探索生成式 AI 在虛實結合 WebXR 中的應用個性化虛擬人物圖像生成

Abstract

本研究旨在探討生成式人工智慧（Generative AI）於虛實整合環境中的應用，並嘗試透過 WebXR 技術，建立一個具有沉浸感與個人化特色的虛擬角色體驗系統。研究以榮格心理學中的 12 原型理論作為角色性格設定的依據，搭配使用者偏好輸入，藉由 Stable Diffusion 模型及多項 LoRA 權重，產製具視覺與心理貼合度的虛擬人物圖像，再進一步導入至 360 度背景場景中，呈現於 WebXR 平台上進行體驗與測試。本研究依據角色生成流程共設計四個階段的使用者體驗問卷，蒐集視覺效果、角色共鳴、風格一致性及沉浸感等面向之回饋，並透過階段性調整與優化，提升生成內容的整體品質與使用者接受度。研究中亦實作臉部替換模組（PULID）與動漫風格控制技術，以解決寫實生成易產生違和感的問題，進一步強化角色的辨識度與親和力。實驗結果顯示，當角色設計能兼顧心理設定與外觀細節，並結合易於理解的互動展示方式時，能有效提升使用者的參與感與沉浸體驗。最終，本研究提出一套整合心理學理論與生成式 AI 圖像技術的角色生成流程，並探討其於虛實整合平台中的應用可能性，期望能為未來的互動設計與個性化內容開發提供新的參考方向。This study explores the application of generative artificial intelligence (Generative AI) in hybrid virtual-real environments, aiming to develop a personalized and immersive virtual character experience system using WebXR technology. Grounded in Carl Jung’s theory of 12 archetypes, the study incorporates users’ preference inputs to generate psychologically resonant and visually personalized character images using the Stable Diffusion model combined with various LoRA weights. These virtual characters are then integrated into 360-degree background scenes and presented on a WebXR platform for user interaction and testing.A four-stage user experience questionnaire was designed to evaluate visual satisfaction, emotional resonance, style consistency, and immersion across different phases of the generation process. Based on the feedback, the system was iteratively optimized to improve both content quality and user acceptance. The study also implements a face-swapping module (PULID) and anime-style LoRA adjustments to address the uncanny valley issues often caused by hyper-realistic generation, thereby enhancing character recognizability and approachability.Experimental results suggest that when character design considers both psychological profiling and visual detailing, and is supported by an intuitive interaction interface, it significantly enhances user engagement and immersion. This research proposes an integrated workflow combining psychological theory and generative image technologies, and discusses its potential for applications within WebXR environments. It aims to offer a new perspective for future developments in interactive design and personalized content creation.