Runway Labs demos video that redesigns itself to fit the frame

Runway Labs posted seven demos of generated video on September 21, and by the next morning the first clip had drawn 129,643 views from an account with 1,310 followers. The clips do something generated video has never done: they reflow when the window changes shape.

The thread's own summary of the trick is one sentence. The model treats window size as another input, the same way it treats a click or a drag. Resize the frame and it renders a new composition for the space it has, not a scaled version of the old one.

Runway Labs' first demo. The frame narrows and the ski run is recomposed for the portrait space, with the chairlift and the skiers still inside it. Video: Runway Labs on X.

That is a different claim from the one the video generation business has been built on. A generated clip is normally an artifact: you prompt, you wait, you get a file. A responsive one is a participant in the layout.

What Is a Responsive Generative Video Interface?

Runway Labs is not a separate company. It is Runway's internal incubator, announced on March 11, 2026, led by co-founder and Chief Innovation Officer Alejandro Matamala Ortiz, with a remit to prototype applications at the frontier of generative video and General World Models. The thread is a prototype demo, not a product launch, and there is nothing to sign up for at the end of it.

Six clips make the argument in six registers. A pack of hot air balloons is recomposed so every balloon survives the narrow frame. A furnished room collapses into a stack of furniture and boxes in a portrait window, then reassembles when the frame widens.

/media/2026/09/390227a6-runway-labs-room-wide-vs-narrow.jpg

A furnished room in a wide frame, and the same furniture packed into a narrow one. Still: Runway Labs on X.

A clothesline regroups. An engineering cross-section of a hydroelectric dam re-lays out its labels and its water level rather than being cropped.

/media/2026/09/9d812912-runway-labs-dam-wide-vs-narrow.jpg

A dam cross-section re-laid out for a portrait frame. Still: Runway Labs on X.

The clip that carries the argument is the last. An annotated architectural section of a staircase is redrawn as a spiral staircase when the frame turns portrait.

/media/2026/09/bfddf79b-runway-labs-straight-vs-spiral-stairs.jpg

The same stair section in two frame shapes. Still: Runway Labs on X.

Video: Runway Labs on X.

Straight stairs becoming a spiral staircase to fit a narrow view, in the thread's own words. That is not resizing and it is not cropping. The model redesigned the object.

Reaction landed on one word. "responsive video" wrote Josh Pigford, in a post that had 20,902 views by the next morning. "UX is changing for better, lot of possibilities this will be unlocking!" wrote AJ. Runway's co-founder and co-CEO, Cristóbal Valenzuela, posted two words of his own: "Generative video interfaces."

The Catch

Runway has already published where this runs out. On Solaris, the Interface World Model it showed on August 31, the company keeps a section headed What It Can't Do Yet, and it lists four problems. Text: "stable, legible text remains one of the hardest problems in video generation, yet interfaces depend on it more than almost any other visual domain." Trust: "a convincing wrong answer is worse than no answer", with grounding generation in verified context as a session unfolds described as "an active research focus". Long sessions: coherence "over extended, open-ended interactions remains an active area of research." Accessibility and integration: a generated interface "still needs to work inside the rest of the software stack, including assistive technologies such as screen readers and accessibility APIs". Elsewhere on the same page Runway is franker about the bill, raising cost among the engineering challenges: "generating every frame is still more expensive than serving a page that was built once", though it says the work that made Solaris real time also made it "orders of magnitude cheaper" than standard video diffusion.

There is no document object model underneath any of it. A generated button is a picture of a button, and a picture of a button is not a button.

Runway's own research page on real-time video generation argues the direction anyway, in language closer to a manifesto: "We think that era is ending. The future of online interaction is real-time video, generated on the fly - responsive, personalized, alive."

The thread is at https://x.com/runwayml_labs/status/2102035097436455021. Solaris early access is a form, with no date, no price and no product attached.