MagicMirror: A lightweight open source face replacement tool

By /

Today we will introduce an open source face-changing project called MagicMirror。It’s created by GitHub users idootop The goal of the lightweight face replacement tool developed is to seamlessly replace one face into another image through image processing technology.

Although its name is “MagicMirror”, it is not a smart mirror, but aImage face-changing system based on deep learning, suitable for image processing, AI project demonstrations, or face-changing technology research.

1. Analysis of the project structure

Let’s start from the source code directory structure to see how this project is organized:

MagicMirror/
├── app/
│ ├── assets/
│ ├── inference.py
│ └── model.py
├── checkpoints/
├── configs/
│ └── default.yaml
├── outputs/
├── results/
├── app.py
├── swap.py
├── README.md

Let’s explain the role of the key parts one by one:

1. app/

Core code directory, including:

  • inference.pyThe main reasoning logic of face replacement. Process input images, detect faces, extract features and complete synthesis.
  • model.py: Load models, define network structure interfaces, etc.
  • assets/: Contains test pictures or auxiliary resources.

2. checkpoints/

Location where the model weight file is stored. Due to its large size, manual download is usually required. Used to load the trained face replacement model.

3. configs/

Contains default configuration files (such as default.yaml), used to control model parameters, path settings, post-processing switches, etc.

4. outputs/

Output the composite result picture.

5. results/

May be used to save intermediate steps or comparison results before and after.

6. Root directory script:

  • app.py: Usually used as a startup file for a Flask or Web service (may be used to deploy an online face-changing API).
  • swap.py: Provide command-line image face-changing entry script.

🧠2. How does it work?

The core process of MagicMirror is roughly as follows:

  1. Load the source face map (source) and target map (target);
  2. Detect faces in images (using face detection models);
  3. Extract facial feature vectors (usually using Face Recognition networks);
  4. Generating face-changing images using pre-trained models;
  5. Carry out integration and post-processing to improve naturalness;
  6. Output image to outputs/ or results/ Contents.

🚀3. How to operate?

Although no detailed explanation is provided in the README, the basic operation can be inferred from the code structure:

Installation dependencies:

pip install -r requirements.txt

Execute the command line to change faces:

python swap.py --source path/to/source.jpg --target path/to/target.jpg

Or run a web service:

python app.py

The prerequisite is to prepare the model weight file in advance and place it in checkpoints/ Under the catalog.

🔍4. Uses and advantages

  • Support offline operation without networking;
  • Based on PyTorch, it is easy to modify and extend;
  • The structure is clear and easy to learn and research;
  • Ideal for creating demos, research projects, or even integrating into web services or desktop applications.

Summary:

MagicMirror is a concise but practical face-changing toolkit that uses deep learning models to complete image-level face replacement tasks.
Whether you are interested in image processing, are studying GAN, want to prototype a face-changing App, or are just simply curious, this project is worth a try.

If you want to understand more about the specific algorithm principles or model structure, you can continue to explore model.py and inference.py Implementation details in.

GitHub：https://github.com/idootop/MagicMirror

Oil tubing:

Scroll to Top