DeepFaceReal-Physics v2.0.0 — Hyper-Realistic AI Avatar

DeepFaceReal-Physics v2.0.0 — Hyper-Realistic AI Avatar with HeyGen-Level Realism

🎯 New Features

6 Core Engine Modules:

• 3D Face Engine — 468 MediaPipe face mesh landmarks, Delaunay triangulation for 3D surface mapping, 6DoF head pose estimation (pitch/yaw/roll/xyz), expression coefficient extraction
• Audio-Driven Talking Head — MFCC/pitch/energy feature extraction, head pose prediction from audio rhythm, speech-synchronized expressions, natural nodding/tilting (inspired by SadTalker/Ditto)
• Enhanced Wav2Lip Lip Sync — Phoneme detection from audio, lip shape prediction matching spoken phonemes, Wav2Lip inference on face region, temporal smoothness
• Natural Eye & Gaze Engine — Saccade simulation (200-300ms intervals), micro-saccades, natural blinks (every 2-4s, 100-400ms), gaze target tracking, pupil rendering
• Conversational Gesture Engine — Speech-rhythm hand gestures, shoulder/head micro-movements, posture variations, configurable intensity
• Real-Time Pipeline — Async pipeline wiring all engines: Audio→Features→TalkingHead→LipSync→EyeGaze→Gestures→Composite. Frame skipping, cached inference, async queues. CPU-optimized.

Professional UI/API:

• Streamlit UI v2 with all engine controls, Heygen Mode preset, mobile camera QR, recording export
• FastAPI v2 with POST /generate/talking-head, POST /animate/face, WebSocket /ws/realtime, POST /config/render
• Windows EXE Builder (build_exe.py) — PyInstaller bundling all modules

Documentation:

• SEO-optimized README with comparison table vs HeyGen/SadTalker
• Full architecture diagram and engine documentation

🔧 Changes

• Added 6 new engine modules to core/
• Updated app.py with v2 controls and Heygen Mode
• Updated api.py with v2 endpoints
• Added build_exe.py for Windows EXE
• Plans/ and strip_comments.py removed
• All files credited: Powered By DeathLegionTeamLK

📊 Performance

• CPU-optimized: Target 15-20 FPS at 640×480
• Frame skipping, cached inference, async queues
• No GPU required

⚠️ Known Limitations

• Wav2Lip model requires separate download to models/
• GFPGAN/Real-ESRGAN enhancement unavailable in sandbox (fallback: OpenCV enhancement)
• Virtual camera (v4l2loopback) requires bare-metal Linux

📦 Download

• Source code: GitHub checkout
• Windows EXE: Build with build_exe.py

---

Powered By DeathLegionTeamLK

DeepFaceReal Physics v1.0.0 — Full Body Tracking + Physics + WhatsApp

🎭 DeepFaceReal Physics v1.0.0 — Full Body Tracking + Physics + WhatsApp

What's New

• Full Body Tracking: MediaPipe Holistic integration with 543 landmarks (33 pose + 21 left hand + 21 right hand + 468 face mesh)
• Physics Simulation Engine: Real-time momentum/inertia, spring physics, hand physics, and face micro-expression simulation
• Parallax Background: 3-layer parallax background system with depth-based blur (far/mid/near planes)
• WhatsApp Mobile: IP Webcam support for Android phone camera integration via MJPEG stream
• Enhanced Face Swap: Physics-based inertia smoothing, hand overlay rendering, blink simulation

Features

• 🔄 Real-time face swap with InsightFace (inswapper_128)
• 🧍 Full body tracking via MediaPipe Holistic
• ⚡ Physics engine (momentum, springs, inertia)
• 🎨 3-layer parallax background system
• 📱 WhatsApp Mobile integration (IP Webcam)
• 🖥️ WhatsApp Desktop support (virtual camera)
• 💬 LLM character responses via OpenRouter
• 🎭 Character management system
• 🎯 Streamlit UI with real-time controls
• ⚙️ FastAPI backend with REST + WebSocket endpoints

Powered By DeathLegionTeamLK

Installation

See README.md for full setup instructions.