Runtime Overview

• Real-time streaming: frames reach clients over WebRTC as they are generated, not after the whole video is done.
• Live interaction: clients change inputs mid-generation. No restart, no re-queue.
• No transport code: you never import a WebRTC library, manage a WebSocket, or encode video.
• Typed, validated inputs: declare the commands your model accepts with types and constraints, and the runtime validates every payload before your handler runs.

​

The simplest model

from pathlib import Path

from reactor_runtime import InputField, Output, ReactorModel, Video, event


class MyOutput(Output):
    main_video: Video


class MyModel(ReactorModel):
    def load(self, config_path: Path | None) -> None:
        self.pipe = load_my_model()
        self.prompt = "a sunny meadow"

    @event(name="set_prompt", description="Scene the model renders")
    async def set_prompt(self, prompt: str = InputField(default="a sunny meadow")) -> None:
        self.prompt = prompt

    async def run(self) -> None:
        while True:
            await self.connected.wait()
            while self.connected.is_set():
                frame = self.pipe.forward(prompt=self.prompt)
                await self.emit(MyOutput(main_video=frame))

​

How the pieces fit

• Tracks are the named media channels the model reads and writes. MyOutput.main_video above is an outbound video track; declare an Input the same way to receive a client’s webcam or microphone.
• run() is your generation loop. It decides what to produce and calls emit() to hand each result to the transport, which throttles the loop to the rate clients are playing it back at.
• Commands are the @event handlers clients call to change what the model is doing. The runtime validates each payload against the handler’s signature before invoking it.
• Messages are the typed payloads your model sends back — progress, state snapshots, or a direct reply to a command.

​

From model to production

​

Next