Cross-disciplinary teams often face the challenges of semantic misalignment and low emotional engagement during the early stages of design. This study proposes an interaction system that integrates visual clustering and physiological sensing to observe semantic alignment and emotional engagement during visual interpretation in team interaction. Although AI image generation tools are increasingly being adopted to facilitate language-based communication, their inherent ambiguity and semantic distortion can lead to misunderstandings and hinder consensus building. At the same time, emotional cues that influence collaboration quality are often embedded in non-verbal interactions; if not detected in time, these can result in withdrawal or breakdowns in communication. To address these issues, this study proposes an interactive system that integrates semantic input, AI image generation, and wearable emotion sensing devices. Using VGG16 for image feature clustering and the Emotion Wings device to record real-time heart rate variability (HRV), the study analyzes the interaction between image style, semantic orientation, and emotional engagement. The results indicate that semantically balanced images help enhance participant engagement and mutual understanding, while semantic bias or character exclusion tend to lead to emotional withdrawal and low consensus.