Video Models Reason Early: Exploiting Plan Commitment for Maze Solving

TL;DR

This paper investigates the internal reasoning process of video diffusion models in maze solving, revealing early plan commitment and introducing a chaining method to improve complex maze solving.

Contribution

It uncovers the early plan commitment behavior in video models and proposes ChEaP, a chaining method that significantly enhances maze-solving accuracy.

Findings

Video models commit to high-level plans within the first few denoising steps.

Maze difficulty correlates with path length, not obstacle density.

ChEaP improves maze-solving accuracy from 7% to 67% on long mazes.

Abstract

Video diffusion models exhibit emergent reasoning capabilities like solving mazes and puzzles, yet little is understood about how they reason during generation. We take a first step towards understanding this and study the internal planning dynamics of video models using 2D maze solving as a controlled testbed. Our investigations reveal two findings. Our first finding is early plan commitment: video diffusion models commit to a high-level motion plan within the first few denoising steps, after which further denoising alters visual details but not the underlying trajectory. Our second finding is that path length, not obstacle density, is the dominant predictor of maze difficulty, with a sharp failure threshold at 12 steps. This means video models can only reason over long mazes by chaining together multiple sequential generations. To demonstrate the practical benefits of our findings, we introduce Chaining with Early Planning, or ChEaP, which only spends compute on seeds with promising early plans and chains them together to tackle complex mazes. This improves accuracy from 7% to 67% on long-horizon mazes and by 2.5x overall on hard tasks in Frozen Lake and VR-Bench across Wan2.2-14B and HunyuanVideo-1.5. Our analysis reveals that current video models possess deeper reasoning capabilities than previously recognized, which can be elicited more reliably with better inference-time scaling.