🧠 Project Overview

DeepFaceEdit is an interactive face generation and editing tool built on top of StyleGAN2-ADA, designed for manipulating facial attributes in a highly intuitive way. It enables users to generate, project, and semantically edit high-quality face images using latent space traversal techniques all through a user-friendly GUI.

Whether you're experimenting with synthetic identity creation or exploring the interpretability of GANs, DeepFaceEdit provides powerful tools to visualize and manipulate the latent space of human faces.

---

✨ Features

• 🧬 Face Generation
  Generate realistic and diverse human faces from random seeds or latent vectors.

• 🛠️ Face Editing
  Modify facial attributes such as age, gender, expression, and more through semantic vector controls.

• 🖼️ Face Projection
  Project real images into StyleGAN2’s latent space using inversion techniques, enabling you to edit real faces.

• 🖱️ Interactive GUI
  Built with FreeSimpleGUI, the interface offers an intuitive editing experience without writing code.

• ⚡ Real-time Preview
  Instantly visualize changes as you adjust latent parameters or apply vector edits.

• 💾 Vector Reuse
  Save and reload latent vectors for consistent editing and iterative experimentation.

---

🔑 Key Capabilities

🎲 Face Generation

Generate unique faces using multiple techniques:

• Random seed generation – Explore diverse identities with different seeds.
• Z-vector based generation – Sample from the standard latent space.
• W-vector based generation – Leverage StyleGAN2’s intermediate latent space for better control.

🧭 Latent Space Manipulation

• Traverse interpretable directions in latent space to control age, gender, smile, and other high-level features.
• Combine multiple edits with fine-grained control over intensity.

📸 Face Projection

• Align real faces using dlib's facial landmarks.
• Perform GAN inversion to embed real images into the latent space for further editing.

---

---

Initially Passed Z Vector for Age (Changed the atribute of smiling) :

Initially Passed Z Vector for Mouth to Nose (Changed the atribute of gender) :

---

📦 Repository

📍 Clone this repo with submodules:

git clone --recurse-submodules https://github.com/Haseebasif7/GAN.git
cd GAN

🧩 The --recurse-submodules flag ensures the stylegan2-ada-pytorch directory is properly cloned.

If you forgot --recurse-submodules, fix it with:

git submodule update --init --recursive

---

🐍 Python Setup

1. ✅ Install Python 3.11

Ensure Python 3.11 is installed. You can download it from: https://www.python.org/downloads/release/python-3110/

Check installation:

python --version

⚠️ On Windows, check "Add Python to PATH" during installation.

2. 🧪 Create and Activate a Virtual Environment

Create a virtual environment in the project root:

python -m venv venv

Activate it:

Windows:

venv\Scripts\activate

Linux/macOS:

source venv/bin/activate

3. 📥 Install Dependencies

Upgrade pip:

pip install --upgrade pip

Then install all required packages:

pip install -r requirements.txt

💡 If you face issues installing dlib, make sure C++ build tools are available:

• Windows: Install Microsoft Build Tools
• Linux/macOS: Install cmake, g++, and Python headers

---

🎯 Download Pretrained Models

Two model files are required in the models/ folder.

✅ 1. ffhq.pkl (StyleGAN2-ADA Pretrained Model)

Download from:

https://nvlabs-fi-cdn.nvidia.com/stylegan2-ada-pytorch/pretrained/ffhq.pkl

Then place it in the models/ folder:

GAN/
├── models/
│   └── ffhq.pkl

✅ 2. shape_predictor_68_face_landmarks.dat (Face Landmark Detector)

Download the compressed file:

http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2

Place it in the models/ folder:

GAN/
├── models/
│   └── shape_predictor_68_face_landmarks.dat.bz2

Decompress it using the included script:

python models/decompress.py

✅ After decompression, you will have:

GAN/
├── models/
│   ├── shape_predictor_68_face_landmarks.dat
│   └── shape_predictor_68_face_landmarks.dat.bz2

---

🧱 Git Submodule Configuration

This project uses the StyleGAN2-ADA repo as a Git submodule.

Here’s the .gitmodules content:

[submodule "stylegan2-ada-pytorch"]
  path = stylegan2-ada-pytorch
  url = https://github.com/NVlabs/stylegan2-ada-pytorch.git

Git reads this automatically when you clone with --recurse-submodules.

---

🧩 Project Components

• Generator – Responsible for generating high-quality face images using the StyleGAN2-ADA model.
• Shifter – Handles latent space edits to apply semantic transformations (e.g., smile, age).
• Projector – Projects real-world face images into the latent space of the generator.
• Controller – Orchestrates the interaction between backend modules and manages workflow logic.
• Graphical Interface (GUI) – A user-friendly front-end built with FreeSimpleGUI for interactive editing and visualization.

## 📁 Final Project Structure (After Setup)

gan-face-editor/
├── core/
│   ├── align_faces.py
│   ├── generator.py
│   ├── projector.py
│   └── shifter.py
├── controller.py
├── facegan-env/
├── gui/
│   ├── main.py
│   └── layouts/
│       ├── interface.py
│       ├── project.py
│       └── sliders.py
├── LICENSE
├── models/
│   ├── ffhq.pkl
│   ├── shape_predictor_68_face_landmarks.dat
│   ├── shape_predictor_68_face_landmarks.dat.bz2
│   └── decompress.py
├── README.md
├── requirements.txt
├── results/
│   ├── age(2).png
│   └── m_n(2).png
├── run.py
├── settings/
│   ├── config.py
│   └── __init__.py
├── stylegan2-ada-pytorch/
├── utils/
│   ├── helpers.py
│   └── __init__.py
├── vectors/
│   ├── age.npy
│   ├── eye_distance.npy
│   ├── eye_eyebrow_distance.npy
│   ├── eye_ratio.npy
│   ├── eyes_open.npy
│   ├── gender.npy
│   ├── lip_ratio.npy
│   ├── mouth_open.npy
│   ├── mouth_ratio.npy
│   ├── nose_mouth_distance.npy
│   ├── nose_ratio.npy
│   ├── nose_tip.npy
│   ├── pitch.npy
│   ├── roll.npy
│   ├── smile.npy
│   └── yaw.npy

🛠️ Technologies Used

• Python
• VGG16
• PyTorch
• StyleGAN2-ADA
• dlib
• FreeSimpleGUI

---