Runway's Solaris renders interactive UIs as live video, no code
Runway's first 'interface world model' paints the UI directly, frame-by-frame — no DOM, no JS. It's 720p and real-time, but early-access-only, so no API yet.
Copy markdown
The image is the app
Solaris renders each frame of the interface as generated video, with no DOM, CSS, or JavaScript underneath — the picture you see is the running app. It's built on Runway's Gen-4.5 video model, with NVIDIA handling real-time inference at up to 720p.
Runway's numbers, still unverified
In Runway's own user study of 250 people across 30 examples, testers preferred Solaris 61% to 24% over a coded UI on instruction-following, and 71% to 21% on 'natural' behavior. No third party has benchmarked it yet, so treat the win as a vendor claim.
You can't build on it yet
There's no API, no weights, and no pricing — just an early-access waitlist while Runway 'works with key partners to launch publicly.' File this as a shift in how UIs might get made, not a weekend project.
Where the demo ends
Runway itself flags the unsolved parts: legible text rendering, factual grounding you can trust, coherence across long sessions, and accessibility-API support. That's the gap between a striking clip and a UI you'd hand to real users.