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

TL;DR

SafeLens introduces a fast-and-slow inference framework for efficient, accurate video content moderation, significantly reducing inference costs while outperforming existing models on real-world benchmarks.

Contribution

The paper presents a novel fast-and-slow inference architecture, a high-quality filtered dataset, and test-time reasoning techniques for improved video guardrails.

Findings

SafeLens outperforms existing open-source and closed-source video guardrails.

It achieves state-of-the-art performance on real-world and AI-generated video benchmarks.

The approach reduces inference costs significantly compared to scaling models or data.

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.