What is Reactor Runtime?
Reactor Runtime is a backend framework that lets researchers deploy interactive video generation models without building infrastructure. The problem? Most researchers spend weeks building networking code, managing servers, and handling sessions—just to demo their models. Amazing models stay locked in papers because deployment is too hard. The solution? Reactor Runtime handles all of it:- Real-time video streaming - WebRTC powered by LiveKit
- Local development - Test with a web frontend instantly
- One-click deployment - Go from local to production seamlessly
- Session management - Multiple users handled automatically
- Model registry - Weights download from S3 automatically
Who is this for?
This documentation is for:- Researchers who want to demo their models interactively without building infrastructure
- Partners integrating custom models into the Reactor ecosystem
- ML Engineers converting existing models for real-time deployment
How It Works
A Reactor model consists of two essential files:Model Manifest
Configuration file that specifies your model class, initialization
arguments, required weights, and version info
VideoModel Class
Python class that wraps your model. You don’t rewrite your model—just add a
thin interface layer