zorrobyte / FamilySafe-NSFW-Finder

This program searches through a directory tree to find images that are not safe for work (NSFW) by using a machine learning algorithm. It then relocates these images to a designated directory.

FamilySafe NSFW Finder

Description

This script scans a directory for images and videos, using a classifier model to detect NSFW (Not Safe For Work) content. If NSFW content is found, the script can optionally move these media files to a specified directory. It supports recursive scanning of directories.

Features

• Scans specified directories for image and video files (.jpg, .jpeg, .png, .gif, .bmp, .tiff, .tif, .webp, .mp4, .avi, .mov, .mkv, .wmv, .flv, .webm, .mpeg, .mpg, .m4v).
• Uses Falconsai/nsfw_image_detection model to classify images as NSFW or safe.
• Can move detected NSFW media files to a chosen directory.
• Option for recursive directory scanning.

Requirements

• Python 3.x
• Anaconda or Miniconda (for TensorFlow and Keras support)

Installation

1. Ensure Python 3.x is installed on your system.
2. Clone the repository or download the script.
3. Create a Conda environment:

conda create -n nsfw_finder python=3.10 cuda-toolkit cudnn tensorflow
conda activate nsfw_finder

4. Set up the conda environment:

conda env config vars set LD_LIBRARY_PATH=$CONDA_PREFIX/lib
conda deactivate
conda activate nsfw_finder

5. Install required Python packages:

pip install -r requirements.txt

Usage

Activate the conda environment before using the script.

Run the script via command line, specifying the source directory and, if desired, a destination folder for detected media files:

conda activate nsfw_finder
python nsfw-finder.py [directory] [options]

Arguments

• directory: The directory to scan.
• -r, --recursive: Scan subdirectories recursively.
• -t, --target <directory>: Optional target directory for NSFW files.
• -d, --frame_delay <seconds>: Set the time gap between each frame analyzed from video files (default is 5 seconds).
  • A lower value will increase processing speed, but may lead to less accurate results.
  • Use this option if you want to prioritize speed over accuracy or need faster scans for large volumes of videos.
• -f, --frame_frequency <frames>: Set the number of frames between each frame analyzed from video files.
  • A higher value will reduce processing time but may decrease accuracy.

Example

python nsfw-finder.py ~/Media --recursive --target ~/.nsfw_media --frame_delay 2.0

or with frame delay option:

# Analyze a video at one frame every 10 seconds (for faster analysis, albeit slightly less accurate)
python nsfw-finder.py ~/Videos/ -d 10

Notes

• NSFW detection accuracy depends on the model.
• Internet access is required to download the model on first run.

License

Licensed under the MIT License - see the LICENSE file for details.

Contributors

• Josep Hervás (https://github.com/gestur1976)
• Ross Fisher (https://github.com/zorrobyte)

Contributions to the project are welcome. Please fork the repository and submit a pull request with your changes.

Disclaimer

This tool is for educational use only. The authors are not responsible for misuse or any resulting damage.