Researchers have developed an artificial intelligence model that generates panoramic images for virtual tourism. Called SGMSD-Net, the system is designed to address a limitation of AI-generated tourism imagery: scenes can appear convincing without accurately representing the location they depict. The findings are reported in the International Journal of Information Quality.
SGMSD-Net uses semantic guidance to interpret the meaning of a scene and diffusion modeling to create images through progressive refinement. According to the research team, the model works across three resolution stages, first establishing the image’s overall structure and then adding details such as textures and lighting.
Tests using the SUN360 and Matterport3D datasets found that SGMSD-Net performed better than four existing approaches in image quality, perceptual similarity and consistency with textual descriptions. Subjective assessments also indicated that the generated scenes were clear and visually consistent.
The model could support virtual tours that let people explore destinations before travelling. The researchers also suggest applications in educational reconstructions of historical sites that are difficult to access or require protection from tourism to support conservation efforts.
Future work will examine how the model could be combined with three-dimensional reconstruction and multimodal artificial intelligence systems. These technologies could process different forms of information to create more interactive representations of virtual destinations.
The study, “Semantic-guided multi-scale diffusion network for immersive tourism content generation,” was authored by Lun Wei and colleagues and published in the International Journal of Information Quality in 2026. DOI: 10.1504/ijiq.2026.157237
Sources: Phys.org.