Skip to main content
Visko Orbis is a real-time steerable video generation model. It produces a continuous stream of video you can control mid-run by sending a new prompt; the picture morphs into your description at the next chunk boundary instead of a hard cut. Start a scene from text alone, optionally anchor the opening frame to a reference image, and let the model generate sound in realtime on a dedicated audio track. Visko-Orbis streams in 33-frame chunks (1.833 seconds), every prompt lands at the next chunk boundary. This model is best for continuous scenes evolving inside one uninterrupted shot, image-anchored openings that hold composition, and runs that want sound generated from the picture without a separate sound generator. The Visko Orbis reference is split across four pages: this overview, the complete command and event schema, the prompt guide for writing prompts the model renders best, and an end-to-end tutorial against the open-source reference frontend. The base wire protocol is the same as every other Reactor model: open a session with the Reactor class, send named commands, and receive events. Visko Orbis has two versions of the same model: reactor/visko-orbis-stable and reactor/visko-orbis-dynamic. See Typed Model SDKs for more information.

At a glance

Key features

Delivery resolutions up to 4k

Pick a delivery tier at any time — 1080p, 2k, or 4k — and main_video upscales the 832×480 raster the model generates. Render options from state.available_resolutions; never hard-code.

Image-anchored openings

Upload a reference image before start to anchor generation to your chosen frame.

Picture-driven audio by default

main_audio streams alongside video; the model generates sound from the picture in realtime.
Beyond these, steer mid-run by sending a new prompt, seed runs for reproducibility, and toggle sound compute at start time. The prompt guide covers how to write the continuous takes that make the morph read as cinematography.

Quick start

The fastest path to a working Visko Orbis Stable app is the create-reactor-app CLI, which scaffolds the reference frontend into a runnable project. You can also clone the reference frontend itself, or follow the tutorial for a guided walkthrough.
Working in Python instead? The CLI is JavaScript-only, so install the SDK with pip install reactor-sdk and follow the Python example below (see the quickstart for the full walkthrough):
The above JS example uses the base SDK. The typed SDK, @reactor-models/visko-orbis-stable, gives you named methods and React hooks, which is discussed in the tutorial. To learn more about Reactor’s typed SDK’s, see Typed Model SDKs.

How it works

On connect the model is live but idle; it won’t produce frames until it has a prompt and you call start. To get your first stream:
  1. Connect to the model. The connection moves disconnected → connecting → waiting → ready.
  2. Optionally set a reference image with set_image.
  3. Set a prompt with set_prompt.
  4. Start generation. start will fail with a command_error until a prompt is set.
  5. Steer mid-run by calling set_prompt again. The picture morphs at the next chunk boundary (~1.8 s) without restarting. Every steering command applies at the next chunk boundary.
  6. Pause / resume / reset. generation_complete (or pause) returns the session to WAITING. The model does NOT auto-restart, so call start again with the same conditions, or reset to clear the prompt and image.