Open framework and dataset for turning video diffusion models into real-time, long-horizon interactive world models.
Updated EU status checked Last shown
Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
SolarWM is a research framework for training long-horizon video world models, described in the paper "SolarWM: Open Data and Scalable Training for Long-Horizon Video World Models" (2026). It converts bidirectional video generation models into causal, few-step student models that can run in real time and generate interactive worlds.
The project page states that one framework supports four base models, and that it is released with an open data pipeline of 1.43M clips (about 25 TB). One shown variant is the SolarWM-Wan-5B causal student, which runs in 4 steps. The page includes comparisons against LingBot V2 and a zero-shot real-time interactive game demo. A YouTube creator summary says worlds stay consistent for over an hour and that models and training recipes are open-sourced; this is not confirmed on the scraped page text.
It is aimed at researchers and developers working on world models and interactive video generation. The scraped material gives no code, weights or license links, so exact release contents, licensing and hardware requirements are not verified here.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Framework converting video models into real-time interactive worlds that stay consistent for over an hour. Works with several video models, with a 1.3M clip, 25TB dataset, models and training recipes all open-sourced.
Closest neighbours by meaning, not just by category.
Model
Research code that streams video generation chunk by chunk from a rough 3D proxy, built on Code World Model.
Model
Real-time interactive world model with compressed memory and hybrid action and semantic control, built on Wan 2.2.
Model · Runway
Runway's real-time interactive world model generating 720p/24fps video with audio, steered by text actions and camera input.
Model
World model that turns a few context images into an editable, physics-simulated 3D world rendered by a video model.
Model
Interactive autoregressive world model generating minute-long playable video with real-time camera control and prompt events.
Model
14B autoregressive world model that generates interactive, camera-controlled video worlds in 3 denoising steps.
