--- title: Runway’s WorldPrompt and the Engineering of Real-Time Worlds description: Runway's CTO Kamil Sindi and Principal Research Scientist, along with co-founder Anastasis Germanidis, explain GWM Worlds 2, a research preview that turns... canonical: https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Runway’s WorldPrompt and the Engineering of Real-Time Worlds | daily.dev og:description: Runway's CTO Kamil Sindi and Principal Research Scientist, along with co-founder Anastasis Germanidis, explain GWM Worlds 2, a research preview that turns... og:url: https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz og:image: https://api.daily.dev/og/posts/V5l0290bz.png og:image:alt: Runway’s WorldPrompt and the Engineering of Real-Time Worlds og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Runway’s WorldPrompt and the Engineering of Real-Time Worlds **[Latent Space](https://daily.dev/sources/latentspace)** · 94 min read · 0 upvotes · 0 comments ## Summary Runway's CTO Kamil Sindi and Principal Research Scientist, along with co-founder Anastasis Germanidis, explain GWM Worlds 2, a research preview that turns video and audio generation into real-time interactive simulation via an autoregressive diffusion approach. The centerpiece is WorldPrompt, a prompting mechanism (not a scripting language) that fixes elements like the first frame while allowing timestamped, real-time actions to steer characters, cameras, and environments. The model streams 720p video at 24fps with 48kHz audio, achieved by fine-tuning a base video model to the WorldPrompt format, post-training for autoregressive generation, and distillation for real-time speed. Key engineering challenges discussed include error accumulation in autoregressive generation, managing GPU memory for infinite context, imperfect long-term memory, and achieving 'counterfactual' realism so different user actions produce equally plausible outcomes. Beyond gaming, use cases include robotics simulation and testing AI agents at scale via synthetic environments. The piece also includes a long-form podcast transcript covering Runway's history from Green Screen rotoscoping tools through Gen-1, Gen-2, the Stable Diffusion controversy, and Gen-3's scramble to match OpenAI's Sora. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is WorldPrompt in Runway's GWM Worlds 2? WorldPrompt is a prompting mechanism, not a programming language, that lets users fix aspects of a generated world (including the first frame) and specify timestamped events or actions, even in real time. It works like a control layer for characters, cameras, and environments, similar to controlling NPCs in a video game, but without scripting or explicit state control. _Anyone tracking real-time generative video tooling can follow how prompting interfaces like this evolve on daily.dev._ ### What resolution and frame rate does Runway's GWM Worlds 2 generate in real time? GWM Worlds 2 streams continuous real-time interactive video at 720p resolution and 24 frames per second, with synchronized audio at 48,000 Hz. It is built by fine-tuning Runway's foundational audio-video model to the WorldPrompt format, post-training it for autoregressive generation, and then using distillation methods to make generation fast enough for real-time playback. _Engineers comparing real-time world model specs can keep tabs on releases like this via daily.dev._ ### What is error accumulation in autoregressive video generation models and why is it a challenge? Error accumulation happens because autoregressive models feed their own generated frames back in as input to generate subsequent frames, so small errors compound over time, degrading quality the longer generation continues. Runway's co-CEO Anastasis Germanidis called this the biggest challenge with autoregressive models, alongside managing GPU memory for infinite-length generations and the model's lack of perfect long-term memory. _Developers evaluating autoregressive generation architectures can follow this research thread on daily.dev._ ## Similar posts on daily.dev - [Runway releases its first world model, adds native audio to latest video model](https://daily.dev/posts/runway-releases-its-first-world-model-adds-native-audio-to-latest-video-model-8471ciojo) · TechCrunch · 1 upvotes · 0 comments - [Runway says its new text-to-video AI generator has ‘unprecedented’ accuracy](https://daily.dev/posts/runway-says-its-new-text-to-video-ai-generator-has-unprecedented-accuracy-acxpsywsu) · The Verge · 0 upvotes · 0 comments - [Runway started by helping filmmakers. Now it wants to beat Google at AI.](https://daily.dev/posts/runway-started-by-helping-filmmakers-now-it-wants-to-beat-google-at-ai--mvyeeepxe) · TechCrunch · 0 upvotes · 0 comments - [Runway Chooses Modal to Power Real-Time Inference for Runway Characters](https://daily.dev/posts/runway-chooses-modal-to-power-real-time-inference-for-runway-characters-alcgf5onl) · Modal · 0 upvotes · 0 comments --- Tags: [#3d](https://daily.dev/tags/3d), [#diffusion-models](https://daily.dev/tags/diffusion-models), [#world-models](https://daily.dev/tags/world-models) [View this post on daily.dev](https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Runway’s WorldPrompt and the Engineering of Real-Time Worlds","url":"https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz"},"datePublished":"2026-09-25T01:37:26.649Z","dateModified":"2026-09-25T01:37:55.398Z","description":"Runway's CTO Kamil Sindi and Principal Research Scientist, along with co-founder Anastasis Germanidis, explain GWM Worlds 2, a research preview that turns...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/de08182c3e7747f67ebfa920d0f7ab20?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/de08182c3e7747f67ebfa920d0f7ab20?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Latent Space","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Latent Space","logo":"https://media.daily.dev/image/upload/s--DnJ9laFj--/f_auto,q_auto/v1780213196/logos/latentspace?_a=BAMAMiWQ0","url":"https://daily.dev/sources/latentspace"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"3d,diffusion-models,world-models","timeRequired":"PT94M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Latent Space","item":"https://daily.dev/sources/latentspace"},{"@type":"ListItem","position":3,"name":"Runway’s WorldPrompt and the Engineering of Real-Time Worlds"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/runway-s-worldprompt-and-the-engineering-of-real-time-worlds-v5l0290bz#faq","mainEntity":[{"@type":"Question","name":"What is WorldPrompt in Runway's GWM Worlds 2?","acceptedAnswer":{"@type":"Answer","text":"WorldPrompt is a prompting mechanism, not a programming language, that lets users fix aspects of a generated world (including the first frame) and specify timestamped events or actions, even in real time. It works like a control layer for characters, cameras, and environments, similar to controlling NPCs in a video game, but without scripting or explicit state control. Anyone tracking real-time generative video tooling can follow how prompting interfaces like this evolve on daily.dev."}},{"@type":"Question","name":"What resolution and frame rate does Runway's GWM Worlds 2 generate in real time?","acceptedAnswer":{"@type":"Answer","text":"GWM Worlds 2 streams continuous real-time interactive video at 720p resolution and 24 frames per second, with synchronized audio at 48,000 Hz. It is built by fine-tuning Runway's foundational audio-video model to the WorldPrompt format, post-training it for autoregressive generation, and then using distillation methods to make generation fast enough for real-time playback. Engineers comparing real-time world model specs can keep tabs on releases like this via daily.dev."}},{"@type":"Question","name":"What is error accumulation in autoregressive video generation models and why is it a challenge?","acceptedAnswer":{"@type":"Answer","text":"Error accumulation happens because autoregressive models feed their own generated frames back in as input to generate subsequent frames, so small errors compound over time, degrading quality the longer generation continues. Runway's co-CEO Anastasis Germanidis called this the biggest challenge with autoregressive models, alongside managing GPU memory for infinite-length generations and the model's lack of perfect long-term memory. Developers evaluating autoregressive generation architectures can follow this research thread on daily.dev."}}]} ```