Virtual Companion

The Virtual Companion is a fully animated 3D character that appears on your screen, making your AI interactions more engaging and human-like. Built with Babylon.js, it features realistic animations, lip-synced speech, and dynamic emotional expressions.

Overview

The companion responds to your interactions with expressions, gestures, and animations that match the conversation context. It exists as a draggable overlay that can be positioned anywhere on your screen.

Display Modes

Switch between full body and portrait views

Animations

Idle, thinking, speaking, and celebration states

Lip Sync

Real-time mouth movements synchronized to speech

Positioning

Drag anywhere on screen with per-preset saved positions

Display Modes

Choose how your companion appears on screen:

Full Body
Portrait

Full Body Mode shows the complete 3D character:

Entire character model visible
Full range of body animations
Hand gestures and full-body expressions
Larger canvas size for detailed animations

Best for: Desktop browsing with larger screens

Switching Display Modes

Open Control Panel

Click the VAssist icon in your browser toolbar

Navigate to Scene Settings

Go to Settings → Scene Configuration

Select Display Mode

Choose between Full Body and Portrait mode

Apply Changes

Changes take effect immediately

import { SceneConfig } from './config/sceneConfig';

// Configure display mode
const config = {
  cameraPreset: 'fullBody', // or 'portrait'
  model: {
    displayMode: 'full',  // 'full' or 'portrait'
    scale: 1.0
  }
};

Animations

The companion uses a sophisticated animation system with multiple states and smooth transitions.

Animation States

Idle

Default State - Subtle breathing and natural movements

Random idle variations
Occasional yawns and stretches
Greeting gestures (waves hello)
Smooth looping animations

Duration: Continuous, switches variants every 10 seconds

// Idle animations
const idleAnimations = [
  'idle_1',      // Primary idle with subtle movements
  'idle_2',      // Secondary idle variant
  'idle_4_short',// Short idle loop
  'yawn_1',      // Yawning animation
  'yawn_2',      // Alternative yawn
  'hi_1',        // Waving hello
  'hi_2'         // Alternative greeting
];

Thinking

Processing State - Shows the AI is working

Thoughtful poses
Hand-on-chin gestures
Contemplative expressions
Auto-switches between variants

Triggered when:

Waiting for AI response
Processing complex queries
Analyzing page content

// Thinking animations
const thinkingAnimations = [
  'thinking_1',  // Thoughtful pose
  'thinking_2'   // Alternative thinking animation
];

Speaking

Active Conversation - Animated speech with lip sync

Body gestures matching emotion
Real-time lip synchronization
Blended with base animation
Facial expressions

Emotion variants:

Excited - energetic gestures
Calm - relaxed movements
Nervous - anxious body language
Angry - emphatic gestures

// Talking animations by emotion
const talkingAnimations = {
  excited: 'talk_excited',
  nervous: 'talk_nervous',
  calm: 'talk_calm',
  angry: 'talk_angry'
};

Celebrating

Success State - Positive feedback animations

Clapping hands
Excited jumping
Happy expressions
One-shot animation

Triggered by:

Task completion
Success messages
Positive user interactions

// Celebration animations
const celebratingAnimations = [
  'clap_1'  // Clapping in celebration
];

Animation Blending

The companion uses smooth animation blending for natural transitions:

import { TransitionSettings } from './config/animationConfig';

// Smooth transitions between animations
const transitionConfig = {
  transitionFrames: 30,  // 1 second at 30fps
  easingCurve: {
    x1: 0.25,  // Bezier control points
    y1: 0.1,
    x2: 0.75,
    y2: 0.9
  }
};

// Blend from idle to thinking
await animationManager.transition({
  from: 'idle_1',
  to: 'thinking_1',
  duration: transitionConfig.transitionFrames
});

Key features:

Overlap blending - Old animation fades out while new one fades in
Bezier easing - Smooth S-curve for natural motion
Loop transitions - Seamless cycling of looping animations
Weight-based mixing - Blend multiple animations simultaneously

Lip-Synced Speech

The companion’s mouth movements are synchronized in real-time with text-to-speech output using VMD (Vocaloid Motion Data) generation.

