Runway

What Runway is

Runway describes itself as building foundational Real-World Intelligence, and says its models are used by brands, companies, agencies and studios^[1]. Its site reports more than 60 million creative users^[1]. The company sells three platforms built on the same models: Runway Creative, a developer platform, and robotics tooling^[1]. Runway Creative is a cloud workspace for generating and editing video, images and audio in one place^[1].

So the company presents itself on two levels at once. On one level it is a tool maker: a browser-based studio where a team can produce and cut audiovisual material without leaving the same workspace^[1]. On another level it is a research lab, and the products are meant to be applications of that research rather than separate lines of business^[1].

Research direction

Runway's research hub states that the goal is general-purpose multimodal simulators of the world, with video as the main input and output modality, supplemented by text and audio^[2]. Within that program, GWM-1 is described as a General World Model built to interact with the real world^[2]. Gen-4.5 is presented as the company's strongest video model for motion quality, prompt adherence and visual fidelity^[2].

A world model in this sense is not quite the same thing as a clip generator. A generator takes a description and returns a finished piece of video. A world model has to accept an action — a text instruction, a camera movement, a mouse event — and produce the next moment of a scene in response, again and again, without the scene falling apart. The research page's framing, with video as the primary modality and text and audio alongside it, follows from that: the system has to keep a world going across several senses at once^[2].

Real-time and interactive systems

GWM Worlds 2, published 3 September 2026, turns video and audio generation into real-time interactive simulation steered by free-form text actions and continuous camera motion^[2]. It generates continuous 720p video at 24 fps and audio at 48,000 Hz that responds to user input^[2].

Solaris, published 1 October 2026, is an interface world model: it generates an interactive UI frame by frame in response to mouse interactions^[2].

Runway Characters, published 19 July 2026, turns a single reference image into a real-time conversational video agent at 24 fps in HD^[2]. It uses 37 milliseconds of model time per frame and takes 1.75 seconds server-side from the end of the user's speech to the start of the character's response^[2].

Where it sits

Because the three product lines rest on the same models, one body of research surfaces in a creator's editing workspace, in a developer's interface, and in robotics tooling^[1]. The published systems each take a different part of the stated goal: GWM Worlds 2 pushes at continuous, steerable audiovisual simulation^[2]; Solaris treats the interface itself as something that can be generated^[2]; Runway Characters turns a still image into a talking video agent^[2].

The numbers attached to these systems are mostly about responsiveness rather than resolution. Continuous 720p at 24 fps and 48,000 Hz audio describe a stream that keeps running while a person acts on it^[2]. The frame-level and response-level figures for Runway Characters — 37 milliseconds of model time per frame, 1.75 seconds from the end of speech to the start of a reply — describe the same concern from the user's side, where a pause of a couple of seconds is the difference between a conversation and a recording^[2]. Read together, the products and the research describe a company aiming at simulation that can be interacted with, using the creative tools as the place where that capability is first used and paid for^[1].