PromptForge-350k: A Large-Scale Dataset and Contrastive Framework for Prompt-Based AI Image Forgery Localization

TL;DR

This paper introduces PromptForge-350k, a large-scale dataset and a contrastive learning framework for detecting AI-generated image forgeries, addressing the lack of specialized datasets and methods in this emerging field.

Contribution

It presents a fully automated annotation framework, constructs the PromptForge-350k dataset, and proposes ICL-Net, a novel contrastive localization network with superior performance.

Findings

Achieves 62.5% IoU on PromptForge-350k, outperforming SOTA by 5.1%.

Maintains IoU drop of less than 1% under common degradations.

Generalizes well to unseen editing models with 41.5% IoU.

Abstract

The rapid democratization of prompt-based AI image editing has recently exacerbated the risks associated with malicious content fabrication and misinformation. However, forgery localization methods targeting these emerging editing techniques remain significantly under-explored. To bridge this gap, we first introduce a fully automated mask annotating framework that leverages keypoint alignment and semantic space similarity to generate precise ground-truth masks for edited regions. Based on this framework, we construct PromptForge-350k, a large-scale forgery localization dataset covering four state-of-the-art prompt-based AI image editing models, thereby mitigating the data scarcity in this domain. Furthermore, we propose ICL-Net, an effective forgery localization network featuring a triple-stream backbone and intra-image contrastive learning. This design enables the model to capture highly robust and generalizable forensic features. Extensive experiments demonstrate that our method achieves an IoU of 62.5% on PromptForge-350k, outperforming SOTA methods by 5.1%. Additionally, it exhibits strong robustness against common degradations with an IoU drop of less than 1%, and shows promising generalization capabilities on unseen editing models, achieving an average IoU of 41.5%.