Multi-Model NSFW Image Scanner

This project provides a comprehensive NSFW (Not Safe For Work) image detection system using multiple AI models and CLIP analysis. It includes tools for converting various NSFW detection models to TFLite format and a scanner utility to analyze directories of images.

Features

• Multiple NSFW detection models:
  • Marqo NSFW Detection (384x384)
  • Falconsai NSFW Detection (224x224)
  • Giacomoarienti NSFW Classifier (224x224)
  • DamarJati NSFW Detection (224x224)
• CLIP-based image content analysis
• HTML report generation with image previews
• Interactive results viewer with full-size image modal
• Multi-model comparison support

Setup

1. Create and activate a virtual environment:

$ python -m venv venv

Activate on Linux/Mac

$ source venv/bin/activate

Activate on Windows

$ .\venv\Scripts\activate

2. Install dependencies: $ pip install -r requirements.txt

Converting Models

Before scanning images, you'll need to convert the models to TFLite format. Run the following conversion scripts:

1. Convert Marqo NSFW Detection model:

$ python convert_nsfw_model.py

2. Convert Falconsai NSFW Detection model:

$ python convert_falconsai_model.py

3. Convert Giacomoarienti NSFW Classifier:

$ python convert_giacomo_model.py

Each script will download the respective model from HuggingFace and convert it to TFLite format. The converted models will be saved in the current directory.

Usage

Basic Scanning

To scan a directory of images using all available models:

$ python scan_directory.py path/to/image/directory

Using a Specific Model

To use only one model for scanning:

$ python scan_directory.py path/to/image/directory --model-type marqo

Available model types:

• marqo
• falconsai
• giaco
• damarscan

Custom Output Location

To specify a custom output report location:

$ python scan_directory.py path/to/image/directory --output custom_report.html

Output

The script generates an HTML report in the scanned directory containing:

• Thumbnail previews of all images
• Detection scores from each model
• CLIP analysis results
• Processing duration for each image
• Interactive full-size image viewer
• Color coding for NSFW detections

Notes

• The scanner supports common image formats: .jpg, .jpeg, .png, .bmp, .gif
• CLIP analysis provides additional content understanding through text-image similarity
• First run will download the CLIP model (cached for subsequent runs)
• Processing time varies based on image count and enabled models

Requirements

• Python 3.7+
• CUDA-capable GPU recommended but not required
• Sufficient disk space for models (~1GB total)