Multi-person image generation affects identities beyond the target subjects
Cross-Person Identity Change Under Source-Coupled Multi-Human Generation: A Controlled UniPortrait Study
Computer Vision and Pattern Recognition
Summary
This paper studies how AI tools that change pictures with multiple people can affect not only the person being changed but also others in the same image. The researchers used a system called UniPortrait and measured how well it preserved each person’s face identity after edits. They found that traditional methods often disrupt identities broadly, while certain newer methods can better maintain both the main target and others. However, the study doesn’t prove which components cause these effects or if these results apply to other systems.
What this means in practice
- •For image processing engineers: Improve multi-person image generators by carefully handling identity preservation for all individuals in a scene.
- •For digital content moderators: Evaluate the risks of unintended identity changes when applying image generation to group photos.
Authors
Edrees Al-Edreesi, Ghaleb Al-Gaphari, Muneer Alsurori, Mohammed Edrees, Fouad Al-Hamdani
Abstract
Multi-human personalization must account for identity changes in people outside the nominal target of a generative operation. We characterize this behavior in UniPortrait, without introducing a new preservation algorithm. Two sequential studies use 119 accepted two-person baseline scenes and three fixed output seeds, measuring continuous reference-based facial similarity with InsightFace \buffalo. In Study~1, source-coupled single-ID target-only generation (A1) produces negative mean transitions for both the protected person ($-0.567904$) and target ($-0.236904$), indicating broad perturbation rather than established successful repair with collateral loss. In Study~2, native dual-ID generation (A2) yields more positive paired contrasts: protected $+0.512918$ (95\% case-cluster CI $[0.471434,0.550259]$) and target $+0.158853$ ($[0.118903,0.197523]$), across 287 jointly measurable pairs from 111 cases. Shared baseline similarity cancels algebraically in these paired contrasts, leaving a matched POST comparison. Validation designed after the primary result freeze finds directional agreement with an independent identity evaluator and blinded human judgments; within-\buffalo\ nuisance models provide supportive sensitivity evidence. The findings characterize intact native regimes, whose references, routing, adapters, and token counts co-vary. They do not establish component causality, categorical identity preservation, successful localized editing, or generalization beyond the evaluated implementation and measurable cohort.