1. Workflow Overview

This workflow specializes in magazine-grade facial expression editing, featuring:

• 6-level parametric control

• Non-destructive editing pipeline

• EXP_DATA reuse system

2. Core Technology

• ExpressionEditor: 3D morphable model-based system

• GQ Portrait Dataset: Built-in fashion photography templates

3. Critical Nodes

4. Processing Pipeline

Input Phase:

• Source image loading (982x1400)

• Auto 68-point facial landmark detection

Editing Stack:

1. Base Layer (Node 10):

   • Mouth slightly open (1.0)

   • Eyebrows lowered (-15)

2. Eye Enhancement (Node 13):

   • Pupil dilation (11.2)

   • "OnlyEyes" isolation mode

3. Dynamic Blending (Nodes 14/17/19/21):

   • Multi-parameter mixing

Output:

• Saves final composite with metadata

5. Parameter Schema（Python）

{
  "eyebrow_raise": 0,     # Range: [-20,20] 
  "eye_blink": 3,         # 0=open, 10=closed
  "intensity": 1.7        # Global multiplier
}

6. Usage Notes

1. Prerequisites:

   • FaceX plugin + dlib library

2. Input Specifications:

   • Min. 800x1200 resolution

   • Front-facing portraits

3. Optimization:

   • CUDA acceleration recommended

   • Layer complex expressions progressively

4. Layer complex expressions progressively

Special Notes

• Optimized for Asian facial features, adjust face_width for Caucasian

• EXP_DATA saving enables batch processing

• Version-lock recommended due to parameter range changes