💻 computer science

Generative-AI-Enhanced XR Data Services for Intangible Cultural Heritage: Culturally Constrained Symbol Generation and User Engagement in Indonesian Batik

This study proposes and evaluates a culturally constrained AI-XR framework for Indonesian Batik that demonstrates how integrating generative AI with cultural-symbol validation significantly enhances perceived service quality, cultural authenticity, and user engagement compared to traditional or non-constrained digital experiences.

Original authors: Xinyi Zhang

Published 2026-09-29

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Xinyi Zhang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine a world where the stories woven into fabric are not just preserved in glass cases, but can be touched, explored, and even reimagined by visitors in real time. This is the promise of digital heritage, a field that seeks to use technology to keep living traditions alive. At the heart of this effort is a concept called "intangible cultural heritage." Unlike a stone statue or a painting, which is a physical object you can hold, intangible heritage consists of the skills, knowledge, and meanings passed down through generations. It is the way a pattern is made, the story behind a symbol, and the community that keeps the practice going. The challenge for researchers is to create digital tools that respect these deep cultural roots. They must avoid turning sacred traditions into mere decorations or generic images. The goal is to build systems that allow people to engage with a culture's soul without distorting its essence.

A researcher turned their attention to Indonesian Batik, a UNESCO-recognized tradition of hand-dyed textiles known for its intricate patterns and deep symbolic meanings. They asked a difficult question: Can artificial intelligence help people explore these patterns in a new way, without losing the cultural truth that makes them special? To find out, they built a digital service that combines three things: a computer program that can generate new patterns, a virtual reality environment where users can walk through a gallery, and a strict set of rules to ensure the computer respects the culture. The researcher did not just want to see if the computer could make pretty pictures; they wanted to know if people would feel that the experience was respectful, authentic, and meaningful.

The researcher began by gathering a massive collection of Batik images, carefully checking that they had the right permissions to use them. They fed these images into a computer system designed to learn the rules of Batik design. This system was not the most powerful or newest AI available; instead, the researcher chose a stable, reliable method that could be closely monitored. The computer learned to create new patterns that looked like Batik, but with a crucial safety net: a "cultural-symbol constraint." This acted like a filter, ensuring the computer did not mix up different regional styles or create nonsense designs that violated the traditional rules. Before anyone saw the results, three experts in Batik culture reviewed hundreds of the computer-generated images. They rated them on a scale of one to seven. The average score was 5.2. This number told the researcher something important: the computer was good enough to be used for guided learning and exploration, but it was not yet skilled enough to replace a human artisan or to claim it was creating a perfect, authentic piece of tradition on its own.

With the technology tested, the researcher invited 531 people from various backgrounds to try out their digital service. The participants were split into three groups. The first group saw a standard website with static images and text. The second group used a virtual reality gallery where they could walk around and look at the patterns in a 3D space, but they could not change anything. The third group used the full "AI-XR" service. They walked through the virtual gallery, but they could also ask the computer to generate new patterns, mix different styles, and see how the designs changed in real time. The researcher measured how much each group felt the service was high-quality, how immersed they felt in the virtual space, and whether they believed the experience was culturally authentic.

The results revealed a clear path to engagement. The group that used the full AI-XR service reported the highest quality of experience. They felt more involved and found the personalization features useful. However, the technology alone did not make the experience successful. The study found that the high quality of the service first made people feel "present," as if they were truly inside the cultural space. This feeling of presence then helped them judge the experience as culturally authentic. When people felt the experience was respectful and true to the tradition, they became more curious and emotionally connected to the culture. This connection, in turn, made them more likely to want to use the service again and to share the experience with others.

Crucially, the study showed that the computer's ability to generate images was not the only thing that mattered. If the system had generated beautiful patterns that felt culturally shallow or disrespectful, the users would not have engaged with them, no matter how advanced the technology was. The "cultural-symbol constraint" and the expert review were essential. They ensured that the computer stayed within the boundaries of the tradition, which allowed the users to trust the experience. The researcher found that the path from a good digital service to a desire to share the culture was a long one, passing through feelings of presence and authenticity. It was not a direct jump from "cool technology" to "cultural love."

The findings offer a cautious but hopeful guide for the future of digital heritage. The researcher concluded that artificial intelligence can be a powerful tool for exploring living traditions, but only if it is used as a supervised assistant rather than an independent creator. The technology must be wrapped in layers of cultural explanation, expert validation, and clear rules. When these conditions are met, digital services can help people feel a genuine connection to traditions like Batik, encouraging them to learn more and pass the knowledge on. The study suggests that the future of preserving culture lies not in replacing human craft with machines, but in using machines to create safe, respectful, and engaging spaces where people can discover the stories behind the patterns for themselves.

Technical Summary: Generative-AI-Enhanced XR Data Services for Intangible Cultural Heritage

Problem Statement
The paper addresses the challenge of digitizing Intangible Cultural Heritage (ICH), specifically Indonesian Batik, which involves not just visual patterns but complex symbolic systems, regional histories, and craft knowledge. Existing digital heritage applications often treat heritage as static archives or focus on visual novelty without ensuring cultural fidelity. Generative AI offers the potential for personalized exploration but carries the risk of producing culturally shallow or inaccurate outputs that detach symbols from their traditions. The core problem is developing a framework that integrates generative AI with Extended Reality (XR) to create a "data service" that supports exploration while maintaining cultural-symbol fidelity and perceived authenticity, rather than merely generating high-quality images.

