Skip to main content
Real-time models provide low-latency video streaming using WebRTC. These models are optimized for interactive applications requiring immediate feedback.

​
Available Models

mirage

Real-time video restyling model

mirage_v2

Real-time video restyling model (v2)

lucy_v2v_720p_rt

Real-time video editing model

lucy_2_rt

Real-time video editing with reference image support

live_avatar

Real-time avatar generation

​
Model Specifications

ModelFPSResolutionDescription
mirage251280x704Real-time video restyling
mirage_v2221280x704Real-time video restyling (v2)
lucy_v2v_720p_rt251280x704Real-time video editing
lucy_2_rt201280x720Real-time video editing with reference image
live_avatar251280x720Real-time avatar generation

​
Usage

​
Accessing Models

Use the models.realtime() factory function to access real-time model definitions: 
import { models } from '@decartai/sdk';

const mirageModel = models.realtime('mirage');
const mirageV2Model = models.realtime('mirage_v2');
const lucyV2VModel = models.realtime('lucy_v2v_720p_rt');
const lucy2Model = models.realtime('lucy_2_rt');
const avatarModel = models.realtime('live_avatar');

​
Model Definition Properties

Each model definition includes: 
interface ModelDefinition {
  name: string;           // Model identifier
  urlPath: string;        // API endpoint path
  fps: number;            // Frames per second
  width: number;          // Video width in pixels
  height: number;         // Video height in pixels
  inputSchema: ZodObject; // Input validation schema
}

​
Creating a Real-time Client

Real-time models require the createRealtimeClient for WebRTC streaming: 
import { createDecartClient, models } from '@decartai/sdk';

const client = createDecartClient({
  apiKey: 'your-api-key',
});

const realtimeClient = client.realtime({
  model: models.realtime('mirage'),
});

// Listen for video frames
realtimeClient.on('frame', (frame) => {
  console.log('Received frame:', frame);
});

// Start streaming
await realtimeClient.connect();

​
Input Requirements

Real-time models accept streaming video input via WebRTC. No additional parameters are required in the input schema.

​
Connection Options

interface RealtimeOptions {
  model: ModelDefinition<RealTimeModels>;
  onFrame?: (frame: VideoFrame) => void;
  onError?: (error: Error) => void;
}

​
Examples

​
Basic Real-time Streaming

import { createDecartClient, models } from '@decartai/sdk';

const client = createDecartClient({
  apiKey: process.env.DECART_API_KEY,
});

const realtimeClient = client.realtime({
  model: models.realtime('mirage_v2'),
});

// Handle incoming frames
realtimeClient.on('frame', (frame) => {
  // Render frame to canvas or video element
  displayFrame(frame);
});

// Handle errors
realtimeClient.on('error', (error) => {
  console.error('Streaming error:', error);
});

// Connect and start streaming
await realtimeClient.connect();

// Send input frames
const stream = await navigator.mediaDevices.getUserMedia({ video: true });
realtimeClient.sendStream(stream);

// Disconnect when done
await realtimeClient.disconnect();

​
Using Lucy 2 RT with Reference Image

import { createDecartClient, models } from '@decartai/sdk';

const client = createDecartClient({
  apiKey: process.env.DECART_API_KEY,
});

const realtimeClient = client.realtime({
  model: models.realtime('lucy_2_rt'),
});

// Set reference image for styling
await realtimeClient.setReferenceImage(referenceImageFile);

// Start streaming
await realtimeClient.connect();

const stream = await navigator.mediaDevices.getUserMedia({ video: true });
realtimeClient.sendStream(stream);

​
Live Avatar Generation

import { createDecartClient, models } from '@decartai/sdk';

const client = createDecartClient({
  apiKey: process.env.DECART_API_KEY,
});

const avatarClient = client.realtime({
  model: models.realtime('live_avatar'),
});

avatarClient.on('frame', (frame) => {
  // Display avatar frame
  renderAvatar(frame);
});

await avatarClient.connect();

// Send audio and video input for avatar generation
const stream = await navigator.mediaDevices.getUserMedia({ 
  video: true, 
  audio: true 
});
avatarClient.sendStream(stream);

​
Type Definitions

type RealTimeModels = 
  | "mirage" 
  | "mirage_v2" 
  | "lucy_v2v_720p_rt" 
  | "lucy_2_rt" 
  | "live_avatar";

function models.realtime<T extends RealTimeModels>(
  model: T
): ModelDefinition<T>;

​
Performance Considerations

  • Latency: Real-time models prioritize low latency over quality
  • FPS: Different models run at different frame rates (20-25 FPS)
  • Resolution: All models output at 720p or similar resolutions
  • Network: Requires stable, low-latency network connection for WebRTC
  • Browser Support: Requires WebRTC support (modern browsers only)

​
Error Handling

realtimeClient.on('error', (error) => {
  if (error.code === 'CONNECTION_FAILED') {
    // Handle connection failure
    console.error('Failed to establish WebRTC connection');
  } else if (error.code === 'STREAM_ERROR') {
    // Handle streaming error
    console.error('Error during streaming:', error.message);
  }
});

realtimeClient.on('connectionStateChange', (state) => {
  console.log('Connection state:', state);
  // States: 'connecting', 'connected', 'disconnected', 'failed'
});

Video Models

Async video generation models

Image Models

Image generation models

Realtime Client

Real-time client API reference

WebRTC Guide

Real-time streaming guide

Build docs developers (and LLMs) love

Get started for freeTalk to us