Thinking in Video: Can Video Generators Really Reason About the Real World?

2026-07-19

AI Summary

This work investigates whether current video generation models possess genuine causal reasoning capabilities or merely replicate superficial patterns. To address this, the authors introduce the “video thinking” paradigm, which leverages generative simulation to model and predict real-world dynamics, and propose a novel Causal-Generative Dual-Judge (CGDJ) evaluation framework. This framework assesses explicit causal understanding through spatiotemporally flattened visual question answering and evaluates implicit causal reasoning via consistency in future video generation and audio-visual alignment. Experimental results reveal that while open-source models can produce plausible dynamics, they lack explicit causal awareness; even advanced closed-source systems, though superior, still exhibit a perception-prediction gap and inconsistencies in audio-visual logical coherence.

Abstract

Recent advances in world models and video generation have given rise to an emerging reasoning paradigm that leverages video generative models to simulate, predict, and reason about real-world dynamics. We redefine this paradigm as Thinking in Video, where video is not merely an output artifact but a medium for constructing, extending, and verifying causal thought. However, this promise remains unverified: convincing rollouts may reflect memorized appearances rather than causal understanding, while existing metrics separate perceptual fidelity from semantic logic. To evaluate whether video generators support such reasoning, we introduce the Causal-Generative Dual-Judge (CGDJ), auditing World Model Consistency from two perspectives. Explicit Causal Perception tests whether a generator reads a video scenario as a reasoning problem through spatio-temporal flattened visual question answering, while Implicit Generative Perception-Prediction Gap evaluates whether it renders the causal consequence as a consistent future video. Applying CGDJ to representative open- and closed-source generators reveals a clear Perception-Prediction Gap: open-source models produce plausible dynamics despite near-zero explicit causal perception, whereas advanced closed-source systems show stronger but still limited alignment between reasoning and generation. Further analysis exposes audio-visual misalignment, where models verbalize correct causal logic more reliably than they render it, challenging the "world simulator" narrative.

Problem

• video generation
• causal reasoning
• world models
• perception-prediction gap
• real-world dynamics

Innovation

• Thinking in Video
• Causal-Generative Dual-Judge
• World Model Consistency
• Perception-Prediction Gap
• Causal Reasoning