Elena Avatar Configuration - Azure AI Foundry

Status: Ready for Configuration
Date: January 2026
Feature: Azure AI Foundry TTS Avatar

---

Overview

Azure AI Foundry’s TTS Avatar feature enables Elena to have a photorealistic video avatar that speaks with natural voice and expressions. This creates a more engaging and interactive user experience.

---

Avatar Features

✅ Available Features

1. Photorealistic Video Avatar
   • High-quality video generation
   • Natural facial expressions
   • Lip-sync with speech
2. Voice Integration
   • Uses Elena’s voice: en-US-JennyNeural
   • Natural-sounding speech
   • Emotion control
3. Resolution Options
   • 720p (standard)
   • 1080p (recommended)
   • 4K (premium)
4. Emotion Control
   • Neutral (default)
   • Happy
   • Sad
   • Angry
   • Custom emotions
5. Background Options
   • Transparent (for overlays)
   • Custom backgrounds

---

Configuration

Method 1: Script (Recommended)

Use the configuration script:

cd /Users/derek/Library/CloudStorage/OneDrive-zimaxnet/code/engram
source .venv/bin/activate
python3 scripts/configure_elena_avatar.py

What it does:

• Fetches current Elena agent definition
• Adds avatar configuration
• Creates new agent version with avatar enabled

Method 2: Azure Portal

1. Go to: Azure AI Foundry → Project zimax → Applications → Elena
2. Navigate to: Avatar settings
3. Configure:
   • Avatar ID: en-US-JennyNeural (matches Elena’s voice)
   • Style: Professional
   • Resolution: 1080p
   • Background: Transparent

Method 3: REST API

# Get access token
TOKEN=$(az account get-access-token --resource "https://ai.azure.com" --query accessToken -o tsv)

# Get current agent
curl -X GET \
  "https://zimax.services.ai.azure.com/api/projects/zimax/agents/Elena?api-version=2025-11-15-preview" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json"

# Create new version with avatar config
curl -X POST \
  "https://zimax.services.ai.azure.com/api/projects/zimax/agents/Elena/versions?api-version=2025-11-15-preview" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "definition": {
      "kind": "prompt",
      "instructions": "...",
      "model": "gpt-5.1-chat",
      "tools": [...],
      "avatar": {
        "avatar_id": "en-US-JennyNeural",
        "style": "professional",
        "emotion": "neutral",
        "resolution": "1080p",
        "background": "transparent"
      }
    }
  }'

---

Avatar Configuration Schema

{
  "avatar": {
    "avatar_id": "en-US-JennyNeural",
    "style": "professional",
    "emotion": "neutral",
    "resolution": "1080p",
    "background": "transparent"
  }
}

Configuration Options

Field	Type	Options	Description
avatar_id	string	Standard avatars or custom ID	Avatar to use (matches voice)
style	string	professional, casual, friendly	Avatar presentation style
emotion	string	neutral, happy, sad, angry	Default emotion
resolution	string	720p, 1080p, 4K	Video resolution
background	string	transparent, custom URL	Background type

---

Integration with Engram

Frontend Integration

Elena’s avatar can be displayed in the frontend:

// AvatarDisplay component already supports Elena
<AvatarDisplay
  agentId="elena"
  isSpeaking={isSpeaking}
  expression={expression}
  visemes={visemes}
  showName={true}
  size="lg"
/>

Backend Integration

When Elena responds via Foundry, the avatar video is automatically generated:

# Foundry agent response includes avatar video URL
response = await foundry_client.invoke_agent(
    agent_id="Elena",
    thread_id=thread_id,
    user_id=user_id,
)

# Response includes:
# - text: Response text
# - avatar_video_url: URL to generated avatar video (if enabled)
# - avatar_audio_url: URL to audio track

---

Custom Avatar Creation

Option 1: Standard Avatar

Use Azure’s pre-built avatars:

