In July 2026, the World Internet Conference released its first-ever Cultural Heritage Digitalization Case Collection — a curated selection of 45 projects from 13 countries. China contributed projects ranging from AI-powered mural restoration at the Yungang Grottoes to satellite-assisted detection of 80,000 ancient underground irrigation wells in Xinjiang's Turpan Basin. What makes this noteworthy is not the individual projects, but the scale: China is now deploying artificial intelligence, 3D scanning, cloud computing, and multispectral imaging across hundreds of heritage sites simultaneously — creating what is arguably the world's largest coordinated effort to digitize cultural heritage.
The timing is not coincidental. China's 15th Five-Year Plan (2026–2030) explicitly prioritizes "high-quality cultural development" and "preservation of national memory" through digital means. The National Cultural Heritage Administration (NCHA) released its first formal regulations governing state-owned cultural relic data in July 2026, establishing a tiered licensing system that allows commercial entities — from game developers to film studios — to access standardized digital heritage resources under negotiated terms. The policy framework is catching up to the technology, and the technology is moving fast.
Why AI Is Needed: The Race Against Decay
Cultural heritage faces a universal problem: everything decays. Stone erodes, pigments fade, wood rots, and ancient manuscripts crumble. For centuries, the only defense was physical conservation — climate-controlled rooms, restricted access, chemical treatments. But these methods are expensive, slow, and ultimately temporary. A 1,500-year-old cave mural does not stop aging just because you put it in a museum.
Digital preservation offers a fundamentally different approach. Instead of trying to stop time, it captures heritage in such extraordinary detail that future generations can study it — even if the physical original eventually degrades. AI supercharges this process by automating tasks that once required years of manual labor: detecting microscopic cracks in stone, reconstructing missing fragments from partial data, translating ancient scripts, and generating navigable 3D environments from flat photographs.
Consider the math. The Dunhuang Mogao Caves contain over 45,000 square meters of murals and more than 2,000 painted sculptures spanning a millennium of Buddhist art. Manually documenting every square centimeter at preservation-grade resolution would take decades. With AI-assisted multispectral imaging and automated stitching algorithms, the Dunhuang Academy has now accumulated over 600 petabytes of digital heritage data — a volume that would be impossible without machine learning pipelines to process, classify, and index the raw scans.
The Technology Stack: How It Actually Works
China's digital heritage efforts draw on a layered technology stack that combines several AI disciplines:
1. Multispectral and Hyperspectral Imaging
At the Yungang Grottoes in Shanxi Province — home to 51,000 Buddhist statues carved between the 5th and 6th centuries — researchers have deployed self-developed 3D hyperspectral equipment that goes far beyond conventional photography. The system captures data across dozens of spectral bands, revealing pigment compositions, subsurface cracks, and water infiltration patterns invisible to the naked eye. AI algorithms then analyze this spectral data to predict where deterioration will occur next, shifting conservation from reactive repair to preventive intervention.
2. LiDAR and Drone Photogrammetry
In the densely forested hills of Fujian Province, traditional archaeological surveys were nearly impossible — vegetation obscured the terrain, and optical imagery could not penetrate the canopy. A team from Xiamen University deployed drone-mounted LiDAR to map sites beneath the tree cover, using deep learning to classify terrain features and identify archaeological anomalies from the point cloud data. The same technique was applied in Turpan, where AI-assisted satellite imagery and object detection algorithms located over 80,000 karez well shafts — an ancient underground irrigation system — that had never been comprehensively mapped before.
3. 3D Reconstruction and Digital Twins
The most visually striking application is full-scale 3D reconstruction. The Yungang Grottoes Academy, in collaboration with Zhejiang University, produced the world's first 1:1-scale 3D-printed mobile replica of an entire cave (Cave No. 12). The replica — which has been exhibited in Shanghai and at the China Cultural Center in Paris — is accurate to sub-millimeter precision. Meanwhile, the Digital Yuanmingyuan project uses imperial archives and historical photographs to create hyper-detailed virtual models of the Old Summer Palace, destroyed in 1860. Visitors at the physical ruins can use AR glasses to see lost pavilions layered over surviving stone foundations.
🔑 The Dunhuang Standard
The Dunhuang Academy has been digitizing the Mogao Caves for over 30 years — long before "AI" became a buzzword. Today, its 600+ petabyte archive is governed by blockchain-based traceability systems that track every data licensing transaction. The academy has established standardized workflows covering data collection, professional processing, secure distributed storage, and classified external authorization. When a game developer or film studio wants to use Dunhuang's digital assets, the blockchain ledger records exactly what was used, by whom, and for what purpose. This model is now being adopted as a national template under the NCHA's new regulations.
