Search papers, labs, and topics across Lattice.
This paper addresses the challenge of personalized person image generation by proposing a dual-branch framework that integrates global appearance control with local facial detail preservation. The authors introduce Dynamic Balancing Scaling (DBS), a fine-tuning strategy that enhances coordination between these branches through adaptive temporal gating and region-aware optimization. Experimental results demonstrate that DBS significantly improves the balance between facial fidelity and overall appearance consistency compared to existing methods, while also introducing a new benchmark, Pexels-100, for evaluating holistic identity preservation.
Achieving a harmonious balance between facial fidelity and overall appearance consistency could redefine personalized image generation.
Personalized person image generation requires preserving subject identity across both local facial details and broader appearance cues. Existing methods typically emphasize only one level of identity information, leading to an inherent trade-off between facial fidelity and overall appearance consistency. To address this, we first propose a simple dual-branch baseline that unifies global appearance control and local facial control within a shared generation framework. This simple combination of different branches yields promising results, but suffers from instability in practice due to uncoordinated branch contributions. To this end, we propose Dynamic Balancing Scaling (DBS), a fine-tuning strategy for improving face and appearance identity coordination. DBS consists of two components: adaptive temporal gating, which dynamically modulates branch contributions along the denoising trajectory, and region-aware optimization, which improves the coordination of facial, appearance, and global supervision. Together, these designs alleviate persistent face-branch over-dominance and encourage more effective appearance-aware guidance. We also introduce Pexels-100, a benchmark for evaluating holistic identity consistency in personalized person generation. Experiments show that DBS achieves a better trade-off between facial fidelity and appearance consistency than existing open-source baselines, while providing a controllable basic framework for holistic identity modeling.
