What is it about?

Across the world, many historic buildings and monuments have been damaged or destroyed by war, natural disasters, and human activity. Once these structures are lost, it is often impossible to fully understand how they originally looked or how people experienced them. This research explores how modern artificial intelligence (AI) tools can help bring these lost places back into view. The study uses text‑to‑image AI, a type of technology that can create detailed pictures from written descriptions. By carefully analysing historical texts, archaeological reports, and architectural records, the researchers created detailed written prompts that describe how heritage sites once looked. These prompts were then used to generate realistic images of monuments that are now partially or completely destroyed. The research demonstrates this approach using well‑known heritage sites such as Palmyra in Syria, Pompeii in Italy, the Buddhas of Bamiyan in Afghanistan, and ancient Maya cities in Mexico. The AI‑generated images were compared with historical evidence and expert knowledge to check how accurate and believable they were. Rather than replacing historians or archaeologists, this approach combines AI with traditional heritage research. It offers a new way to visualise the past, support education, and help future conservation efforts by making lost heritage easier to understand and imagine.

Featured Image

Why is it important?

This work is timely because cultural heritage is being lost at an unprecedented rate due to conflict, climate change, and urban development. Traditional reconstruction methods rely heavily on photographs or surviving structures, which are often incomplete or unavailable. What makes this research unique is its use of written historical descriptions as the primary source for reconstruction, rather than images alone. By showing that AI can turn carefully researched text into credible visual reconstructions, the study opens new possibilities for preserving heritage digitally—even when physical evidence is scarce. The approach also introduces a structured methodology that combines AI image generation, expert review, and quantitative image evaluation. This makes the process more transparent and reproducible, helping build trust in AI‑assisted heritage conservation. In the long term, this work could support museums, educators, policymakers, and conservation professionals in protecting and sharing cultural heritage with wider audiences.

Perspectives

Working on this research was particularly meaningful because it sits at the intersection of engineering, artificial intelligence, and cultural heritage. Seeing AI used not just for efficiency or automation, but to help reconnect people with lost history, was deeply motivating. One of the most rewarding aspects was discovering how much historical detail can be unlocked through careful reading of texts and collaboration across disciplines. This work reinforced the idea that AI is most powerful when it supports human expertise rather than replacing it. I hope this publication encourages further collaboration between engineers, historians, archaeologists, and digital humanists, and sparks thoughtful discussion about how emerging technologies can be used responsibly to protect our shared cultural memory.

Prof Tatiana Kalganova
Brunel University

This publication was an important early step in my PhD research on AI-assisted digital heritage reconstruction. As first author, I was interested not only in whether text-to-image AI could generate visually convincing images, but whether it could be used in a more disciplined way to support the reconstruction of damaged or lost heritage from historical descriptions, archaeological records, and architectural evidence. The main challenge was methodological. Heritage reconstruction is not the same as ordinary image generation. A visually impressive output is not enough. The process must remain connected to evidence, expert interpretation, and transparency about what is known, what is uncertain, and what is speculative. This paper allowed us to explore that challenge through case studies including Palmyra, Pompeii, the Buddhas of Bamiyan, and ancient Maya cities. The work also shaped the direction of my later research. It made clear that AI-generated reconstructions should not be treated as final or authoritative replacements for lost monuments. They are better understood as visual hypotheses: images that can support discussion, education, conservation, and further expert review. For me, the value of this publication lies in showing both the promise and the responsibility of using generative AI in cultural heritage. AI can help us imagine and communicate the past, but it must be guided by historical evidence, human expertise, and respect for the cultural meaning of damaged or destroyed sites.

Kawsar Arzomand
Brunel University

Read the Original

This page is a summary of: From ruins to reconstruction: Harnessing text-to-image AI for restoring historical architectures, Challenge Journal of Structural Mechanics, June 2024, Tulpar Academic Publishing,
DOI: 10.20528/cjsmec.2024.02.004.
You can read the full text:

Read

Contributors

The following have contributed to this page