From Preservation to Commercialization: The BoGuan Model
In September 2025, Huawei and the Shaanxi Culture Industry Investment Group (SCG) unveiled BoGuan — the world's first commercial multimodal large language model (LLM) dedicated to cultural tourism. Trained on a dataset of 1.2 petabytes — including 31 million images, 4.4 million minutes of video, 2.18 million minutes of audio, 510 3D models, and 960 million pieces of structured text — BoGuan represents a shift from pure preservation to active commercialization.
The model powers an AI travel companion agent available through the GO-SHAANXI app, which had reached over 4 million users by March 2026. It also generates digital IP for intangible cultural heritage — including a cartoon character called Tang Biaobiao, designed by integrating local heritage elements with stone carvings of the Six Steeds of Zhao Mausoleum. Sales of related digital collectibles and creative products have exceeded 2 million yuan (approximately $275,000).
Zhang Beiyuan, a dough sculpture artisan in Xi'an, reported that using BoGuan reduced the time to complete a traditional sculpture from two to three months down to less than a week. The model helps generate design references, suggest color palettes based on historical data, and simulate the final product before physical work begins. This is not AI replacing craftsmanship — it is AI compressing the research and iteration phases that traditionally consumed most of an artisan's time.
Tencent's Jingdezhen Project: Gaming Meets Heritage
In July 2026 — coinciding with the UNESCO inscription of Jingdezhen's Handicraft Porcelain Industry Sites — Tencent launched Digital Jingdezhen: Porcelain Craft Adventure, an AI-powered cultural heritage game. The project uses AI-assisted Procedural Content Generation (PCG) to create large-scale digital recreations of Jingdezhen's five major porcelain heritage sites and historical production scenes.
The underlying dataset is substantial: over 30,000 documentary records, more than 5,000 ceramic gene specimens, and data on nearly 1,000 representative ceramic artifacts from collections worldwide. Tencent applied Optical Character Recognition (OCR), Natural Language Processing (NLP), and knowledge graph technologies to transform decades of scattered historical records, heritage documentation, and craft specifications into a structured, queryable digital resource.
What makes the Jingdezhen project different from a typical museum app is the concept of "living heritage." Users do not just browse static exhibits — they participate in virtual porcelain-making, learning the 72-step traditional process through interactive gameplay. The AI companion provides contextual explanations, and the game's procedural generation ensures that no two visits are identical. This is a deliberate strategy: rather than treating heritage as a frozen artifact to be observed from a distance, it treats it as a living tradition that can be experienced and carried forward.
💡 The Black Myth: Wukong Effect
When the video game Black Myth: Wukong became a global hit, it did more than sell copies — it triggered a surge of tourism to the real-world temples and grottoes digitally reproduced in the game. The Shanxi Province tourism board reported a 50% increase in visitors to sites featured in the game. NCHA officials now cite Black Myth: Wukong as proof that "authoritative cultural data is the genuine wellspring of cultural innovation." The game's success demonstrated that high-fidelity digital heritage assets have commercial value far beyond academic archives — they can drive tourism, inspire creative industries, and introduce global audiences to cultural sites they might never have discovered otherwise.
Can AI Really Capture Intangible Heritage?
For all the impressive numbers, AI faces a hard limit when it comes to intangible cultural heritage — the practices, rituals, oral traditions, and embodied knowledge that cannot be reduced to pixels and polygons.
At Hongcun, an 800-year-old village in Anhui Province and a UNESCO World Heritage site, researchers mapped the ancient water system using drone thermal imaging — but the project only succeeded because they also interviewed seven senior residents who still recite the hydraulic maintenance chants passed down since the Southern Song dynasty. The resulting AI model does not just locate pipe blockages; it surfaces seasonal maintenance windows validated by both infrared anomalies and ritual calendars. The algorithm found the flaw; the elders knew when to fix it.
Similarly, in Lijiang, Naxi elders retain contractual veto power over every AI-generated transcription of Dongba script inscriptions before inclusion in the UNESCO-endorsed digital archive. During the Qixi Festival in Zhouzhuang, AI-powered audio analysis of 300 hours of field recordings identified subtle tonal shifts in love-song performances across three generations — but the insight was only useful because it informed a co-designed workshop where elders taught teens vocal techniques while engineers helped them record variations in real time.
The pattern is consistent: the most effective AI heritage projects in China embed anthropologists, folklorists, and community elders before algorithm design begins. The technology captures what can be measured; the community provides what cannot. Together, they produce something neither could achieve alone.
Global Implications: A Template for Heritage Preservation?
China's approach to AI-driven heritage preservation is attracting international attention — not because it is unique, but because it is systematic. The 2026 World Internet Conference case collection included projects from Italy, Spain, and Germany alongside Chinese ones, and the underlying challenges are universal. Every country with ancient heritage faces the same problems: decay, funding shortages, and the tension between preservation and public access.
