GuardT2I: Defending Text-to-Image Models from Adversarial Prompts

TL;DR

GuardT2I introduces a generative moderation framework using large language models to detect and mitigate adversarial prompts in text-to-image models, significantly improving safety without harming performance.

Contribution

This work presents a novel generative approach with LLMs for adversarial prompt detection in T2I models, surpassing existing moderation solutions.

Findings

Outperforms OpenAI-Moderation and Azure Moderator in adversarial scenarios

Enhances T2I safety without degrading image generation quality

Utilizes LLMs for conditional transformation of text prompts

Abstract

Recent advancements in Text-to-Image (T2I) models have raised significant safety concerns about their potential misuse for generating inappropriate or Not-Safe-For-Work (NSFW) contents, despite existing countermeasures such as NSFW classifiers or model fine-tuning for inappropriate concept removal. Addressing this challenge, our study unveils GuardT2I, a novel moderation framework that adopts a generative approach to enhance T2I models' robustness against adversarial prompts. Instead of making a binary classification, GuardT2I utilizes a Large Language Model (LLM) to conditionally transform text guidance embeddings within the T2I models into natural language for effective adversarial prompt detection, without compromising the models' inherent performance. Our extensive experiments reveal that GuardT2I outperforms leading commercial solutions like OpenAI-Moderation and Microsoft Azure Moderator by a significant margin across diverse adversarial scenarios. Our framework is available at https://github.com/cure-lab/GuardT2I.