Skip to content
QbitAI · WeChat

Peking University Open-Sources Unified World Model Framework for Synthesis and Reasoning Tasks

北大开源统一世界模型框架:多类合成推理任务一套搞定

Peking University DCAI and Kuaishou Kling open-sourced OpenWorldLib for four task types: video generation, 3D modeling, VLA control, and multimodal reasoning. Its Pipeline coordinates Operator, Reasoning, Synthesis, Representation, and Memory modules, supporting forward and stream execution. The key test is whether unified interfaces cut cross-task reproduction cost.

Why it matters: Peking University DCAI and Kuaishou Kling open-sourced OpenWorldLib, putting video generation, 3D modeling, VLA robot control and multimodal reasoning into one pipeline, split into operator, reasoning, synthesis, representation and memory modules.

Read the original ↗Export Markdown