Skip to content

Latest commit

 

History

History
477 lines (333 loc) · 12.3 KB

README.md

File metadata and controls

477 lines (333 loc) · 12.3 KB

Prompt Inspector 🔎

Inspect AI image generation metadata from Discord uploads. Enhanced fork with powerful security features and streamlined setup.

Deploy on Railway

📑 Table of Contents

Features

Core Features

  • 🔍 Comprehensive Metadata Parsing - Supports Forge/A111, 200+ ComfyUI nodes (FLUX, PixArt, Griptape, etc.) as well as SwarmUI etc.
  • ✨ JPG Upload Support Where Discord fails, there is a R2 bucketing support for Jpeg/WebP Upload/Processing support.
  • 1️⃣2️⃣3️⃣ Multiple Interaction Styles - Numbered emoji reactions, slash commands, or context menus
  • 📦 Batch Processing - Handle 6+ images with a single reaction
  • 👥 PluralKit Support - Automatically resolves proxied messages
  • 💾 Lightweight - Runs metadata parser in separate process (~100MB RAM)

AI Features

  • ✨ /describe - Generate AI descriptions (Danbooru tags or natural language)
  • 💬 /ask - Conversational AI with context memory
  • 💬 /coder - Coding Help with AI context.
  • 💬 /techsupport - Will ask you if you turned it off and on again.
  • 🔄 Multi-Provider - Gemini + Claude with automatic fallback

Security Features

  • 🛡️ Anti-Scam Detection - Behavior-based crypto/wallet spam detection
  • 🚫 Cross-Posting Protection - Instant ban for spam across multiple channels
  • 🔒 Malware Prevention - Magic bytes check on attachments and embeds
  • 📊 Smart Scoring - Context-aware detection (allows emotional spam from trusted users)

Screenshots & Examples

Bot in Action

Metadata extraction and display Metadata Extraction
View comprehensive metadata from AI-generated images with clean embeds 		Settings management Per-Server Settings
Configure features per-server with an intuitive settings UI
Fun commands Fun Commands
Interactive commands like /goodnight with random GIFs and messages 		AI commands AI-Powered Commands
/describe, /ask, /coder, /techsupport with Gemini + Claude integration
Context menus and reactions
Multiple Interaction Styles
Use emoji reactions, slash commands, or right-click context menus

Quick Start

Local Setup 
# Clone and setup
git clone https://github.com/Ktiseos-Nyx/PromptInspectorBot.git
cd PromptInspectorBot
python3 -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -r requirements.txt

# Configure
cp config.example.toml config.toml
# Edit config.toml with your settings

# Run (new modular version)
python3 main.py
Docker Deployment 
# Build
docker build -t prompt-inspector-bot .

# Run with .env file
docker run -d --env-file .env prompt-inspector-bot
Environment Variables (Required) 
BOT_TOKEN=your_discord_bot_token

# Optional but recommended
GEMINI_API_KEY=your_gemini_key
ANTHROPIC_API_KEY=your_claude_key

How to Use

Metadata Inspection

  1. Post an image in a monitored channel
  2. Click emoji reactions:
    • 1️⃣ 2️⃣ 3️⃣ 4️⃣ 5️⃣ - View individual image metadata (1-5 images)
    • 📦 - Download all metadata as files (6+ images)
  3. Or use commands:
    • /metadata <image> - Parse uploaded image
    • Right-click image → "View Prompt"

AI Features

  • /ask <question> - Chat with AI (remembers context per user)
  • /describe <image> <style> - Generate AI tags/descriptions

Security System

🛡️ Anti-Scam Features

What It Detects

Type 1: Wallet Scammers

  • Currency symbols in username (£, €, ¥, ₿)
  • Hoisting characters (=, !, #)
  • ALL CAPS crypto spam
  • Keywords: WALLET, SOL, PAY, DEAD TOKENS

Type 2: Screenshot Spammers

  • 4+ images cross-posted to multiple channels
  • Gibberish text or empty messages
  • Auto-generated usernames (word.word1234_5678)

How It Works

Score Action
100+ Instant Ban - User banned, all messages deleted
75-99 Delete + Alert - Message removed, admins notified
50-74 Watchlist - Logged for monitoring

Detection Methods

✅ Magic Bytes Check - Scans attachments AND embeds for malware ✅ Cross-Posting - Same message in 2+ channels = ban ✅ Gibberish Detection - Context-aware (allows "AAAA" from users with roles) ✅ Username Analysis - Hoisting, currency symbols, auto-generated patterns ✅ Role Tracking - CATCHER role exploitation detection

Configuration

# config.toml
CATCHER_ROLE_ID = 1336289642789470228  # Self-assignable role scammers exploit
TRUSTED_USER_IDS = [123456789]         # Bypass security (mods, bots)
ADMIN_CHANNEL_ID = 1234567890          # Ban notification channel

Automatic Bypasses:

  • ✅ Server owners
  • ✅ Trusted users (configured above)

AI Provider Setup

🤖 Multi-Provider System

The bot supports both Gemini and Claude with automatic fallback!

Quick Setup

# Set API keys for providers you want
GEMINI_API_KEY=your_gemini_key
ANTHROPIC_API_KEY=your_claude_key

# Provider priority (tries first, falls back to next)
LLM_PROVIDER_PRIORITY=claude,gemini

Gemini Configuration

GEMINI_PRIMARY_MODEL=gemini-flash-latest
GEMINI_FALLBACK_MODELS=gemini-flash-latest,gemini-2.5-pro,gemini-2.5-flash

Model Recommendations:

Model Speed Quality Free Tier Best For
gemini-flash-latest ⚡⚡⚡ Good 15/min Recommended
gemini-2.5-pro ⚡ Excellent 2/min Complex tasks
gemini-2.5-flash ⚡⚡⚡ Good 15/min ⚠️ Overly strict

Claude Configuration

CLAUDE_PRIMARY_MODEL=claude-3-5-haiku-20241022

Model Recommendations:

Model Speed Cost (per 1M tokens) Best For
claude-3-5-haiku-20241022 ⚡⚡⚡ $0.25 / $1.25 Recommended
claude-3-5-sonnet-20241022 ⚡⚡ $3 / $15 Higher quality

Budget tip: $5 starter credit = ~1,500 image descriptions with Haiku!

