YouTube AI Detector

Crowdsourced AI detection for YouTube. A Chrome extension that lets the community flag AI-generated videos — when 3 people independently report the same video, the button turns orange for everyone.

No account. No tracking. Fully open source.

Install from Chrome Web Store · Landing page

---

How it works

1. Install the extension — one click, no sign-up. A random anonymous UUID is generated locally and never leaves your machine.
2. Flag AI videos — an "AI slop" button appears next to YouTube's like/dislike bar on watch pages. Click to report, click again to undo.
3. Community consensus — when 3 independent users flag the same video, the button turns orange for everyone. Your own reports turn it red immediately (configurable).

What you can configure

Click the extension icon to open settings:

• Button theme — choose from 4 presets (Default, Neon, Stealth, Toxic) or pick custom colors
• Custom colors — set your own flagged/reported colors with a color picker

Developer settings (selfFlag, devMode) live in extension/config.js and aren't exposed in the UI.

---

Project structure

extension/                Chrome extension (Manifest V3, vanilla JS)
├── manifest.json         Extension config
├── config.js             Hardcoded dev settings (selfFlag, devMode)
├── logger.js             Conditional console logging
├── content.js            Injects report button on YouTube watch pages
├── background.js         Service worker — API calls, caching, install ID
├── popup.html/js/css     Settings popup (themes, colors)
├── styles.css            Injected styles for report button + tooltip
└── icons/                Extension icons (SVG source + 16/48/128px PNGs)

backend/                  Rust API server (Axum + SQLx + PostgreSQL)
├── src/
│   ├── main.rs           Axum router, CORS, static files, tracing
│   ├── lib.rs            Library crate (shared between server + CLI)
│   ├── config.rs         Environment-based configuration
│   ├── db.rs             PostgreSQL queries (batch fetch, toggle report)
│   ├── cache.rs          In-memory video cache (DashMap + TTL)
│   ├── handlers.rs       Route handlers (videos, reports, stats)
│   ├── models.rs         Request/response types
│   ├── threshold.rs      AI classification logic (report_count >= 3)
│   ├── rate_limit.rs     Per-install-id token bucket
│   └── bin/
│       └── cli.rs        Dev CLI for adding/removing test reports
├── static/
│   └── index.html        Landing page (served at /)
└── migrations/
    ├── 001_initial.sql   Initial schema (videos + votes)
    └── 002_reports.sql   Migrated from voting to report-only model

---

Setup

Prerequisites

• Rust (stable, >= 1.70)
• PostgreSQL 18 (Docker recommended)
• Chrome or any Chromium-based browser
• cargo-watch (optional, for hot reload in dev)

1. Database

docker run -d \
  --name yab-db \
  -e POSTGRES_USER=yab \
  -e POSTGRES_PASSWORD=yab \
  -e POSTGRES_DB=youtube_ai_blocker \
  -p 5433:5432 \
  --restart unless-stopped \
  postgres:18

Port 5433 is used to avoid conflicts with any local Postgres on 5432.

2. Backend

cd backend
cargo run

Migrations run automatically on startup. The server starts on http://localhost:3000.

The default database URL is postgres://yab:yab@localhost:5433/youtube_ai_blocker — override with the DATABASE_URL env var if needed.

Dev mode (hot reload)

cd backend
cargo watch -x run

When compiled in debug mode (debug_assertions), request/response tracing is logged to the console with method, path, status, and latency.

Environment variables

Variable	Default	Description
DATABASE_URL	postgres://yab:yab@localhost:5433/youtube_ai_blocker	PostgreSQL connection string
PORT	3000	HTTP server port
CACHE_TTL_SECS	0	In-memory cache TTL (0 = disabled)

Dev CLI

A separate binary for testing. Add/remove reports with random user IDs:

cd backend

# Add 5 reports to a video
cargo run --bin yab-cli -- add dQw4w9WgXcQ 5

# Remove 2 reports
cargo run --bin yab-cli -- remove dQw4w9WgXcQ 2

# Check current status
cargo run --bin yab-cli -- status dQw4w9WgXcQ

3. Chrome extension

1. Go to chrome://extensions
2. Enable Developer mode (top right)
3. Click Load unpacked and select the extension/ folder

The extension talks to http://localhost:3000/api by default. Change API_BASE in background.js for production.

---

API

All endpoints are under /api. Reports require an X-Install-Id header (UUID).

GET /api/videos/batch?ids=id1,id2,...

Fetch report data for up to 50 videos at once.

{
  "videos": {
    "dQw4w9WgXcQ": { "report_count": 5, "is_ai": true },
    "abc123": { "report_count": 1, "is_ai": false }
  }
}

GET /api/videos/:video_id

Single video lookup.

{
  "video_id": "dQw4w9WgXcQ",
  "report_count": 5,
  "is_ai": true
}

POST /api/report

Toggle a report. If you already reported, your report is removed. Otherwise, it's added.

curl -X POST http://localhost:3000/api/report \
  -H "Content-Type: application/json" \
  -H "X-Install-Id: 550e8400-e29b-41d4-a716-446655440000" \
  -d '{"video_id": "dQw4w9WgXcQ"}'

{
  "success": true,
  "reported": true,
  "info": { "report_count": 5, "is_ai": true }
}

Rate limit: 20 requests per minute per install ID.

GET /api/stats

Aggregate counts (used by the landing page).

{
  "total_videos": 42,
  "total_reports": 128
}

Docker deployment

The backend uses a multi-stage Dockerfile with cargo-chef for optimized layer caching.

cd backend
docker build -t yab-backend .

How it works

Stage	Purpose	Cached unless...
planner	Extracts a dependency-only recipe from Cargo.toml/Cargo.lock	Always runs (<1s)
builder	cargo chef cook compiles all deps, then cargo build compiles the app	cook invalidates on dep changes; build runs on every source change (~7s)
runtime	debian:bookworm-slim with just the binary, migrations, and static files	Binary changes

On a source-only change (no new dependencies), rebuilds take ~7 seconds instead of recompiling everything from scratch. The final image is ~137 MB.

Coolify

Set the build pack to Dockerfile with:

• Dockerfile location: /backend/Dockerfile
• Build context: /backend
• Environment variables: DATABASE_URL, PORT (default 3000), CACHE_TTL_SECS

---

Tech stack

• Extension: Vanilla JS, Chrome Manifest V3, chrome.storage.local
• Backend: Rust, Axum, Tokio, SQLx
• Database: PostgreSQL 18
• Caching: DashMap (in-memory, configurable TTL)
• Middleware: tower-http (CORS, tracing, static file serving)

Tests

cd backend
cargo test

Threshold logic has unit tests in backend/src/threshold.rs.

License

MIT