Runway launches Solaris, an AI model that generates interactive app interfaces in real time without code

AI startup Runway has launched Solaris, a pioneering "Interface World Model" that generates fully interactive 720p software interfaces in real time without any underlying code. Built on Runway's Gen-4.5 video model, Solaris treats rendered pixels as the application itself, responding directly to user clicks and drags. While Runway's internal evaluations show strong user preferences over traditional code, the system faces major hurdles in text rendering, accessibility, and probabilistic determinism.

Solaris Interface World Model launch

• ▪Runway launched Solaris as a limited research preview on August 31, 2026, offering early access to organizations through an online request form.
• ▪Runway announced the launch of Solaris, an artificial intelligence model in a new category called Interface World Models, on August 31, 2026.
• ▪Solaris generates fully interactive software interfaces frame by frame in real time at 720p resolution without using underlying code like HTML, CSS, or JavaScript.

Code-free frame generation architecture

• ▪In Runway's internal evaluations, participants preferred Solaris over a coded interface generated by Claude Opus 5 in 61% of instruction-following and 71% of natural-behavior comparisons.
• ▪Solaris is built on top of Runway's Gen-4.5 video generation model, using a language model to handle reasoning and semantic prompts while the world model renders the visual frames.
• ▪Solaris processes user interactions, such as clicks, drags, and text inputs, directly as conditioning signals to generate the next visual frame in the generation loop.

Real-time interactive speed engineering

• ▪Runway trained the distilled Solaris model on its own outputs to stabilize visual quality and mitigate compounding errors over long interactive sessions.
• ▪Runway collaborated with NVIDIA on compute infrastructure to optimize real-time frame-by-frame generation speeds for Solaris.
• ▪To achieve interactive speeds, Runway retrained Gen-4.5 to generate frames autoregressively and applied diffusion distillation to compress the required denoising steps.

World Models research lineage

• ▪Runway raised $315 million in a Series E funding round in February 2026, led by General Atlantic, to expand its world model capabilities.
• ▪Runway previously announced GWM-1, its first General World Model initiative for gaming and robotics simulation, in December 2025.
• ▪The concept of world models in artificial intelligence traces back to Jürgen Schmidhuber's 1990 work on recurrent neural networks and David Ha's 2018 paper.

Shopping tutorial visualization applications

• ▪Runway proposes using Solaris to generate diverse, novel interface layouts to train AI agents, aiming to improve their robustness on unfamiliar websites.
• ▪Runway identifies online shopping and interactive tutorials as near-term use cases where Solaris can adapt layouts and visualize steps dynamically for individual users.

Text reliability deployment constraints

• ▪Generated pixel frames from Solaris do not automatically interoperate with screen readers, accessibility APIs, or traditional semantic software stacks.
• ▪Technical reviewers note that Solaris faces a determinism problem because a system generating visual behavior probabilistically cannot guarantee stable states for critical inputs.
• ▪Runway identifies stable and legible text rendering as a primary open technical challenge for Solaris, as interfaces are highly text-dependent.