Researchers have proposed a new framework called H-GIM, or Human-Centered Generative Immersive Multimedia, designed to address the scalability limitations of traditional extended reality production pipelines. Published in Multimedia Tools and Applications, the study introduces a system powered by generative and multimodal artificial intelligence that aims to reduce the reliance on manual content creation and static interaction models. The framework is intended to enable real-time adaptation to user behavior in educational, entertainment, and collaborative environments.
The core architecture of H-GIM integrates multimodal foundation models capable of processing textual, audio, visual, and spatial signals. These components are coordinated through a router-based orchestration layer, which allows the system to manage diverse data inputs simultaneously.
A significant technical contribution of the study is the introduction of a synthetic data generation pipeline. The authors note a lack of large-scale training datasets specifically for extended reality applications. To overcome this, the pipeline automatically produces high-fidelity 3D assets using procedurally guided generative models.
The generation of 3D assets within H-GIM employs a hybrid methodology to ensure physical plausibility. Geometry is produced via procedural seeds that undergo physics validation, ensuring that the structural integrity of the assets remains consistent with real-world laws. In contrast, the texture maps are generated using SDXL-Turbo, a specific diffusion model. The authors explicitly clarify that SDXL-Turbo is used only for texture maps, ensuring that the system does not claim end-to-end diffusion-based mesh generation.
Beyond asset generation, the framework includes an affective interaction module that dynamically adapts multimedia complexity based on user cognitive load and behavioral signals. This module allows the system to monitor how users interact with the immersive environment and adjust the content accordingly.
The researchers evaluated the framework using a functional prototype to measure its performance in practical scenarios. The experimental results demonstrated that H-GIM achieves an average asset-generation time of 1.1 minutes, with a standard deviation of 0.2 minutes.
In addition to generation speed, the study measured the system’s responsiveness and user experience. The mean interaction latency was recorded at 42 milliseconds, indicating a low delay between user input and system response. The prototype also achieved a user-presence score of 0.81, with a standard deviation of 0.04, suggesting a high level of perceived immersion among participants.
A user study involving 50 participants further validated the effectiveness of the H-GIM framework. The results indicated statistically significant improvements in the immersive experience compared with conventional extended reality workflows, with a p-value of less than 0.001. This statistical significance underscores the potential of generative and multimodal AI to enhance the quality of immersive multimedia systems.
The findings of this study highlight the potential of generative and multimodal AI to enable scalable, adaptive, and human-centered immersive multimedia systems.
Subject of Research: Development and evaluation of H-GIM, a human-centered generative immersive multimedia framework using AI for adaptive XR systems.
Article Title: H-GIM: a human-centered generative immersive multimedia framework for adaptive immersive multimedia systems
Article References: Chebbi, I., Abidi, S., & Ayed, L. B. (2026). H-GIM: a human-centered generative immersive multimedia framework for adaptive immersive multimedia systems. Multimedia Tools and Applications, 85(11), Article 842. https://doi.org/10.1007/s11042-026-22003-0
Image Credits: AI Generated
DOI: 10.1007/s11042-026-22003-0
Keywords: Immersive Multimedia, Generative AI, Extended Reality, 3D Asset Generation, Human-Computer Interaction, Multimodal AI, XR Development, Cognitive Load, H-GIM, human-centered, generative, immersive
News Source: Denise Maddox. (November 9, 2026). New AI Framework Generates Immersive 3D Assets in Minutes. Scienmag.