How Lip Sync Works

TTS Generation

Kokoro.js generates speech audio from AI response text

Phoneme Analysis

Audio is analyzed to extract phoneme timing data

VMD Creation

Phonemes are mapped to mouth shapes (visemes) and converted to VMD format

Animation Playback

VMD animation is applied to the character’s mouth morphs in real-time

import { VMDHandler } from './services/VMDHandler';
import { TTSService } from './services/TTSService';

// Generate speech with lip sync
const { audio, vmd } = await TTSService.synthesize(text, {
  generateVMD: true,
  voice: 'af_heart',
  speed: 1.0
});

// Play audio and sync mouth movements
await virtualAssistant.speak({
  audio,
  vmdData: vmd,
  baseAnimation: 'talk_calm'
});

Viseme Mapping

The companion uses standard viseme shapes for realistic mouth movements:

Viseme	Phonemes	Example Words
A	ah, aa	father, hot
E	eh, ae	bed, cat
I	ih, iy	bit, see
O	oh, ao	go, caught
U	uh, uw	book, food
M	m, p, b	mom, pet, big
F	f, v	fox, van
TH	th, dh	think, this
S	s, z	sit, zoo
T	t, d, n	top, dog, no
L	l	love
R	r	red

Lip sync requires the TTS feature to be enabled. See Configuration for TTS setup.

Positioning & Customization

The companion can be freely positioned and customized to suit your workflow.

Dragging & Positioning

Click and Hold

Click on the companion’s canvas area

Drag to Position

Move the companion anywhere on your screen

Release to Lock

Position is automatically saved per display mode

Position Memory:

Each display mode (Full Body / Portrait) remembers its own position
Positions saved per browser profile
Persists across browser sessions

import { PositionManager } from './babylon/managers/PositionManager';

// Save position for current preset
await PositionManager.savePosition({
  preset: 'fullBody',
  x: window.innerWidth - 400,
  y: window.innerHeight - 600
});

// Load saved position
const position = await PositionManager.loadPosition('fullBody');

Customization Options

Customize the companion’s appearance and behavior:

Canvas Size

Adjust the companion’s display size:

Full Body: 400x600px (default)
Portrait: 300x400px (default)
Custom sizes available in settings

{
  canvas: {
    width: 400,
    height: 600,
    autoResize: true
  }
}

Character Scale

Scale the 3D model independently of canvas:

Range: 0.5 to 2.0
Default: 1.0
Affects model size only, not UI elements

{
  model: {
    scale: 1.2,  // 120% of original size
    position: { x: 0, y: -0.5, z: 0 }
  }
}

Animation Speed

Adjust animation playback speed:

Range: 0.5 to 2.0
Default: 1.0
Affects all animations globally

{
  animation: {
    globalSpeed: 1.0,
    transitionDuration: 30  // frames
  }
}

Lighting & Effects

Configure scene lighting and visual effects:

Ambient light intensity
Directional light position
Shadow quality
Background transparency

{
  scene: {
    ambientLightIntensity: 0.8,
    directionalLight: {
      intensity: 1.0,
      position: { x: 1, y: 1, z: 1 }
    },
    backgroundColor: 'transparent'
  }
}

Built with Babylon.js

The Virtual Companion leverages Babylon.js, a powerful 3D rendering engine, for smooth animations and realistic character rendering.

Technical Architecture

Rendering Engine
MMD Model Support
Animation System

Babylon.js provides:

WebGL-based 3D rendering
Hardware-accelerated graphics
Efficient animation system
Advanced physics and lighting

import * as BABYLON from '@babylonjs/core';

// Create Babylon.js scene
const engine = new BABYLON.Engine(canvas);
const scene = new BABYLON.Scene(engine);

// Optimize for performance
scene.skipPointerMovePicking = true;
scene.autoClear = false;
engine.enableOfflineSupport = false;

MikuMikuDance (MMD) format support:

PMX model loading
VMD animation playback
Morph targets for facial expressions
Bone-based skeletal animation

import { MmdRuntime } from '@babylonjs/mmd';

// Load MMD model
const mmdRuntime = new MmdRuntime(scene);
const modelMesh = await mmdRuntime.loadModel(
  'model.pmx',
  { buildSkeleton: true }
);

// Play VMD animation
await mmdRuntime.playAnimation('motion.vmd');

