
This paper presents a hybrid workflow for the artificial intelligence (AI)-assisted three-dimensional (3D) modelling of lost or fragmentarily documented architectural heritage assets from sketches, historical photographs and partial graphic references. The approach addresses cases in which conventional photogrammetry or 3D scanning cannot provide a complete record because the object has disappeared, survives only partially or is documented through indirect sources. The methodology is applied to three domestic heritage elements associated with the House of Amaro Pargo in Tenerife, Spain: a lookout window, an oven and a wooden trough. The workflow combines documentary interpretation, AI-assisted image generation and editing, image-to-3D mesh generation, manual geometric correction, texturing at the visual-reference stage and contextual integration in digital or physical outputs. The case studies compare different generative tools and evaluate the resulting models as interpretative assets rather than as metrically validated reconstructions, according to their correspondence with visible source features, visual plausibility, geometric and technical usability, mesh closure and suitability for digital visualisation or 3D printing. The results show that generative AI (GenAI) can support the production of preliminary 3D hypotheses, especially when the available evidence is incomplete and a complete metric restitution is not feasible. However, they also reveal limitations related to documentary uncertainty, variability between tools, hidden or ambiguous geometries, mesh errors and the need for manual review. The paper concludes that AI should not be understood as a substitute for rigorous surveying, photogrammetry, 3D scanning or expert modelling, but as an intermediate and revisable tool within a supervised workflow. Its main contribution lies in documenting the practical possibilities, recurrent limitations and manual interventions involved in transforming AI-generated outputs into usable 3D assets for the interpretation and visualisation of domestic heritage from fragmentary sources.
This article examines how artificial intelligence (AI), especially generative AI, is reshaping architectural design, historic buildinginformation modelling (HBIM), and digital reconstruction in cultural heritage contexts. It clarifies a central distinction: AI outputs may be visually persuasive, but they become architectural or heritage knowledge only when they can be validated against historical evidence, spatial and material constraints, and conservation requirements. The article brings togethertwo bodies of literature: generative AI in architectural representation, early design, spatial exploration, and studio practice; and AI-supported HBIM, scan-to-HBIM, semantic enrichment, digital twins, and evidence-based reconstruction. Methodologically, this study is a thematic narrativeliterature review and conceptual synthesis, not an empirical case study, systematic review, or bibliometric analysis. The literature is analysed through three lenses: design stage, information form, and epistemic claim. The review shows that generative AI is most visible in conceptual design, visualisation, prompt-based exploration, and rapid production of alternatives. By contrast, AI-supported HBIM research is concerned mainly with point-cloud processing, semantic segmentation, scan-to-HBIM automation, ontology-based enrichment, data quality, and heritage-management workflows. This comparison reveals a persistent gap between thespeed of representation and thereliability of information. In heritage contexts, the gap is critical because HBIM models and reconstructions can guide conservation decisions, interpretation, and public memory.The article concludes that AI-supported heritage modelling should be assessed usingevidence, uncertainty, authenticity, provenance, metadata, paradata, and expert validation, rather thanvisual plausibility alone. Its original contribution is to frame evidence-based generative HBIM as an operational validation workflow. In this workflow, AI-generated elements remain provisional until they are linked to sources, marked by uncertainty, checked againstsurvey data, archival evidence, typological and material knowledge, and expert judgement, and integrated into HBIM as traceable information rather than as anautonomous reconstruction.
