What if your app interface was generated frame by frame?

The research

A September 1 research paper from Runway introduces Solaris, an interface world model. The authors describe generating an interactive screen frame by frame in response to mouse actions, with a language model helping interpret intent. Their proposal changes both the appearance and behavior of the interface dynamically, rather than relying only on predefined screen states. This is a research claim from the authors, not evidence that ordinary production apps can safely replace their code with generated video. The paper’s public abstract describes a possible direction for software, rather than a general consumer launch.

Source: Runway researchers: Solaris, arXiv paper ↗

Our practical take

The fun question is what a screen could do if it adapted around your actual task. A beginner might need labels and examples; someone repeating a familiar job might want fewer controls. But visual flexibility brings a design challenge: people build confidence by learning where things live and what clicks do. If a button changes meaning or location unpredictably, a beautiful interface can become tiring to use. Adaptation should help someone make progress, and that requires more than generating a plausible next frame.

Borrow the question today

Our suggested exercise needs no access to Solaris. Pick an app you use regularly and write down its three most common tasks. Sketch a beginner view and a frequent-user view using the same underlying information. Keep critical actions in predictable places and explain any changes. Ask a friend to complete the same task on both sketches, then notice where they hesitate. This gives you a practical way to explore adaptive interfaces without confusing a research demonstration with a working service. The most interesting future screen may be the one that removes a small frustration reliably, rather than changing everything each time you touch it.