Runway wants a cinematic desktop. For now it lives in a demo reel
Runway Labs showed Project Continuum, a research desktop where some windows are generated video that reshapes itself as you click and resize. It runs on Runway's Solaris model, but the company has published no latency, frame rate or access date, and its own notes say legible text and long sessions are unsolved.

Halfway through Runway's five-minute walkthrough of Project Continuum, a cross-section of a leaf sits in a window on an ordinary-looking desktop. Someone drags the window narrower, and instead of squashing the picture, the scene rearranges itself to fit, with the labels for chloroplast, stoma, CO2 in and O2 out still neatly in place. Runway says that window is not an app drawing a diagram. It says the leaf is video, generated on the spot by a model that watches the cursor and decides what the next frame should look like.
Runway posted the demo on X on Thursday afternoon, calling Continuum "a new operating system research application built around real-time video interfaces" from Runway Labs, its internal incubator. The company's own verb was "sharing an early look", and that is the right register for it. There is no blog post, no paper, no product page, no waitlist and no date. What exists is a 52-second sizzle reel, the 5:22 walkthrough on the Runway Labs account and a short thread, and it is a genuinely interesting look at where Runway thinks screens are going.
Four ideas, one model underneath
The thread names four modes. Portals are animated overlays that appear while you browse, so a NASA article about fog spawns a moving view of the Golden Gate, with "soft, dream-like edges that mark where generated content begins". Responsive Video Interfaces are the leaf trick: resize a region and "the model recomposes the scene with spatial awareness". Interactive Worlds promise to "expand any media into an interactive real-time world simulation" for "people and agents", shown as a Mars habitat with rovers. Visual Thinking is pitched as a harness for coding agents that draws each step and tool call as it happens, and in the walkthrough a chat agent asked for a 3D-printable mount answers with a rendered part beside its text.
All of it rests on Solaris, which Runway introduced on Monday 31 August as an "Interface World Model". Solaris is built on Gen-4.5, the same video model Runway sells to filmmakers, and treats clicks and drags as conditioning for the next frame, the way a video model treats a text prompt. A language model decides what should happen and prompts Solaris, which renders it frame by frame. Runway says it built Solaris for real-time interaction, coherence across a whole session and "visual quality that holds at 720p".
The evidence Runway offers for the approach is a user study it ran itself: 250 participants, 30 interaction examples and nearly 7,500 head-to-head judgments comparing Solaris with an interface coded by Claude Opus 5. People preferred Solaris for following the instruction in 61% of comparisons against 24%, and for behaving naturally in 71% against 21%. It is a vendor study on a small set of examples and nobody has replicated it, but it is more than most interface demos come with.
What the videos do not show
Runway has published no latency, frame rate, resolution or hardware figure for Continuum. A widely shared rewrite of the launch says "technical reporting cites" a 720p target and per-frame latency below 500 milliseconds. The Solaris post says something different: it names 720p as a quality goal and notes that interactions "stop feeling interactive somewhere around half a second of delay". That is a threshold Runway is aiming at, not a number it has reported, so the honest answer to how fast Continuum runs is that nobody outside Runway knows.
The videos are also edited, which is normal for a launch and worth saying plainly. About half of the sizzle reel is white title cards ("What if your operating system was cinematic"), and the footer reads v0.0.1-ALPHA. In the walkthrough, the menu-bar clock shows 3:28 PM at the start, at 1:40 and at 5:00. Either the session was cut together or the desktop chrome around those regions is a static mock-up. Neither means the demo is fake, though it does mean the footage cannot prove the regions run live and unbroken for five minutes.
Then there are those leaf labels. In the Solaris post, Runway lists "stable, legible text" as one of the hardest unsolved problems in video generation and suggests a hybrid in which image models render text-heavy views during pauses. Crisp, steady captions inside a region Runway calls generated video could be that hybrid, a still or an overlay. We cannot tell from the footage, and Runway has not said.
Two more items from Runway's own list matter here. On trust, the company writes that "a convincing wrong answer is worse than no answer". On long sessions, it says coherence over extended use "remains an active area of research". An operating system is a long session by definition. That is why Continuum reads best as what Runway calls it, a research shell where selected regions are generated while normal software keeps running around them, rather than an OS anyone could install.
Not the first generated screen
Generated interfaces have a short history already. NeuralOS, an academic project posted in July 2025, renders an Ubuntu desktop frame by frame from mouse and keyboard input, and Runway's own Solaris paper cites it. Google DeepMind says Genie 3 holds a world together for minutes at 24 frames a second in 720p. What Continuum adds is the mix: generated regions dropped into an otherwise conventional desktop, each one doing a specific job, which is a more practical idea than generating the whole screen.
Why a film-tools company is building a desktop
Continuum is the third thing Runway has shown in about a month that is not aimed squarely at filmmakers. Solaris arrived on 31 August, and on Wednesday 30 September Runway launched Runway Ads, which it says "runs and optimizes the creative side of paid advertising programs end to end". Runway Labs itself was set up in March to explore "the next generation of applications for generative video", led by co-founder Alejandro Matamala Ortiz. None of this was hidden. When Runway raised a $315 million Series E in February, led by General Atlantic with NVIDIA, Adobe Ventures and AMD Ventures, it said the money would go into world models for new products and industries.
Runway still sells Gen-4.5 and its character tools to creators, so reading this as Runway leaving film overstates it. A fairer reading, and it is our inference, is that Runway is building a world-model company, with film as one of several customers paying for the same research.
What it means if you make work
In practice, for anyone making work this week, nothing changes, because there is nothing to use: Continuum has no access, and Solaris and its predecessor GWM-1 are request-only research previews. What is worth watching is Interactive Worlds. If a vendor can turn a still, a frame or a thumbnail into a scene you can walk through in real time, that is live previs, where a director blocks a shot inside a generated set instead of waiting minutes for a clip. That is a prediction, labeled as one, and it depends on the speed and long-session problems Runway itself says are open.
The useful habit now is to keep your look in references Runway's tools already take, meaning the stills, style frames and start frames that Solaris uses to anchor a scene. Runway's answer to its own trust problem is grounding generation in real reference imagery, so the creators with a clean, consistent reference library will be the ones ready to step inside their own worlds if this ships.
The questions for Runway are simple ones. How fast is it, on what hardware, and are those labels generated? Until the company publishes a frame rate or opens access, Continuum is a well-made film about a computer, which is a fitting thing for a video company to have shipped.
Timeline
- Date: Thu 11 Dec 2025. What happened: Runway releases GWM-1, its general world model, built on Gen-4.5; Source: Runway research post
- Date: Tue 10 Feb 2026. What happened: $315M Series E led by General Atlantic, reported $5.3B valuation; Source: Runway news; TechCrunch
- Date: Wed 11 Mar 2026. What happened: Runway Labs incubator announced, led by Alejandro Matamala Ortiz; Source: Runway news
- Date: Mon 31 Aug 2026. What happened: Solaris introduced as an Interface World Model; paper on arXiv; Source: Runway news; arXiv 2609.00776
- Date: Wed 30 Sep 2026. What happened: Runway Ads launches; AI Summit attendees in San Francisco see an early Continuum preview; Source: Runway news; Runway staff post on X
- Date: Thu 1 Oct 2026, 15:44 UTC. What happened: Project Continuum sizzle reel and 5:22 walkthrough posted; Source: @runwayml and @runwayml_labs on X