Video Models Can Reason with Verifiable Rewards

TL;DR

This paper introduces VideoRLVR, a reinforcement learning approach for optimizing video diffusion models to perform verifiable reasoning tasks with explicit constraints, improving reliability and rule consistency.

Contribution

It presents a novel recipe combining rule-based feedback with diffusion models, including an efficient Early-Step Focus strategy, to enhance verifiable reasoning in videos.

Findings

VideoRLVR outperforms supervised baselines on Maze, FlowFree, and Sokoban.

Dense decomposed rewards improve performance in low-success-rate scenarios.

The approach surpasses proprietary and open-source models on reasoning benchmarks.

Abstract

Video diffusion models have made rapid progress in perceptual realism and temporal coherence, but they remain primarily optimized for plausible generation rather than verifiable reasoning. This limitation is especially pronounced in tasks where generated videos must satisfy explicit spatial, temporal, or logical constraints. Inspired by the role of reinforcement learning with verifiable rewards (RLVR) in reasoning-oriented language models, we introduce VideoRLVR, a practical recipe for optimizing video diffusion models with rule-based feedback. VideoRLVR formulates video reasoning as the generation of verifiable visual trajectories and consists of an SDE-GRPO optimization backbone, dense decomposed rewards, and an Early-Step Focus strategy for efficient training. The Early-Step Focus strategy restricts policy optimization to the early denoising phase, reducing training latency by about 40% while preserving performance. We evaluate VideoRLVR on Maze, FlowFree, and Sokoban, three procedurally generated domains with objective success criteria. Across these tasks, VideoRLVR consistently improves over supervised fine-tuning baselines, with dense decomposed rewards proving especially important in low-success-rate settings. Our RL-optimized model also outperforms the evaluated proprietary and open-source video generation models on these verifiable reasoning benchmarks and out-of-domain benchmarks. These results suggest that verifiable RL can move video models beyond perceptual imitation toward more reliable rule-consistent visual reasoning.