Code World Model 2.0: proxy-guided renderer streams consistent 30-minute video

_akhaliq · x · 2026-11-19

The paper 'Code World Model 2.0: Towards Infinite Controllable Video Generation' (arXiv 2611.22925) shares Invoker-Render, its video renderer: it takes a structural proxy, a reference image, and a text prompt — a coarse proxy can synthesize natural gait from character positions, while a detailed proxy guides individual limb movements. For streaming, Progressive Forcing distills the causal renderer to four denoising steps per chunk, with training on its own generated history to reduce drift; a continuous 30-minute racing demo stays visually consistent while following the input trajectory.

Original post →

More from Multimodal

Multimodal channel →