Releases·Model Release
Astronex-World 1.0: Real-Time Interactive World Model Foundation
Astronex-World 1.0 is an open controllable video world-model foundation supporting text-to-video and image-to-video generation under camera trajectories, continuous actions, and inserted text events. Built on a 5B prior, it streams 832x480 video at 24 fps in real time on a single GPU and outperforms larger 13.6B–14B models on WBench. Its reserved action interfaces enable post-training for embodied intelligence and autonomous driving.
0 engagement1 sourceThu, Sep 17, 2026, 10:38 AM
AI Overview
- Astronex-World 1.0 is an open controllable video world-model family built on the Wan2.2-TI2V-5B prior, offering bidirectional and block-causal models.
- The causal model generates 832x480 video at 24 fps in real time on one NVIDIA L20 48 GB GPU; all five training stages run on two L20s.
- It scores 73.5 on WBench Navi and 70.0 on WBench Full, beating the 13.6B LongCat-Video and 14B Helios and trailing the 22B LTX-2.3 by under one point.
- PRoPE injects camera intrinsics/extrinsics and a 64-dimensional action stream modulates every Transformer layer; reserved interfaces enable post-training for embodied intelligence and autonomous driving.
Related
Releases·Model Release
AlayaWorld open-source video world model released with 720p 24 FPS streaming and camera control
AlayaWorld is a full-stack, open-source video world model that supports 720p, 24 FPS streaming video generation with camera control and text-driven event generation. It is built around four core properties: interaction, consistency, stability, and runtime.
AlayaWorld is a full-stack, open-source video world model that supports 720p, 24 FPS streaming video generation with camera control and text-driven event generation. It is built around four core properties: interaction, consistency, stability, and runtime.
64 engagement1 sourceredditWed, Jul 22, 2026, 03:46 AM
Releases·Model Release
Runway launches GWM Worlds 2 research preview for real-time interactive world simulation
Runway announced GWM Worlds 2, a research preview for real-time interactive world simulation built on its foundational audio-visual generation model. The system generates continuous 720p video at 24 fps with 48 kHz audio that responds dynamically to user inputs like movement, dialogue, and weather changes, rather than relying on pre-scripted clips. It matters to practitioners as an early playable world model in the growing real-time generative simulation space.
Runway announced GWM Worlds 2, a research preview for real-time interactive world simulation built on its foundational audio-visual generation model. The system generates continuous 720p video at 24 fps with 48 kHz audio that responds dynamically to user inputs like movement, dialogue, and weather changes, rather than relying on pre-scripted clips. It matters to practitioners as an early playable world model in the growing real-time generative simulation space.
69 engagement1 sourceredditThu, Sep 3, 2026, 06:01 PM
Developments·Paper
4DStreamCtrl: Interactive Video Generation with Online 4D Control
Researchers propose 4DStreamCtrl, a method for interactive video generation that enables real-time, streaming control of both camera and object motion in 3D. It addresses limitations of prior approaches that handle only camera or 2D trajectories, or run offline at fixed lengths.
Researchers propose 4DStreamCtrl, a method for interactive video generation that enables real-time, streaming control of both camera and object motion in 3D. It addresses limitations of prior approaches that handle only camera or 2D trajectories, or run offline at fixed lengths.
0 engagement1 sourcearxivWed, Aug 26, 2026, 07:49 AM
Developments·Paper
H3-World paper turns MiniMax-H3 video generator into interactive world model
Researchers introduce H3-World, a framework that converts the 33B MiniMax-H3 video generator into an interactive world model controlled through language. The approach builds on MiniMax-H3's existing zero-shot natural-language control of character behavior and camera motion, refining it into precise, temporally grounded control without adding dedicated action modules. It suggests language is emerging as a natural interface for steering large video generators.
Researchers introduce H3-World, a framework that converts the 33B MiniMax-H3 video generator into an interactive world model controlled through language. The approach builds on MiniMax-H3's existing zero-shot natural-language control of character behavior and camera motion, refining it into precise, temporally grounded control without adding dedicated action modules. It suggests language is emerging as a natural interface for steering large video generators.
0 engagement1 sourcearxivTue, Sep 1, 2026, 05:21 PM
Developments·Paper
ForgeWM: Progressive Causal Training for Few-Step Action-Conditioned Video World Models
Researchers introduce ForgeWM, a progressive framework that converts bidirectional action-conditioned video generators into efficient few-step world models via domain adaptation. It addresses the challenge of aligning discrete keyboard and continuous mouse inputs with temporally compressed latent chunks during causal training and autoregressive rollout, enabling low-latency interactive video synthesis.
Researchers introduce ForgeWM, a progressive framework that converts bidirectional action-conditioned video generators into efficient few-step world models via domain adaptation. It addresses the challenge of aligning discrete keyboard and continuous mouse inputs with temporally compressed latent chunks during causal training and autoregressive rollout, enabling low-latency interactive video synthesis.
0 engagement1 sourcearxivFri, Aug 14, 2026, 07:13 AM