AnimationManager handles:

State machine for animation states
Smooth blending between animations
Composite animation mixing
Automatic idle variation switching

import { AnimationManager } from './babylon/managers/AnimationManager';

const animManager = new AnimationManager(mmdRuntime);

// Play animation with blending
await animManager.playAnimation('thinking_1', {
  transitionDuration: 30,
  loop: true,
  weight: 1.0
});

// Blend multiple animations
await animManager.playComposite([
  { id: 'idle_1', weight: 0.6 },
  { id: 'talk_calm', weight: 0.4 }
]);

Performance Optimization

The companion is optimized for smooth performance:

Efficient Rendering

60 FPS target frame rate
Culling for off-screen elements
LOD (Level of Detail) system
Adaptive quality settings

Smart Loading

Lazy loading of animations
Cached animation files
Progressive model loading
Resource pooling

Memory Management

Automatic garbage collection
Texture compression
Animation buffer reuse
Scene disposal on unmount

GPU Acceleration

WebGL 2.0 support
Hardware skinning
GPU-based morphing
Shader optimization

Interaction & Behaviors

The companion responds intelligently to different contexts:

Automatic State Changes

User Action	Companion Response
Send chat message	Transitions to Thinking state
Receive AI response	Plays Speaking animation with lip sync
Enable voice mode	Listens (idle) then responds (speaking)
Complete task	Plays Celebration animation
Idle for 30+ seconds	Occasional yawns and stretches
Hover over companion	Subtle acknowledgment gesture

Emotion-Based Animations

The AI can specify emotions in responses, and the companion reacts accordingly:

// AI response includes emotion metadata
const response = {
  text: "I'm excited to help you with that!",
  emotion: "excited"  // Maps to 'talk_excited' animation
};

// Companion uses corresponding animation
await companion.speak(response.text, response.emotion);

Emotion mapping can be customized in animationConfig.js. See Configuration for details.

Accessibility

The companion can be disabled for users who prefer a minimal interface:

Open Control Panel

Click the VAssist toolbar icon

Go to Scene Settings

Navigate to Settings → Scene Configuration

Toggle Companion

Disable “Show Virtual Companion”

With the companion disabled:

All AI features remain fully functional
Chat interface works normally
Voice mode still available
Lower resource usage

Next Steps

Chrome AI APIs

Learn about the AI powering the companion’s intelligence

Configuration

Customize the companion’s appearance and behavior

Get Started

Features

Guides

Virtual Companion

Overview

Display Modes

Animations

Lip Sync

Positioning

Display Modes

Switching Display Modes

Animations

Animation States

Animation Blending

Lip-Synced Speech

How Lip Sync Works

Viseme Mapping

Positioning & Customization

Dragging & Positioning

Customization Options

Built with Babylon.js

Technical Architecture

Performance Optimization

Efficient Rendering

Smart Loading

Memory Management

GPU Acceleration

Interaction & Behaviors

Automatic State Changes

Emotion-Based Animations

Accessibility

Next Steps

Chrome AI APIs

Configuration

Build docs developers (and LLMs) love

Get Started

Features

Guides

Documentation Index

​Overview

Display Modes

Animations

Lip Sync

Positioning

​Display Modes

​Switching Display Modes

​Animations

​Animation States

​Animation Blending

​Lip-Synced Speech

​How Lip Sync Works

​Viseme Mapping

​Positioning & Customization

​Dragging & Positioning

​Customization Options

​Built with Babylon.js

​Technical Architecture

​Performance Optimization

Efficient Rendering

Smart Loading

Memory Management

GPU Acceleration

​Interaction & Behaviors

​Automatic State Changes

​Emotion-Based Animations

​Accessibility

​Next Steps

Chrome AI APIs

Configuration

Build docs developers (and LLMs) love

Overview

Display Modes

Switching Display Modes

Animations

Animation States

Animation Blending

Lip-Synced Speech

How Lip Sync Works

Viseme Mapping

Positioning & Customization

Dragging & Positioning

Customization Options

Built with Babylon.js

Technical Architecture

Performance Optimization

Interaction & Behaviors

Automatic State Changes

Emotion-Based Animations

Accessibility

Next Steps