faceswap: a deep learning tool for recognizing and swapping faces in pictures and videos

What it solves

FaceSwap is designed to recognize and swap faces in images and videos using deep learning. It provides a cohesive toolset for users to create realistic face replacements, whether for movies, social commentary, or AI experimentation.

How it works

The software follows a three-step pipeline:

1. Extract: It identifies and extracts faces from raw photos or video frames.
2. Train: It uses the extracted face data to train a deep learning model to recognize the features of two different faces.
3. Convert: It applies the trained model to swap the faces in the original source material.

Users can operate the software via a command-line interface or a graphical user interface (GUI).

Who it’s for

It is intended for VFX artists, programmers, engineers, activists, and hobbyists who want to experiment with AI techniques or create content for ethical and legitimate purposes.

Highlights

• Cross-platform support: Runs on Windows, Linux, and MacOS.
• Hardware acceleration: Supports NVIDIA GPUs via CUDA and AMD GPUs via ROCm (on Linux).
• Multiple models: Includes various generative models like Phaze-A, Villain, and OHR.
• Integrated GUI: Offers a visual interface for users who less comfortable with the command line.