NSFW-Guard

Ever used a bunch of wildcards and ended up with a result you didn’t want to explain to your roommates? I built a little workaround for that.

I took the logic we usually use for detection and repurposed it to keep my local generations under control. It’s a bit of a "hack," but it works for me and I figured some of you might find it useful too!


🛡️ NSFW Image Guard for ComfyUI

A lightweight, robust security node designed to filter and censor NSFW content in real-time. Perfect for creators who stream, work in public spaces, or want to organize their generations automatically.

Key Features

• Flexible Censorship: Choose between Blur, a custom Image overlay, or a simple Black-Screen block.

• Silent Logging: Automatically saves flagged original images to a dedicated folder (output/nsfw_logs/) for private review.

• Detection Engine: Powered by standard detection and detailed segmentation models (e.g., for specific anatomy).

• Console Feedback: Color-coded terminal logs provide instant status updates and alerts.

1. Setup

• Install: Extract the folder ComfyUI-NSFW-Guard into your ComfyUI/custom_nodes/ directory.

• Models: Place your YOLO .pt detection or segmentation models in ComfyUI/models/ultralytics/.

  • (Recommended: Anime NSFW Detection/ADetailer All-in-One)

• Custom Assets: (Optional) To use a custom warning, place an image named stop.jpg (or .png/.webp) inside the node's folder.

2. How to Use

• Wiring: Connect your image source (e.g., VAE Decode) to the images input. Route the IMAGE output to your Preview or Save nodes.

• Model Selection: Choose your detection model from the model_name list.

• Configuration: Adjust the threshold (sensitivity) and pick your preferred filter_mode.

3. Filter Modes

• Blur: Applies a heavy Gaussian blur (radius 60). Keeps the color palette and composition visible while obscuring all details.

• Custom_Image: Replaces the blocked image with your custom stop.jpg file.

• Black: Replaces the content with a solid dark frame and a red "NSFW CONTENT BLOCKED" warning.

4. Settings Explained

• threshold:

  • 0.35 - 0.45: High sensitivity; recommended for maximum safety.

  • 0.60 - 0.80: Standard detection; reduces false positives.

• Save: Toggles the background saving of flagged images to output/nsfw_logs/.

• debug_mode: Toggles terminal notifications (Red for blocks, Green for status).

5. Terminal Feedback

• 🔴 [NSFW Guard] ALERT: NSFW content detected! Displays in red when a block occurs.

• 🟡 [NSFW Guard] WARNING: Triggered if "Custom_Image" mode is active but no stop.jpg is found in the node directory.

• 100% success rate is technically impossible to guarantee.

VibeCoded.

So it can have Bugs.

Write me and i ask Gemini or Chatgpt to remove the bugor do it by yourself.