This article presents the findings of the first comprehensive project that applied virtual archaeology techniques to document, reconstruct in three dimensions, and ethically manage pre-Columbian human remains housed at the National Museum of Archaeology (MUNARQ) in La Paz, Bolivia. The project digitized 47 individuals from the MUNARQ anthropological collection, including 16 complete mummified individuals and 31 human skulls with intentional cranial modification and trepanation. The non-invasive methodology employed high-resolution photogrammetry techniques. Data capture was performed in RAW format using a full-frame Nikon D750 with a fixed 50 mm lens, tripod mounting, a manual rotating turntable, and LED lighting. For each specimen, about 250 photographs were acquired along concentric circular paths at multiple heights and angles to ensure full surface coverage and stable correspondence. Three-dimensional (3D) reconstructions were generated through Structure-from-Motion (SfM) and Multi-View Stereo (MVS) processing in Agisoft Metashape and RealityScan software, followed by metric scaling based on direct measurements. Dissemination-ready models were optimized through controlled decimation and texture baking to produce visually consistent versions with 4K textures and reduced polygon counts suitable for web visualization. Optimized meshes were standardized at about one hundred thousand polygons. The resulting dataset comprises 11750 processed images and about one terabyte of models, textures, and documentation, providing a permanent record of the current state of preservation and reducing the need for repetitive physical handling to ensure the collection’s long-term preservation. For dissemination, MUNARQ implemented a differential access protocol on Sketchfab, publishing crania openly while classifying complete mummies as restricted content that requires explicit user consent to view. We conclude that 3D archiving, combined with ethical governance and sustainable digital preservation practices, strengthens preventive conservation and enables responsible research and “democratic access” to sensitive anthropological heritage in Latin America.
The digitisation of cultural heritage through photogrammetry has transformed the preservation and dissemination of artworks. However, this technique can produce three-dimensional (3D) models with excessively complex geometry, making them inefficient for real-time applications in virtual reality (VR), on mobile platforms, and in web-based viewers. This limitation restricts access to digital heritage, particularly for users with limited hardware and those in regions with technological restrictions. To address this limitation, this study explores optimisation through retopology, using three 3D software tools, such as Instant Meshes, Blender, and ZBrush. These tools were selected because of their ability to reduce the number of polygons without a significant loss of detail. Their effectiveness was subsequently evaluated through a comparative analysis. The study focuses on the optimisation of 3D models of artworks from the Cathedral Museum of Santiago de Compostela: Santiago Crowned, David, and Christ in Majesty. A mixed-methods approach is used to evaluate reductions in mesh size and polygon count, as well as rendering performance, across the different optimisation workflows. The results identified Instant Meshes as the best-performing tool, as it achieves an effective balance between automation and geometric preservation. Blender ranked second, as it offers more precise control for complex models. Moreover, none of the optimised models showed visual degradation on web platforms such as Sketchfab or in Unreal Engine 5, which ensured visual authenticity and compatibility with mobile devices. Finally, this research highlights the role of 3D optimisation in democratising access to cultural heritage by enabling the creation of lightweight and high-quality digital replicas that institutions can use to disseminate art globally, support education, and safeguard artworks.
This study evaluates the performance of three-dimensional Gaussian Splatting (3DGS) for the documentation and virtualisation of architectural cultural heritage. Six datasets acquired in different areas of the Mosque–Cathedral of Córdoba were analysed, representing a variety of capture strategies and scene scales. The datasets were obtained using different acquisition devices, including a drone, a Digital Single-Lens Reflex (DSLR) camera, a mobile phone, and a 360° camera. The images were photogrammetrically orientated using a Structure-from-Motion (SfM) workflow in RealityScan, and the estimated camera intrinsics and poses were subsequently exported to COLMAP format. The datasets were processed using 3DGS, and the training behaviour was analysed across two hardware configurations (RTX3070Ti and RTX5070 GPUs). The study evaluated reconstruction quality and convergence behaviour using three complementary image quality metrics: Peak Signal-to-Noise-Ratio (PSNR), Structural Similarity Index Measure (SSIM), and Learned Perceptual Image Patch Similarity (LPIPS). Additionally, training time and computational efficiency were analysed to determine how scene complexity and hardware configuration affect the optimisation process. The results show that most of the improvement in perceptual quality occurs during the early stages of training. In the majority of cases, 90–95% of the perceptual improvement is achieved between 10000 and 20000 iterations, while the remaining iterations contribute mainly to minor visual refinements. The analysis also demonstrates that the hardware configuration does not significantly influence the final reconstruction quality, although it strongly affects computational performance in large scenes due to the progressive densification of Gaussian primitives during training. Overall, the results suggest that the number of training iterations can be significantly reduced without substantial loss of perceptual quality. These findings highlight the potential of GS as an efficient methodology for the digital documentation and visualisation of architectural heritage, combining high visual fidelity, relatively short training times, and the capacity for interactive rendering for research, dissemination, and educational applications.