• en-US-JennyNeural (matches Elena’s voice)
• Professional appearance
• Ready to use immediately

Option 2: Custom Avatar

Create a custom avatar with Elena’s likeness:

1. Record Training Video:
   • Record video of talent (with consent)
   • Professional lighting and setup
   • Clear audio
2. Upload to Foundry:
   • Go to: Azure AI Foundry → Avatar Studio
   • Upload training video
   • Configure settings
3. Train Avatar Model:
   • Foundry trains the model
   • Takes several hours
   • Notifies when complete
4. Deploy Avatar:
   • Assign to Elena agent
   • Test and verify

Requirements:

• Consent from talent
• High-quality video recording
• Professional setup

---

Usage Examples

Generate Avatar Video

from azure.identity import DefaultAzureCredential
from azure.ai.projects import AIProjectClient

client = AIProjectClient(
    endpoint="https://zimax.services.ai.azure.com/api/projects/zimax",
    credential=DefaultAzureCredential(),
)

agent = client.agents.get(agent_name="Elena")
openai_client = client.get_openai_client()

# Get response with avatar
response = openai_client.responses.create(
    input=[{"role": "user", "content": "Tell me about the GTM strategy"}],
    extra_body={
        "agent": {"name": agent.name, "type": "agent_reference"},
        "avatar": {
            "enabled": True,
            "resolution": "1080p",
            "emotion": "neutral"
        }
    },
)

# Response includes avatar video URL
avatar_video_url = response.avatar_video_url
print(f"Avatar video: {avatar_video_url}")

Display Avatar in Frontend

// When Elena responds, display avatar video
if (response.avatar_video_url) {
  setAvatarVideoUrl(response.avatar_video_url);
  setIsSpeaking(true);
}

// In component
{avatarVideoUrl && (
  <video
    src={avatarVideoUrl}
    autoPlay
    loop={false}
    className="elena-avatar-video"
  />
)}

---

Cost Considerations

Avatar Generation Costs

• Standard Avatar: Lower cost, pre-built
• Custom Avatar: Higher cost, requires training
• Resolution: 4K costs more than 1080p
• Usage: Pay per video generated

Optimization Tips

1. Use Standard Avatar: Lower cost than custom
2. 1080p Resolution: Good balance of quality and cost
3. Cache Videos: Reuse avatar videos for common responses
4. Batch Generation: Generate multiple videos at once

---

Testing

Test Avatar Generation

# Test avatar in Azure Portal
# Azure AI Foundry → Project 'zimax' → Applications → 'Elena' → Test

# Or use REST API
curl -X POST \
  "https://zimax.services.ai.azure.com/api/projects/zimax/applications/Elena/protocols/openai/responses?api-version=2025-11-15-preview" \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "input": [{"role": "user", "content": "Hello Elena!"}],
    "extra_body": {
      "agent": {"name": "Elena", "type": "agent_reference"},
      "avatar": {"enabled": true}
    }
  }'

Verify Avatar Quality

1. Check video resolution
2. Verify lip-sync accuracy
3. Test emotion expressions
4. Validate background transparency

---

Next Steps

1. Run Configuration Script:

   python3 scripts/configure_elena_avatar.py
   

2. Test Avatar:
   • Use Azure Portal test interface
   • Verify video generation
   • Check quality and sync
3. Integrate with Frontend:
   • Update AvatarDisplay component
   • Add video playback
   • Handle avatar URLs
4. Customize (Optional):
   • Create custom avatar
   • Adjust emotions
   • Upgrade to 4K

---

Summary

✅ Avatar Feature Available in Azure AI Foundry
✅ Configuration Script Ready (scripts/configure_elena_avatar.py)
✅ Integration Points Identified (frontend and backend)
✅ Documentation Complete

Elena can now have a photorealistic video avatar that speaks with natural voice and expressions, creating a more engaging user experience.

---

Last Updated: January 2026