ShortSwap

Complete end-to-end automation of YouTube Shorts face-swapping with both CLI and GUI interfaces.

Features

• CLI Interface - Fully automated command-line pipeline
• GUI Interface - User-friendly graphical interface with desktop integration
• Resume Capability - Resume incomplete jobs that failed mid-processing
• YouTube Download - Automatic download of YouTube Shorts
• Face Swapping - Batch face-swap all video frames using InsightFace
• GPU Acceleration - CUDA support for faster processing
• Gender Filtering - Target specific genders for face swapping
• Person Targeting - Swap faces of specific individuals using reference images
• Watermarking - Optional watermark on output videos
• Progress Tracking - Real-time progress monitoring

What It Does

1. Downloads YouTube Short (or uses local video file)
2. Extracts video into frames
3. Batch face-swaps all frames using InsightFace + inswapper
4. Rebuilds video from swapped frames
5. Adds original audio to final video

Components

Core Scripts

• shortswap.sh - Main automated CLI pipeline
• shortswap_gui.py - Graphical user interface
• run_shortswap_gui.sh - GUI launcher script
• resume_shortswap.sh - Resume incomplete/failed jobs

Dependencies

• impostr_cli.py - Face-swapping engine
• youtube_shorts_downloader.py - YouTube download component

Documentation

• ShortSwap_README.md - CLI documentation
• ShortSwap_GUI_README.md - GUI documentation
• impostr_cli_README.md - Face-swapping component docs
• youtube_shorts_downloader_README.md - YouTube downloader docs

Requirements

System Dependencies

• Python 3.7+
• ffmpeg
• CUDA toolkit (optional, for GPU acceleration)

Python Dependencies

pip install insightface onnxruntime opencv-python Pillow numpy
# For GPU support:
pip install onnxruntime-gpu

AI Model

Download the inswapper_128.onnx model and place it in models/ directory:

• Model: inswapper_128.onnx
• Size: ~554MB
• Download: InsightFace Model Zoo

Quick Start

CLI Usage

# Basic usage with YouTube URL
./shortswap.sh --url <YOUTUBE_URL> --face <SOURCE_IMAGE>

# With GPU acceleration
./shortswap.sh --url <YOUTUBE_URL> --face <SOURCE_IMAGE> --cuda

# With local video file
./shortswap.sh --video <VIDEO_FILE> --face <SOURCE_IMAGE>

# With gender filtering
./shortswap.sh --url <YOUTUBE_URL> --face <SOURCE_IMAGE> --gender male --cuda

GUI Usage

# Launch GUI
./run_shortswap_gui.sh

# Or directly
python3 shortswap_gui.py

Resume Failed Jobs

# Resume an incomplete job
./resume_shortswap.sh <job_directory> <source_face_image> --cuda

Desktop Integration

The GUI includes desktop integration with:

• Desktop launcher icon
• Application menu entry
• Icon: /home/panda/Pictures/logos/shortswap_logo.jpg

Desktop files:

• ~/Desktop/shortswap.desktop
• ~/.local/share/applications/shortswap.desktop

Performance

• CPU Processing: 7-22 minutes for 15-second video
• GPU Processing: 45 seconds - 4 minutes for 15-second video

Advanced Features

Gender Filtering

Target only male or female faces:

./shortswap.sh --url <URL> --face <IMAGE> --gender male
./shortswap.sh --url <URL> --face <IMAGE> --gender female

Person-Specific Targeting

Swap face of a specific person using a reference image:

./shortswap.sh --url <URL> --face <SOURCE> --target-person <REFERENCE_IMAGE>

Watermarking

Add watermark to output videos:

./shortswap.sh --url <URL> --face <IMAGE> --watermark

Architecture

ShortSwap Pipeline:
├── YouTube Download (youtube_shorts_downloader.py)
├── Frame Extraction (ffmpeg)
├── Face Swapping (impostr_cli.py + InsightFace)
├── Video Reconstruction (ffmpeg)
└── Audio Integration (ffmpeg)

Troubleshooting

Computer Freezes During Processing

If the computer freezes mid-processing, use the resume script:

./resume_shortswap.sh /path/to/job/directory /path/to/face/image --cuda

This will:

• Identify incomplete/failed frames
• Re-process only the missing frames
• Complete the video assembly

GPU Out of Memory

If you encounter GPU memory errors:

1. Use CPU mode (remove --cuda flag)
2. Close other GPU-intensive applications
3. Reduce video resolution before processing

No Face Detected

If frames have no detectable faces, they will be saved as-is without swapping.

License

This project uses InsightFace for face detection and swapping. Please review their license terms.

Credits

Built with:

• InsightFace - Face detection and swapping
• ONNX Runtime - Model inference
• yt-dlp - YouTube downloading
• FFmpeg - Video processing

---

Generated with Claude Code

Co-Authored-By: Claude [email protected]