Realtime models
Runway GWM Worlds 2
NewRunway’s realtime world model — interactive worlds streamed as continuous 720p/24fps video with synced 48kHz audio.
runway.com · listing ≠ endorsement · re-checked 2026-10-02 · Outdated or wrong? Report this entry
- Operator
- Runway
- Launched
- Sep 3, 2026 (research preview)
- Model
- Autoregressive diffusion, GWM-1 lineage (Gen-4.5)
- Output
- 720p · 24fps · 48kHz audio, interactive sessions
- Format
- WorldPrompt control — characters, camera, events
- Ceiling
- Degrades after minutes; no scripted state
What is Runway GWM Worlds 2?
Runway’s research preview (announced Sep 3, 2026) turns high-fidelity video-and-audio generation into a real-time interactive simulation: an autoregressive diffusion model from the GWM-1 lineage (post-trained on Gen-4.5) that generates indefinitely rather than as a fixed clip. A WorldPrompt control layer steers characters, camera and environment — with timestamped events and a fixable first frame — while the session is live. Runway itself flags the format’s standing limit: autoregressive error accumulation degrades the world after a few minutes, and there is no scripted state like a game engine.
What Runway GWM Worlds 2 does
- Realtime interactive worlds — continuous 720p video at 24fps with synced 48kHz audio
- Autoregressive diffusion: generates over time, no preset clip length
- WorldPrompt control layer — steer characters, camera and environment, with timestamped events
- Built on the GWM-1 lineage, post-trained on Runway’s Gen-4.5 base model
- Research preview (announced Sep 3, 2026), not a shipped product
Who is Runway GWM Worlds 2 for?
- Exploring playable, steerable generative worlds that render live while you direct them
- Comparing realtime world models (Genie 3, Odyssey-2 Pro, World Labs RTFM) on spec
- Prototyping live-directed generative formats before wiring them to a broadcast
Pros and cons of Runway GWM Worlds 2
Pros
- Video and audio generated in sync, in real time, from a major video-AI lab
- WorldPrompt gives concrete, time-stamped control rather than one-shot prompts
- Candidly documented limits — Runway discusses error accumulation openly
Cons
- Worlds degrade after a few minutes as autoregressive errors accumulate
- No scripted state or game-engine-style logic — agents and users see pixels only
- A research preview: access, pricing and durability are all unproven
What people are saying
Verbatim quotes from public coverage — collected 2026-10-02.
“The biggest challenge with autoregressive models is error accumulation.”
Frequently asked questions
What is Runway GWM Worlds 2?
Runway’s realtime world model: an autoregressive diffusion model that streams interactive 720p/24fps worlds with synced 48kHz audio, announced as a research preview on Sep 3, 2026.
What is WorldPrompt?
Runway’s prompting/control layer for GWM Worlds — it steers characters, camera and environment, supports timestamped events, and can fix elements like the first frame.
Can GWM Worlds 2 run forever like an AI TV channel?
Not yet — Runway says autoregressive error accumulation degrades the world after a few minutes, unlike the always-on H3-based streams on fal.
MiniMax H3
Open-weight 33B video model (Hailuo 3.0) — 15s clips with native stereo audio; the substrate under this wave.huggingface.co
fal H3 Max Live
fal’s realtime-tuned H3 Max — five seconds of 768p in under three, faster than playback.fal.ai
Krea Realtime 14B
Open 14B autoregressive model for real-time, long-form streaming video — distilled from Wan 2.1, ~1s to first frame.krea.ai
Decart MirageLSD
“Live-stream diffusion” — restyle a live feed (camera, game, call, screen) frame-by-frame from a text prompt at near-zero latency.decart.ai
Lightricks LTX Video
Open latent-diffusion video model that generates clips faster than real time — the engine several chat-driven live loops run on.github.com
ByteDance Seed Seaweed APT2
Real-time interactive video research — an 8B model that emits a latent frame per forward pass at 24fps.seaweed-apt.com