Membership Inference Attacks for Face Images Against Fine-Tuned Latent Diffusion Models

TL;DR

This paper presents a membership inference attack method to determine if specific face images were used to fine-tune latent diffusion models, highlighting privacy risks in generative image models.

Contribution

It introduces a novel MIA approach leveraging auxiliary data and watermarks, demonstrating effectiveness against fine-tuned LDMs in black-box scenarios.

Findings

Auxiliary data improves attack performance

Watermarks enhance inference accuracy

Longer fine-tuning reduces prompt influence

Abstract

The rise of generative image models leads to privacy concerns when it comes to the huge datasets used to train such models. This paper investigates the possibility of inferring if a set of face images was used for fine-tuning a Latent Diffusion Model (LDM). A Membership Inference Attack (MIA) method is presented for this task. Using generated auxiliary data for the training of the attack model leads to significantly better performance, and so does the use of watermarks. The guidance scale used for inference was found to have a significant influence. If a LDM is fine-tuned for long enough, the text prompt used for inference has no significant influence. The proposed MIA is found to be viable in a realistic black-box setup against LDMs fine-tuned on face-images.