What Is Runway Solaris? Real-Time UI Model

Inside Runway Solaris: Frame-by-Frame Interfaces at 720p

Runway Solaris is an Interface World Model from Runway that generates interactive software interfaces frame by frame in real time, responding to clicks, drags, and other inputs without relying on conventional UI code. Runway announced Solaris on August 31, 2026, as the first model in its Interface World Models family. The system is currently described as an early-access research project, not a generally available application, API, or open-weight model.

Key Takeaways

• Runway Solaris generates visual interfaces directly instead of producing HTML, CSS, JavaScript, or a conventional component tree.
• The model responds continuously to user actions, including clicks, drags, and typed instructions.
• Runway positions Solaris as an Interface World Model, extending world-model techniques from simulated environments to software screens.
• Available technical reporting describes a 720p output target and sub-500-millisecond per-frame latency target, but these figures are not independently verified.
• Reported evaluations include 250 participants and 7,500 judgments, with Solaris preferred in 61% of instruction-following comparisons against Claude Opus 5.
• Solaris is not generally available, and Runway has not confirmed public pricing, an API, weights, licensing terms, or a launch date.

What Is Runway Solaris?

Runway Solaris is a generative system for creating and modifying interactive interfaces as users operate them. A conventional website uses code to decide what appears after an action. Solaris instead generates the next visual frame from the current scene and the user's input.

Runway calls this category an Interface World Model. The term applies the world-model idea of predicting what happens next in an environment to software interfaces. In a physical simulation, an action might move an object or change a camera view. In Solaris, the action changes a screen, interface element, or simulated application state.

The distinction matters for AI engineering. Solaris is not primarily a chatbot that writes React code. It is also not a prerecorded video with clickable hotspots. Its described design treats the generated visual output as the application surface itself.

Runway introduced the system through its research materials on August 31, 2026. The official research page is the primary resource for the project’s positioning and access request. Runway’s Solaris research page

Runway Solaris remains early-stage. The public material describes limited testing by request, while pricing, production guarantees, and deployment details remain unpublished.

Runway Solaris at a Glance

Specification	Current information
Developer	Runway
Type	Interface World Model
Modality	Interactive visual interface generation
Output resolution	720p reported in available technical coverage
Interaction	Clicks, drags, typing, and other user inputs
Context window	Not yet confirmed
Reported latency target	Sub-500 milliseconds per frame, unconfirmed
Pricing	Not yet confirmed
Availability	Early access by request; not generally available
Public API or SDK	Not yet confirmed
Weights	Not yet confirmed
License	Not yet confirmed
Local deployment	Not yet confirmed

The 720p and sub-500-millisecond figures come from early technical reporting rather than a published model card. They should be treated as reported targets, not guaranteed service specifications.

How Runway Solaris Works / What Makes It Different

Solaris’s core change is the removal of the usual interface translation layer. In a standard software workflow, a designer creates a visual concept, then engineers encode it into markup, styles, components, event handlers, and application logic. A browser or application runtime renders the result.

Solaris is described as using Frame-by-Frame Rendering instead. Each new frame is generated as a visual continuation of the previous frame. The interface changes because the model predicts how the scene should respond to an action, not because a predefined event handler selects a coded state.

Runway’s broader world-model work provides the conceptual foundation. Available reporting connects Solaris to GWM-1 and Gen-4.5, with the video-generation machinery adapted for interactive software. GWM-1 is described as an autoregressive general world model that generates sequential frames and responds to controls.

The reported architecture contains three relevant ideas:

1. Autoregressive Generation: each frame depends on the preceding interaction history and visual state.
2. Step Distillation: a longer denoising process is compressed into fewer generation steps to support interactive speed.
3. Self-Training: the fast model is trained on its own outputs to help preserve visual quality across longer sessions.

These details come from early technical accounts and are not accompanied by a public architecture paper in the supplied material. The practical objective is clear: reduce the time between an input and the next rendered frame.

Solaris also introduces Interaction Conditioning. Clicks, drags, and other controls act as signals that condition the next frame. A user dragging an object is therefore not merely selecting a coded element. The action becomes part of the model input that determines the next visual output.

One early description calls this idea Redefining the Mouse. The phrase captures the product theory: a mouse gesture can become a physical action inside a generated environment rather than a command routed through a fixed menu system.

Runway Solaris treats the generated frame as the interface rather than as a preview of code.

This approach could preserve visual fidelity because it avoids converting a design into an intermediate representation. It also creates serious engineering tradeoffs. Pixels are difficult to inspect, diff, test, index, secure, and connect to existing software systems. A generated interface can look convincing while displaying incorrect text, changing inconsistently, or failing to expose accessibility information.

What You Can Do With Runway Solaris

The demonstrated and discussed use cases focus on interfaces that benefit from direct manipulation.

• Virtual storefronts: A user could enter a visual showroom, select clothing, and drag a garment onto an image of themselves.
• Room design: A person could move furniture or change a sofa’s color through natural-language instructions and direct manipulation.
• Interactive tutorials: A generated environment could demonstrate a scientific process, such as combustion, while allowing the learner to experiment.
• Visual cooking workflows: Ingredients could be dragged into a bowl to create a salad rather than added through fixed buttons.
• Agent training: Agents could learn to operate changing interfaces instead of memorizing static screen locations.
• Explorable scenes: Early community demonstrations showed objects changing in response to gestures. One post described enlarging a tree by dragging upward, enlarging a mural by dragging diagonally downward, moving a lamp, and opening a style-selection UI after a click. These are community observations, not controlled tests. Example discussion of Solaris interactions

The most promising applications are prototypes, simulations, tutorials, and highly visual experiences. Production systems that require deterministic billing, authentication, permissions, database writes, or legal records still depend on conventional software infrastructure.