What distinguishes China's approach is the combination of three factors: government policy coordination (the 15th Five-Year Plan and NCHA data regulations), major technology company investment (Tencent, Huawei, Alibaba), and a willingness to commercialize digital assets to fund further preservation. The tiered licensing system — separating public welfare, non-profit academic, and commercial use — provides a revenue model that many other countries lack.
Italy's RUME cultural management company has built a similar "four-layer technology architecture plus five-stage implementation process" for digital heritage, and Spain's Selene photometric stereo system is advancing the digitization of flat artifacts like manuscripts and engravings. But neither operates at the scale of China's multi-site, multi-institution deployment. The question for global heritage conservation is whether China's model — large-scale, state-coordinated, commercially funded — can be adapted to other governance contexts.
What the Data Tells Us
| Project | Technology | Scale | Status (2026) |
|---|---|---|---|
| Dunhuang Mogao Caves | Multispectral imaging, blockchain traceability | 600+ PB archive, 30+ years | Ongoing, global licensing |
| Yungang Grottoes | 3D hyperspectral, 1:1 3D printing | 51,000 statues, Cave 12 replica exhibited globally | Active, mobile exhibitions |
| Jingdezhen Porcelain | AI game, knowledge graph, PCG | 30,000+ records, 5,000+ ceramic specimens | Launched July 2026 |
| BoGuan LLM (Xi'an) | Multimodal LLM, 5G-A network | 1.2 PB dataset, 4M+ users | Commercial deployment |
| Digital Yuanmingyuan | AR, 3D virtual reconstruction | Full park virtual model | On-site AR experience |
| Turpan Karez Wells | Remote sensing, AI object detection | 80,000+ wells mapped | Completed, monitoring |
| Tianlongshan Caves | Digital repatriation, 3D reassembly | 240+ stolen statues, 150+ abroad | Ongoing, cross-border |
The Limits Nobody Talks About
For all the progress, there are genuine constraints. A single high-resolution photogrammetry scan of the Forbidden City's Meridian Gate generates 8.2 terabytes of raw data. Processing it into navigable 3D models with semantic segmentation — distinguishing original Qing-era glazed tiles from 1950s repairs — requires GPU clusters and domain-specific training. The bottleneck is not data collection; it is contextual alignment. An AI can detect pigment degradation on Dunhuang murals, but only local conservators know whether that fading correlates with seasonal humidity shifts tied to historic monsoon corridors.
Hardware dependency is another constraint. Real-time dialect translation devices used in remote Tibetan Buddhist monasteries work reliably only within 30 meters of a calibrated microphone array. Edge computing devices need solar charging in off-grid sites. And the digital divide between well-funded UNESCO sites and smaller, less-visited heritage locations remains wide — the same AI tools that work at the Mogao Caves are not automatically deployable at a rural temple with no broadband connection.
There is also the question of what gets digitized and what gets left behind. AI training datasets favor well-documented, visually rich heritage — stone carvings, murals, architecture. Intangible practices — the exact pressure needed to press tofu during a Spring Festival ceremony, the unspoken hierarchy of who pours tea first during ancestral rites — resist capture. These require embodied learning, something no algorithm currently simulates.
"The most compelling cultural experiences emerge where technology recedes — not as magic, but as infrastructure. When an AI model flags a stress fracture in a 1,200-year-old stone bridge, the repair crew arrives with chisels forged by the same family for six generations. The algorithm found the flaw; the craftsperson knows how to mend it." — Field observation, Fenghuang Ancient Town, 2025
Conclusion: More Than a Backup Drive
China's AI-driven cultural heritage preservation is often described in terms of backup — creating digital copies in case the physical originals are lost. But the reality is more interesting. These projects are not just preserving the past; they are creating new ways to experience it, study it, and build businesses around it.
A visitor wearing AR glasses at Yuanmingyuan sees a lost palace reconstructed over ruins. A game developer in Chengdu uses officially licensed digital assets from Dunhuang to build a historically accurate virtual world. A dough sculpture artisan in Xi'an reduces a two-month project to one week using AI-assisted design tools. A student in Nairobi explores the Mogao Caves through a web browser. These are not alternative versions of heritage — they are extensions of it, made possible by the same technology stack that also serves as insurance against decay.
The question is not whether AI can preserve cultural heritage. It clearly can, and China is demonstrating that at scale. The more interesting question is what happens when digital heritage becomes so immersive, accessible, and commercially valuable that it changes how people relate to the physical originals. When a 3D-printed cave replica is accurate to sub-millimeter precision, does it matter whether you saw the original or the copy? When a VR experience of the Forbidden City is more detailed than what a tourist could see in person, does the physical visit become less essential?
These are not problems to solve — they are tensions to manage. China's approach — combining state policy, corporate investment, academic rigor, and community participation — is one answer to that challenge. Whether it becomes a global template depends less on the technology and more on whether other countries can build the institutional frameworks to support it.