Face swapping on Mac M1, M2, and M3: A Guide to the Roop Project

Note: The Roop project has been discontinued by its creator. The latest open-source technology for face swapping on Mac systems is FaceFusion. While the instructions in this article for the Roop Project will still work in most cases, it’s recommended to use the latest technology.

To install FaceFusion on your system, follow the instructions in this link:

Face swapping, the process of seamlessly replacing one face with another in images or videos, has become increasingly accessible thanks to advances in deep learning and projects like Roop. While traditionally resource-intensive, Roop aims to bring face swapping capabilities to powerful machines like the Apple Silicon-powered M1, M2, and M3 Macs.

Setting Up Your Environment

Install Conda: Download and install Miniconda or Anaconda to manage your Python environment. Check below mentioned article.

Create Environment:

conda create -n roop-faceswap python=3.10
conda activate roop-faceswap

Clone the Roop Repository:

git clone https://github.com/s0md3v/roop.git
cd roop

Install FFmpeg:

brew install ffmpeg

Install Initial Dependencies:

pip install -r requirements.txt

Troubleshooting M1-Specific Issues:

If you encounter errors during installation, you likely need to adapt the requirements.txt file for compatibility. Edit requirements.txt and paste below lines:

# PyTorch should be installed according to the official instructions for M1 Macs from the PyTorch website

numpy==1.24.3
opencv-python==4.8.0.74  # Ensure this version is compatible with M1, or consider using opencv-python-headless if GUI is not required
onnx==1.14.0

# insightface should be verified for compatibility or installed from source if needed
insightface==0.7.3

psutil==5.9.5
tk==0.1.0  # GUI libraries may have issues on M1; verify compatibility
customtkinter==5.2.0  # Same as above regarding GUI libraries

# Conditional installation for tkinterdnd2 based on platform; this should be fine as is
tkinterdnd2==0.3.0; sys_platform != 'darwin' and platform_machine != 'arm64'
tkinterdnd2-universal==1.7.3; sys_platform == 'darwin' and platform_machine == 'arm64'

pillow==10.0.0

# Specific onnxruntime version for M1 Mac
onnxruntime-silicon==1.13.1; sys_platform == 'darwin' and platform_machine == 'arm64'

# TensorFlow installation should be adapted for M1 Macs, possibly using a version built for ARM64 or TensorFlow-Metal
tensorflow==2.13.0  # Verify compatibility or consider TensorFlow-Metal for M1 Macs

opennsfw2==0.10.2  # Verify compatibility
protobuf==4.23.4
tqdm==4.65.0

# gfpgan should be verified for compatibility or installed from source if needed
gfpgan==1.3.8

Additionally, install PyTorch and its related libraries with:

pip install torch torchvision torchaudio

Running Roop:

Once all dependencies are successfully installed:

python run.py

Keep in mind that the first time you run Roop, it will download necessary models, which might take some time depending on your internet connection.

It will open a window:

Select a face as a source (image) and in Select a target as a target (image or video) and click Start.

Enjoy Face Swap

Conclusion:

Running faceswap projects like Roop on the latest Apple silicon requires careful configuration to ensure compatibility and performance. By following the steps outlined in this guide, users can set up their Mac M1, M2, or M3 for efficient faceswap operations with the Roop project. This opens up exciting possibilities for creators and developers in various multimedia and creative fields.