FLUID: Training-Free Face De-identification via Latent Identity Substitution

2025-11-21

AI Summary

This work addresses the challenge of identity removal in facial images without requiring model training. The proposed method leverages the latent space (h-space) of pre-trained diffusion models to simultaneously eliminate identity information while preserving non-identity attributes. Its core innovation lies in modeling identity editing as a semantic displacement process in latent space, inspired by chemical substitution, and introducing an implicit identity editing paradigm. A reagent loss function is designed to jointly optimize attribute preservation and identity suppression, integrated with both linear and geodesic editing strategies for robust navigation of the latent manifold. Crucially, identity directions are discovered and edited solely via optimization—no fine-tuning or additional training is needed. Experiments on CelebA-HQ and FFHQ demonstrate superior trade-offs between thorough identity removal and high fidelity in pose, expression, and illumination. Both qualitative and quantitative evaluations significantly outperform existing state-of-the-art methods.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We present FLUID (Face de-identification in the Latent space via Utility-preserving Identity Displacement), a training-free framework that directly substitutes identity in the latent space of pretrained diffusion models. Inspired by substitution mechanisms in chemistry, we reinterpret identity editing as semantic displacement in the latent h-space of a pretrained unconditional diffusion model. Our framework discovers identity-editing directions through optimization guided by novel reagent losses, which supervise for attribute preservation and identity suppression. We further propose both linear and geodesic (tangent-based) editing schemes to effectively navigate the latent manifold. Experimental results on CelebA-HQ and FFHQ demonstrate that FLUID achieves a superior trade-off between identity suppression and attribute preservation, outperforming state-of-the-art de-identification methods in both qualitative and quantitative metrics.

Problem

• Substituting facial identity in diffusion model latent space
• Discovering identity-editing directions via optimization with reagent losses
• Achieving superior identity suppression while preserving facial attributes

Innovation

• Training-free face de-identification via latent substitution
• Identity editing as semantic displacement in latent space
• Linear and geodesic editing schemes on latent manifold