Skip to main content
A weathered stone gateway arching over a dirt track through misty highlands
The Model API Reference is the source of truth for what each model accepts and emits: every command, every parameter, every event. For robotics policies, see Robotics: observation and action contracts, embodiment compatibility, and simulator and robot integration guides. This is distinct from the SDK Reference, which documents the JavaScript and Python client libraries (sessions, connect/disconnect, file uploads, raw event sending). Learning how to use Reactor is one thing; learning how to drive a specific model is another. For background on what a Reactor model is, see What is a Reactor model?.
Each model’s reference is hand-written here, one overview, schema, prompt guide, and tutorial per model. A daily sync flags a page when the model’s schema changes upstream, so the prose stays in step with what’s deployed.

​
Available models

FastH3

Prompt-driven video clips with joint native audio, delivered as a curated playout queue or one endless take.

HappyOyster

Permanent explorable worlds built from a prompt: play them with held controls or direct them with text instructions.

LingBot World 2

Image-anchored navigable world model with two-axis WASD driving, directed camera moves, and live prompt steering.

LingBot

Real-time navigable video model with WASD movement, look controls, and live prompt steering.

X2

Live reference-guided video transformations with character, clothing, style, and trajectory control.

Helios

Interactive real-time video generation with autoregressive chunked diffusion and image-to-video support.

LongLive-2.0

Autoregressive multi-shot video generation: direct a sequence with soft shots and hard cuts, live or scheduled.

SANA-Streaming

Real-time streaming video editing: surgical text-driven edits to uploaded clips or a live webcam feed.

Visko Orbis Stable

Per-chunk prompts morph the scene live without cutting. Image-anchored openings and picture-driven audio by default.

LTX

Turns a photo and a script into a video-and-audio take of that person speaking, lip-synced in one generation pass.

Vidu S2-Avatar

A real-time conversational character built from one photo: talk to it by voice or text, and it answers in lip-synced video.

Vidu S2-Editing

Edits your live camera from one reference image: restyle the scene, try on a garment, swap in a character, or replace the background.
Each model section splits its reference across four pages: an overview, the schema, a prompt guide, and a tutorial.

​
What you’ll find per model

  • Every command the model accepts, with parameter types and constraints
  • Every event the model emits, with payload shapes
  • Track capabilities (which media streams are publishable or receivable)
  • Any model-specific preconditions or ordering rules

​
Working without a typed SDK

If a model doesn’t have a typed SDK published yet, or you’re prototyping against a model in development, drive it from the base Reactor class using sendCommand() and raw event payloads. The Model API Reference tells you exactly what those payloads should look like.