemon is responsive to config changes without requiring a restart:

How It Works

• All tasks reload config at each iteration
• Sleep durations capped at 60 seconds (was hours)
• When config file is edited, changes are picked up within 60 seconds
• API signal (if frontend added) triggers instant reload
• Graceful error handling: continues with old config if validation fails

Config Change Examples

Edit config file:

nano config/config.yml        # Change schedule, libraries, etc
# Daemon picks up changes within 60 seconds, no restart needed

From API/Frontend:

from googlarr.main import signal_config_reload
signal_config_reload()  # Daemon recalculates immediately

Responsiveness

| Task | Delay | |------|-------| | Config file changes | ~60 seconds | | API signal | Instant | | New items in Plex | ~60 seconds |

---

Development

Portions of this project were created with AI code generation (ChatGPT and Claude Code). I am a veteran software engineer however, and have code inspected the output and run extensive testing on multiple libraries.

I also use Claude Code to manage git because I am bad at git.

Recent Improvements

• Improved face detection: Fixed coordinate rescaling bug on large images, full 16-point eye contours for more accurate placement
• Web UI: Status dashboard + library browser with status filter on port 8721
• Poster status badges: Colour-coded per-poster status in library view, hover to preview prank
• Unified code paths: Apply/restore logic extracted to reusable functions (no more duplication)
• Config validation: Required keys checked on startup with clear error messages
• CLI tools: Manual control over apply/restore/reset operations
• Enhanced status: Shows schedule info, failed items, and retry counts
• Retry system: Failed items automatically retry up to 3 times
• Responsive daemon: Config changes picked up within 60s without restart
• Override controls: Apply/Restore buttons to bypass schedule from UI

License

MIT. See LICENSE.

Portions of overlay.py and detector.py were adapted from MIT-licensed sources. See LICENSES/ for attribution.