FaceBlend

A Python face morphing tool that creates seamless face composites by blending facial features from two images. Uses MediaPipe for 468-point facial landmark detection and OpenCV for image processing.

Example

Image 1	Image 2	Result

Features

• Automatic face detection and alignment
• Skin tone matching between faces
• Delaunay triangulation for smooth morphing
• EXIF orientation correction
• Batch processing support
• Configurable blend ratios

Installation

git clone https://github.com/yourusername/FaceBlend.git
cd FaceBlend
pip install -r requirements.txt

Quick Start

Single Pair

python face_blender.py person1.png person2.png -o result.jpg

Batch Processing

1. Add face images (PNG/JPG) to the Faces/ directory
2. Edit Faces.txt with pairs to blend:

   Alice, Bob
   Person1, Person2
   

3. Run:

   python batch_blend.py

Results are saved to Results/.

CLI Options

Option	Description	Default
-o, --output	Output filename	blended_result.jpg
-s, --size	Output size in pixels	600
-b, --blend	Blend ratio (0.0-1.0)	0.5

Blend Ratio

• 0.0 = Features mostly from image 1
• 0.5 = Equal blend from both images
• 1.0 = Features mostly from image 2

Project Structure

FaceBlend/
├── face_blender.py   # Core morphing engine
├── batch_blend.py    # Batch processing script
├── requirements.txt  # Python dependencies
├── Faces.txt         # Face pairs configuration
├── Faces/            # Input images (PNG/JPG)
└── Results/          # Output images

Requirements

• Python 3.8+
• Frontal or near-frontal face photos
• Clear, well-lit images

How It Works

1. Detects 468 facial landmarks using MediaPipe
2. Normalizes face orientation and skin tones
3. Creates Delaunay triangulation mesh
4. Warps triangles using affine transforms
5. Blends features with multi-tier masking
6. Composites hair from image 1 onto final result

Limitations

• Requires clearly visible, roughly frontal faces
• Extreme profile angles may fail detection
• Hair is always taken from the first image

License

MIT License - see LICENSE for details.