Methodology
The study employs a mixed-method design comprising four integrated components:

Dataset Construction: A curated corpus of 3,850 Batik images was assembled from public-domain, open-license, and institutionally authorized sources. Images were screened for resolution (>512x512), duplicates, and watermarks, then annotated by two independent cultural experts (inter-rater agreement $\kappa=0.84$ ) into motif categories (e.g., Parang, Kawung, Mega Mendung).
Generative Model & Architecture: The system utilizes a four-layer architecture: Cultural Data, Generative AI, XR Service, and Evaluation.
- Generation: A WGAN-GP (Wasserstein GAN with Gradient Penalty) model was selected for training stability on moderate-sized datasets. It uses a generator with five deconvolution layers and a critic with five convolution layers.
- Personalization: An AdaIN (Adaptive Instance Normalization) module, based on a pre-trained VGG-19 encoder, enables style transfer and user-controlled style mixing.
- Constraints: A ResNet-50 classifier (fine-tuned to 84.2% accuracy) screens outputs for motif recognizability. Outputs are subject to a cultural-symbol constraint module and expert validation.
- XR Interface: The system is deployed via WebXR (Unity 2022.3), allowing spatial navigation and interaction.
Expert Validation: 300 generated outputs were evaluated by three Batik heritage experts using a 7-point rubric covering motif recognizability, symbolic coherence, and cultural respectfulness.
User Experiment: A randomized controlled trial with $N=531$ participants compared three conditions: (1) 2D Digital Exhibition, (2) XR-only Exhibition, and (3) AI-XR Generative Service. The AI-XR condition bundled XR interaction with AI motif generation and style personalization.
Analysis: Structural Equation Modeling (SEM) with 5,000 bootstrapped samples tested a sequential mechanism: Perceived Service Quality $\rightarrow$ Immersive Presence $\rightarrow$ Perceived Cultural Authenticity $\rightarrow$ Cultural Engagement $\rightarrow$ Continuance/Transmission Intention.

Key Contributions

Framework Integration: The paper proposes a culturally constrained AI-XR data-service framework that links a curated image corpus, generative models, expert validation, and user feedback loops, moving beyond standalone image generation to a structured service architecture.
Technical Positioning: It provides a transparent, non-overstated evaluation of WGAN-GP and AdaIN for heritage contexts. The author explicitly acknowledges that these methods are not state-of-the-art compared to StyleGAN or Latent Diffusion models but are chosen for their interpretability, stability, and suitability for constrained cultural datasets.
Theoretical Mechanism: The study offers a parsimonious empirical model distinguishing between expert-rated cultural-symbol fidelity (a technical validation) and user-perceived cultural authenticity (a psychological judgment). It demonstrates that service quality drives engagement through a sequence of presence and authenticity, rather than through technology novelty alone.

Results

Algorithmic Performance: The combined WGAN-GP + AdaIN model achieved a Fréchet Inception Distance (FID) of 78.3 and a Kernel Inception Distance (KID) of 0.041, outperforming DCGAN and standalone WGAN-GP baselines.
Expert Fidelity: Expert evaluation yielded a mean score of 5.2/7. The author interprets this as "moderate/qualified plausibility," suitable for guided educational exploration under expert curation, but insufficient to replace traditional craft or claim high-fidelity reproduction.
User Experiment: Participants in the AI-XR condition reported significantly higher perceived digital heritage service quality, personalization, and visual-generative quality compared to 2D and XR-only conditions.
SEM Findings: All hypothesized paths were supported ( $p < .001$ ). Perceived service quality positively affected immersive presence ( $\beta=0.55$ ), which in turn influenced perceived cultural authenticity ( $\beta=0.34$ ), leading to cultural engagement ( $\beta=0.46$ ), and finally predicting continuance intention and heritage transmission intention.
Mediation: Serial indirect effects were statistically significant but small in magnitude, indicating that while the service quality initiates the chain, cultural transmission intention is a distal outcome requiring more than short-term exposure.

Significance and Claims
The paper claims its primary significance lies in shifting the focus from "image novelty" to "accountable AI-XR service design" for living heritage.

Modest Technical Claims: The author explicitly states they do not claim a new image-generation benchmark. The contribution is the integration of a lightweight, constrained generative module into a heritage service architecture.
Bounded Feasibility: The study concludes that AI-generated Batik motifs are not yet mature enough to replace human curation or traditional authority. Instead, they are viable as tools for "guided educational exploration" when accompanied by cultural explanation, expert curation, and transparent labeling.
Cultural Constraints: The findings emphasize that generative quality alone is insufficient; meaningful engagement depends on culturally constrained validation and the user's perception of authenticity. The system is designed to support, not replace, community-based transmission and craft knowledge.
Service Logic: The research reframes digital heritage from static documentation to a dynamic data service where generation, curation, and user feedback operate in a cyclic loop to ensure cultural acceptability.

Technical Summary: Generative-AI-Enhanced XR Data Services for Intangible Cultural Heritage

More like this