How Runway Solaris Compares

The closest comparison in the available material is Claude Opus 5 used to create an interface through a code-based workflow. Read the reference page on Claude Opus 5

Approach	What happens after interaction?	Main strength	Main open risk
Runway Solaris	The model generates the next visual frame	Direct visual adaptation and rapid prototyping	Latency, consistency, text accuracy, and integration
Claude Opus 5 in a coded workflow	Generated code is executed by conventional software tools	Inspectable logic and established integrations	Design-to-code translation and implementation effort
Traditional UI engineering	Prewritten application logic renders a known state	Determinism, testing, accessibility hooks, and system integration	More work for unusual or highly visual interactions

Early reporting claims that Solaris outperformed GPT-4o, Gemini 2.5 Pro, Claude Opus 5, and Fable 5 on interface reconstruction using structural-similarity and visual-feature metrics. Those claims are not independently verified in the supplied material.

A separate reported user study involved 250 participants and 7,500 judgments. Solaris was preferred to Claude Opus 5 in 61% of instruction-following comparisons. A Threads post gives a different figure, saying 72% of participants preferred Solaris for natural interaction. The tests may measure different tasks, and neither figure has a published independent methodology in the available sources.

Runway Solaris is best understood as a visual interaction prototype, not a confirmed replacement for production software.

Availability: How to Access Runway Solaris

Runway Solaris is not generally available. The current path is an early-access request through Runway’s official research materials. The request form is not documented as a self-serve trial, and submitting it does not guarantee access.

Runway has not published a public Solaris endpoint, model identifier, SDK, authentication method, rate limit, billing rule, or hardware requirement. It has also not published a general launch date or stated whether access will require a particular Runway subscription.

For a currently listed visual-generation option while Solaris access remains limited, Runway Video Generation is a separate catalog entry and should not be treated as Solaris access.

The practical access checklist is straightforward:

1. Use Runway’s official research page rather than an unofficial access offer.
2. Describe the interface, simulation, or agent-training problem to be tested.
3. Treat approval as research access unless Runway publishes commercial terms.
4. Do not assume that an existing Runway plan includes Solaris.
5. Do not budget for production use until pricing, quotas, and service conditions are public.

What We Don’t Know Yet

Important technical and commercial questions remain open:

• What is Solaris’s exact model architecture and training data composition?
• Does it maintain a persistent state representation across long sessions?
• What is the measured median and tail latency under real workloads?
• How does quality change at different resolutions or interaction rates?
• Can it render reliable text, tables, forms, and dense dashboards?
• Does it expose accessibility semantics, keyboard navigation, or an accessibility API?
• Can generated interfaces call external APIs or write to databases safely?
• How are authentication, permissions, payments, and private data handled?
• Is the model available through an API, browser, SDK, or local package?
• Will Runway provide weights, a license, or any form of customer-controlled deployment?
• How much does each interactive session cost to operate?
• Can generated scenes be cached or exported into conventional software?

These questions define whether Solaris becomes a practical application layer or remains a research demonstration. The recurring limitations in early coverage are text rendering, factual grounding, session coherence, accessibility integration, latency, and operating cost.

Frequently Asked Questions

What is Runway Solaris?

Runway Solaris is an Interface World Model that generates interactive software interfaces frame by frame in response to clicks, drags, and other user inputs. Runway announced it as an early-access research system rather than a generally available product.

Is Runway Solaris open source?

Runway Solaris is not confirmed to be open source. Runway has not published weights, a license, or a local deployment package in the materials currently available.

How does Runway Solaris work?

Runway Solaris generates the next visual interface frame from the preceding scene and the user’s interaction instead of asking a browser to execute conventional HTML, CSS, or JavaScript. Available technical reporting describes Autoregressive Generation, Step Distillation, and Self-Training as parts of the approach, but a public model card has not been provided.

How much does Runway Solaris cost?

Runway Solaris has no publicly confirmed price. Runway has not published subscription rates, API charges, quotas, or usage-based billing for Solaris.

Can I use Runway Solaris?

Runway Solaris is available only through an early-access request process, and general public access is not confirmed. Runway has not published a public API, SDK, launch date, or guaranteed invitation policy.

Is Runway Solaris an AI video model?

Runway Solaris uses video-model-style sequential frame generation, but its stated purpose is interactive interface generation rather than fixed video creation. The system responds to user actions while generating the next frame, which makes it closer to a visual software environment than a conventional video editor.

Runway Solaris vs Claude Opus 5?

Runway Solaris generates interface pixels directly, while Claude Opus 5 was used in reported comparisons to create interfaces through a code-based workflow. Solaris may reduce the design-to-code translation step, but its reliability, integrations, accessibility, and operating cost remain unconfirmed.

What to watch next: Runway’s next access update, a public model card or API specification, and independent tests of latency, text accuracy, session consistency, and operating cost. Pricing and deployment terms will determine whether Solaris can move beyond demonstrations into real software workflows.

Building similar interactive visual experiences? On kie.ai you can try Runway Video Generation, Gemini Omni 1.1 Flash, and Wan 3.0 Video.