SafeLens: Deliberate and Efficient Video Guardrails with Fast-and-Slow Screening

2026-05-17

AI Summary

This work addresses the challenge of balancing safety and computational efficiency in large-scale video content moderation. The authors propose a fast-slow hybrid审核 architecture that processes simple videos through an efficient fast path while reserving a more computationally intensive slow path for complex cases requiring deep reasoning. To enhance training efficacy, they introduce an influence-guided data filtering mechanism to construct a high-quality, compact training set. Furthermore, they integrate structured chain-of-thought (CoT) prompting to strengthen the test-time reasoning capabilities of vision-language models. Evaluated on both real-world and AI-generated video benchmarks, the proposed method significantly outperforms state-of-the-art open- and closed-source moderation systems, achieving leading performance while substantially reducing inference costs.

Abstract

The rapid growth of online video platforms and AI-generated content has made reliable video guardrails a key challenge for safety and real-world deployment. While most videos can be screened through fast pattern recognition, a small subset requires deeper reasoning over temporally complex content and nuanced policy constraints. Existing approaches typically rely on large vision-language models applied uniformly across all inputs, resulting in high inference costs and inefficient allocation of computation. We propose SafeLens, a video guardrail framework that introduces a fast-and-slow inference architecture for efficient and accurate content moderation with variable computational cost across inputs. Additionally, we construct a high-quality dataset by applying influence-guided filtering to the SafeWatch Dataset, retaining only 2.4% of the original data. To further address limitations of training-time scaling, we enable test-time reasoning by augmenting the filtered data with structured Chain-of-Thought traces. Across real-world and AI-generated video benchmarks, SafeLens achieves state-of-the-art performance, outperforming strong open-source video guardrails (e.g., SafeWatch-8B, OmniGuard-7B) and closed-source models (e.g., GPT-5.4, Gemini-3.1-pro) while significantly reducing inference cost, demonstrating that efficient design serves to be more effective than scaling data or model size alone.

Problem

• video guardrails
• content moderation
• efficient inference
• AI-generated content
• computational cost

Innovation

• fast-and-slow inference
• video guardrails
• Chain-of-Thought reasoning
• influence-guided filtering
• efficient content moderation