EDGE-Shield: Efficient Denoising-staGE Shield for Violative Content Filtering via Scalable Reference-Based Matching

TL;DR

EDGE-Shield is a scalable, training-free content filtering method for Text-to-Image models that efficiently blocks violative content during denoising, significantly reducing processing time while maintaining accuracy.

Contribution

It introduces an embedding-based matching approach and an $x$-pred transformation to improve scalability and early-stage classification in content filtering.

Findings

Achieves approximately 79% reduction in processing time for Z-Image-Turbo.

Achieves approximately 50% reduction in processing time for Qwen-Image.

Maintains filtering accuracy across different generative model architectures.

Abstract

The advent of Text-to-Image generative models poses significant risks of copyright violation and deepfake generation. Since the rapid proliferation of new copyrighted works and private individuals constantly emerges, reference-based training-free content filters are essential for providing up-to-date protection without the constraints of a fixed knowledge cutoff. However, existing reference-based approaches often lack scalability when handling numerous references and require waiting for finishing image generation. To solve these problems, we propose EDGE-Shield, a scalable content filter during the denoising process that maintains practical latency while effectively blocking violative content. We leverage embedding-based matching for efficient reference comparison. Additionally, we introduce an \textit{$x$}-pred transformation that converts the model's noisy intermediate latent into the pseudo-estimated clean latent at the later stage, enhancing classification accuracy of violative content at earlier denoising stages. We conduct experiments of violative content filtering against two generative models including Z-Image-Turbo and Qwen-Image. EDGE-Shield significantly outperforms traditional reference-based methods in terms of latency; it achieves an approximate $79\%$ reduction in processing time for Z-Image-Turbo and approximate $50\%$ reduction for Qwen-Image, maintaining the filtering accuracy across different model architectures.