> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tavus.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Models

## Phoenix: Face Rendering Model

Phoenix is built on a Gaussian diffusion model that generates **lifelike digital faces with natural facial movements, micro-expressions, and real-time emotional responses**.

### Key Features

<CardGroup cols={2}>
  <Card title="Full-Face Animation" icon="face-smile">
    Dynamically generates full-face expressions, micro-movements, and emotional shifts in real time.
  </Card>

  <Card title="True Realism" icon="stars">
    Achieves the highest fidelity by rendering with pristine identity preservation.
  </Card>

  <Card title="Driven Emotion" icon="masks-theater">
    Adjusts expressions based on context, tone, and conversational cues.
  </Card>
</CardGroup>

## Raven: Perception Model

Raven is the first contextual perception system that **enables machines to see, hear, reason, and understand like humans in real-time**, interpreting emotions, speaking tone, body language, and environmental context to enhance conversation.

### Key Features

<CardGroup cols={2}>
  <Card title="Emotional Intelligence" icon="face-smile">
    Interprets emotion, intent, and expression from both visual cues and vocal tone - detecting sarcasm, frustration, excitement, and more.
  </Card>

  <Card title="Ambient Awareness" icon="gear">
    Continuously analyzes visual and audio streams to detect presence, environmental changes, and user state in real-time.
  </Card>

  <Card title="Callout Key Events" icon="eye">
    Monitors for specified gestures, objects, behaviors, or audio cues (like tone shifts) and triggers functions automatically.
  </Card>

  <Card title="Multi-channel Processing" icon="face-viewfinder">
    Processes camera feeds, user audio, and screen share so Raven has full context for the call.
  </Card>
</CardGroup>

### Screen share

When Raven is enabled, participants can share their screen in the conversation and the PAL can see that shared content - not only the camera feed. No extra setup is required beyond using Raven. See [Screen Share](/sections/conversational-video-interface/pal/screen-share) for how it works and [Perception](/sections/conversational-video-interface/pal/perception) for layer configuration.

## Sparrow: Conversational Turn-Taking Model

Sparrow is a transformer-based model built for **dynamic, natural conversations**, reading tone, rhythm, and subtle cues to decide when to listen, wait, or speak. It is designed to reduce PAL interruptions, handle user backchannels and interruptions gracefully, respond quickly, and maintain conversational flow in the presence of background noise.

See [Conversational Flow](/sections/conversational-video-interface/pal/conversational-flow#1-turn_detection_model) for configuration.

### Key Features

<CardGroup cols={2}>
  <Card title="Conversational Awareness" icon="waveform-lines">
    Understands meaning, tone, and timing to respond naturally like a human.
  </Card>

  <Card title="Turn Sensitivity" icon="comments">
    Understands human speech rhythm, capturing cues and pauses for natural interactions.
  </Card>

  <Card title="Heuristics & ML" icon="chart-network">
    Adapts to speaking styles and conversation patterns using heuristics and machine learning.
  </Card>

  <Card title="Optimized Latency" icon="rocket-launch">
    Delivers ultra-fast response times for seamless real-time conversation.
  </Card>
</CardGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.