NSFW/Artistic Content Mode

If Gemini blocks artistic content (PG-13/R-rated, not NC-17):

NSFW_PROVIDER_OVERRIDE=claude

This skips Gemini entirely for /describe and uses only Claude.

R2 Upload Feature (Optional)

📤 Cloudflare R2 Integration

Upload JPEG/WebP files to extract metadata without Discord stripping it.

Features

  • 🔒 Rate Limited - Generous rate limiting to prevent CPU overload.
  • 🗑️ Auto-Delete - 30-day retention
  • 📏 Size Limited - 10MB max to keep to Non-Nitro.

Setup

  1. Create R2 Bucket in Cloudflare
  2. Set environment variables:
R2_ACCOUNT_ID=your_account_id
R2_ACCESS_KEY_ID=your_access_key
R2_SECRET_ACCESS_KEY=your_secret_key
R2_BUCKET_NAME=your_bucket_name
UPLOADER_URL=https://your-pages.pages.dev/uploader.html
  1. Deploy uploader.html to Cloudflare Pages
  2. Set lifecycle rule for 30-day auto-deletion:
    • Prefix: uploads/
    • Action: Delete object
    • Days: 30

Security

  • ✅ Rate limiting (5/day per user)
  • ✅ File validation (10MB, JPEG/WebP only)
  • ✅ Presigned URL expiry (1 hour)
  • ✅ Auto-deletion (30 days)
  • ⚠️ Users warned about security risks

Configuration

⚙️ Full Configuration Options

Server & Channel Configuration

# Allowed servers (empty = all servers)
ALLOWED_GUILD_IDS = [123456789, 987654321]

# Monitored channels (empty = all channels)
MONITORED_CHANNEL_IDS = []

# Per-channel features
[channel_features]
1234567890 = ["metadata", "describe"]
9876543210 = ["ask"]

Bot Behavior

REACT_ON_NO_METADATA = false
EMOJI_METADATA_FOUND = "🔎"
EMOJI_NO_METADATA = "⛔"
SCAN_LIMIT_BYTES = 10485760  # 10MB

AI Configuration

LLM_PROVIDER_PRIORITY = ["claude", "gemini"]
GEMINI_PRIMARY_MODEL = "gemini-flash-latest"
CLAUDE_PRIMARY_MODEL = "claude-3-5-haiku-20241022"

Security Configuration

CATCHER_ROLE_ID = 1336289642789470228
TRUSTED_USER_IDS = []
ADMIN_CHANNEL_ID = 0

Permissions

Core Permissions (Required)

  • ✅ Read Messages/View Channel
  • ✅ Send Messages
  • ✅ Read Message History
  • ✅ Add Reactions
  • ✅ Attach Files

Security Permissions (Optional)

  • 🛡️ Ban Members - Auto-ban scammers
  • 🛡️ Manage Messages - Delete spam

Troubleshooting

Common Issues

Bot not responding:

  • Check bot is online
  • Verify permissions (see above)
  • Check logs for errors

Describe feature not working:

  • Ensure GEMINI_API_KEY or ANTHROPIC_API_KEY is set
  • Check API quotas
  • Review logs for API errors

Images not processed:

  • Check file format (PNG, JPEG, WebP)
  • Verify size is under SCAN_LIMIT_BYTES
  • Ensure channel is in MONITORED_CHANNEL_IDS (if configured)

Security system banning legit users:

  • Add user ID to TRUSTED_USER_IDS
  • Server owners are automatically trusted
  • Users with roles can post emotional spam ("AAAA")

Forking This Bot

🍴 Important Notes for Forks

If you're forking this bot for your own use, please update the following:

Required Changes

  1. Ko-fi Links - Replace with your own: - uploader.html: Lines 112, 114

    • bot_enhanced.py: Search for ko-fi.com/OTNAngel and ko-fi.com/duskfallcrew
    • Update to your Ko-fi/donation links

  2. Discord Invite - Replace with your server:

    • uploader.html: Line 117 - discord.gg/HhBSvM9gBY
    • bot_enhanced.py: Search for discord.gg/HhBSvM9gBY
    • Update to your support server invite

  3. Ko-fi Role ID - Set your role ID:

    • config.toml: Add KOFI_SUPPORTER_ROLE_ID = your_role_id
    • Get from Discord → Server Settings → Roles → Right-click role → Copy ID

Optional (But Appreciated)

  • Keep the Ko-fi link in uploader.html if you want to support the original developer
  • Credit this fork in your README
  • Share improvements back via pull requests!

Rate Limits (Configurable)

You can adjust these in bot_enhanced.py:

MAX_UPLOADS_PER_MINUTE = 10  # Burst protection
MAX_UPLOADS_PER_DAY_FREE = 100  # Free users
MAX_UPLOADS_PER_DAY_SUPPORTER = 500  # Ko-fi supporters

Legal

Quick Summary:

  • ✅ Extract metadata, send to you
  • ✅ Delete images immediately after processing
  • ❌ Don't store images or metadata
  • ❌ Don't track or sell data

🤖 This bot is a fork of PromptInspectorBot with significant enhancements.