Enhancing Single-Image Facial Demorphing using Multimodal Large Language Models

2026-05-25

AI Summary

Existing face fusion detection methods struggle to recover the original identities, limiting their utility in forensic applications. This work proposes the first reference-free facial defusion framework that jointly reconstructs two source faces through a coupled diffusion model. The key innovation lies in directly leveraging intermediate-layer semantic embeddings from multimodal large language models (MLLMs) as conditioning signals for the diffusion process—bypassing the information loss inherent in conventional text-generation-and-re-encoding pipelines. Notably, these MLLM embeddings are found to be more identity-discriminative than features from vision transformers. Operating entirely in the RGB domain, the method enables end-to-end denoising without latent space transformations and achieves a 30–40% performance gain over current approaches under rigorous evaluation protocols.

Abstract

Face recognition systems are increasingly vulnerable to morphing attacks, where a composite image is crafted to match multiple identities, enabling unauthorized access and identity fraud. Existing detection methods identify morphed images but cannot recover constituent images or identities, limiting their forensic utility. This paper presents a novel reference-free facial demorphing framework that leverages Multimodal Large Language Models (MLLMs) to guide a coupled diffusion-based reconstruction process. Our key innovation lies in extracting semantic embeddings from intermediate MLLM layers to condition the demorphing, providing high-level reasoning about facial attributes and identity cues that complement low-level pixel information. We formulate demorphing as a coupled conditional generation problem, where both constituent faces are synthesized jointly through a denoising diffusion model operating directly in the RGB domain, ensuring inter-identity consistency while preserving fine-grained perceptual details. Unlike prior approaches that rely on compressed latent representations or assume identity overlap between training and testing sets, our method bypasses lossy text generation-reencoding cycles by directly utilizing MLLM hidden states as conditioning signals, enabling the denoising network to attend to subtle visual cues such as hair, background, and facial textures. Ablation studies further reveal that middle MLLM layers encode more identity-discriminative representations, RGB-domain demorphing outperforms latent-space approaches by 30--40\% at strict operating points, and full MLLM embeddings provide substantial advantages over raw ViT features through enhanced semantic structuring from multimodal pretraining.

Problem

• facial demorphing
• morphing attacks
• face recognition
• identity fraud
• forensic recovery

Innovation

• Multimodal Large Language Models
• Facial Demorphing
• Diffusion Models
• Semantic Embeddings
• RGB-domain Reconstruction