Virtual reality (VR) is creating new opportunities for fashion performance by transforming runway presentation from a fixed physical event into an interactive, spatial and visually responsive experience. This study proposes and evaluates a vision-guided VR fashion stage system designed to improve audience immersive experience through markerless motion tracking, avatar synchronisation and dynamic stage visualisation. A mixed-methods design was adopted. Quantitative data were collected from 400 valid participants who experienced the VR fashion performance, with constructs including VR stage environment design, computer-vision-based motion tracking accuracy, interactivity, visual realism of avatars and garments, synchronisation of fashion elements and audience immersive experience. SmartPLS was used for reliability, validity and structural equation modelling. Qualitative feedback was obtained through 20 semi-structured interviews and analysed thematically using NVivo. In parallel, the technical performance of the real-time pose-estimation and VR tracking framework was validated against a human motion-capture ground-truth dataset. The results indicate that all five experiential factors positively affect audience immersion, with VR stage environment design, interactivity and motion tracking accuracy showing the strongest effects