Runway introduced Solaris on August 31 as its first Interface World Model: instead of emitting HTML/CSS/JS, it synthesizes the interface frame by frame in real time on a Gen-4.5 base. A language model decides what should happen next while the world model renders responses to clicks, drags, and other input; Runway targets session coherence at interactive ~720p speeds. In a 250-person study, Solaris was preferred over coded interfaces on about 61% of instruction-following and 71% of natural-behavior comparisons. Public access is request-only early access for now.
Key Takeaways
- โThe UI is generated frame by frame with no intermediate code layer
- โAn LLM reasons about behavior while Solaris renders interactive responses in real time
- โPreferred over coded UIs in ~61% instruction and ~71% naturalness judgments; early-access requests open
Discussion & Comments
0Sign in to join the discussion
Connect with AI developers to exchange benchmark insights.