AI Recreates Wang Xianzhi's Ancient Love Scroll From 250 Surviving Characters

From scattered fragments to a scroll of 919 characters

About 250 surviving characters have become the foundation for a digital reconstruction of one of China's celebrated works of calligraphy. Peking University's Centre for Chinese Font Design and Research announced that it had completed a continuous scroll of 919 characters recreating Wang Xianzhi's rendition of Rhapsody on the Luo River Goddess, known as Luoshen Fu.

The project brings together calligraphic font design and artificial intelligence to fill gaps in a work dating back roughly 1,600 years. Researchers studied the surviving characters, then used a system that generates characters to extend the calligrapher's style across the complete poem.

Wang created his version in the fourth century, during the Eastern Jin era. Its graceful movement and expressive use of ink made it famous, but only fragments survived. The university's reconstruction gives readers a continuous version of a work previously encountered through incomplete remains.

The result is a digital scroll, with the surviving writing providing the basis for newly generated characters. Researchers examined stroke weight, brush rhythm, character structure and spacing to reproduce the visual relationships that give the work its flowing appearance.

Roughly a quarter of the calligraphy remained legible across dispersed fragments. Using the approximate count of 250 surviving characters against the completed total of 919, those surviving forms account for about 27 percent of the reconstructed scroll. The scale of the missing material made the task much larger than filling a few small gaps.

The project also connects two different artistic achievements: Cao Zhi's poem about an encounter with a river goddess and Wang Xianzhi's later interpretation of that poem through brushwork. Understanding both helps explain why reconstructing the writing involves more than recovering readable words.

The poet and the calligrapher behind Luoshen Fu

Cao Zhi, who lived from AD 192 to 232, was a prince of the state of Cao Wei during the Three Kingdoms period and a celebrated poet of its court. He wrote Luoshen Fu around AD 222 after a journey near the Luo River.

The prose poem describes a meeting with the river goddess, presenting her as an ideal of beauty and virtue. Its blend of courtly romance and spiritual longing inspired painters and calligraphers for centuries. Versions of Luoshen Fu became a touchstone of literary and artistic culture.

Wang Xianzhi, who lived from AD 344 to 386, served as an official and became one of China's greatest calligraphers. He was the son of Wang Xizhi, another towering figure in the art. Father and son are known together as the Two Wangs.

The pair refined running script and cursive script, forms of writing that give brush movement and connections between strokes a central role. Wang Xianzhi developed an elegant rhythm and experimented with character structure, linked strokes and placement to create a dynamic flow. Later dynasties studied his work as a standard of taste.

The main dates place the poem, the calligraphy and a later heritage project in their historical settings:

• Around AD 222: Cao Zhi wrote Luoshen Fu after a journey near the Luo River.
• Fourth century: Wang Xianzhi created his celebrated calligraphic rendition during the Eastern Jin era.
• 2019: A collaboration at Yongle Palace in Shanxi began using AI tools to revive the colors of a Yuan dynasty mural.

Many works associated with the Two Wangs survive through copies and rubbings. Gaps in that historical record have left important pieces incomplete, making the surviving forms especially valuable for studying how the artists wrote.

How researchers taught a system to follow the brush

Peking University's team began with a close examination of approximately 250 authentic characters. Researchers measured the direction and curvature of strokes, changes in pressure, the weight of ink and the proportions of the smaller components, called radicals, within each character.

The analysis also considered how Wang balanced dense areas of ink with open space. A character's appearance depends on those internal relationships, while the movement of a whole passage depends on the size, placement and spacing of successive characters.

The system learned patterns in how strokes began and ended, how lines tapered and how characters occupied their writing space. It also studied how Wang curved and pressed his brush and linked characters in running script.

A generative model then used those patterns to produce candidates for characters missing from the surviving fragments. In simple terms, the software studied examples of the calligrapher's hand and proposed how other characters could look when written in a similar manner.

The task went beyond brightening faded ink or making an existing image easier to see. It required producing absent written forms and arranging them into a continuous composition whose rhythm followed the surviving passages.

Human judgment shapes the finished scroll

Experts reviewed the proposed characters, adjusted the system's parameters and selected versions that best matched Wang's hand. This process, often called keeping a human in the loop, places people inside the cycle of generation and revision.

The review helps control mechanical repetition and stylistic drift, where generated writing gradually moves away from the qualities of the surviving examples. Each character must work on its own while also fitting the movement of the surrounding text.

The completed layout follows the spacing and rhythm of the surviving lines. Readers can encounter the poem as a cohesive scroll while understanding which passages preserve surviving forms and which have been reconstructed.

Peking University described the project in a public statement as an experiment joining traditional art with contemporary tools:

This endeavour vividly illuminates the fusion potential of AI plus culture, where traditional calligraphic art gains fresh modes of expression in the digital age.

The account leaves some practical details open: it gives no release date, public access arrangements or schedule for further review. The announced achievement is the completion of the digital rendition.

Greek inscriptions show the value of working together

The Luoshen Fu project belongs to a broader wave of AI work on damaged historical writing. Deep neural networks, systems that learn patterns from examples, have helped scholars suggest missing text, identify where inscriptions originated and estimate when they were made.

Ithaca, a system trained on ancient Greek inscriptions, provides a numerical example of how software and historians can complement each other. The model alone restores damaged text with accuracy of around 62 percent.

Historians working without its assistance achieved accuracy of about 25 percent in the comparison described. When they used Ithaca as a guide, their accuracy rose to roughly 72 percent, an increase of about 47 percentage points.

Ithaca also suggests likely places of origin and time windows for inscriptions. Those suggestions give researchers a starting point for examining historical context alongside the surviving words.

The relationship to Luoshen Fu lies in the working method: software generates possibilities, and specialists assess them. For the Chinese scroll, the central task is reproducing a calligraphic style; for damaged inscriptions, it includes deciding which words belong in a gap and where the text fits historically.

Latin inscriptions and sealed Roman scrolls

Google DeepMind's Aeneas extends AI assistance to Latin inscriptions. It combines text and images, searches large collections for comparable inscriptions and proposes reconstructions for longer missing passages.

Tests with historians found that Aeneas improved restoration and attribution tasks and was useful in most cases. Attribution means identifying such details as where an inscription came from or the period in which it was produced.

Historians weigh the system's proposals against context, style and material evidence before accepting a change. That keeps interpretation with people who can examine how a suggested reading fits the object and its history.

The carbonized Herculaneum scrolls present a different challenge. Sealed since AD 79, these fragile papyrus rolls contain writing that researchers seek to read without physically unrolling them.

Advanced imaging and machine learning help detect faint letters inside the rolls. Scanning provides information from within the material, while AI assists with finding traces of text. Careful validation and open methods help maintain trust in the resulting readings.

Chinese paintings and murals become digital test cases

Chinese universities and cultural institutions are also applying AI to paintings and murals. At Zhejiang University, a team led by art and archaeology scholar Tang Tan developed a deep learning model that connects information from poems and paintings.

The approach is described as cross modal because it works across different forms of information, in this case language and images. By learning how classical poetry describes landscapes and colors, the model proposes palettes suited to the intended appearance of pigments that have been lost.

The researchers demonstrated fast, reversible recoloring on works including Dong Yuan's Residents on the Outskirts of Dragon Abode. Conservators can use such digital versions to preview and discuss color choices before taking any physical action.

At Yongle Palace in Shanxi, a collaboration launched in 2019 used AI tools to revive the faded colors of Chaoyuan Tu. The Yuan dynasty mural stretches nearly one hundred meters and depicts Taoist figures arranged in a layered procession.

Digital enhancement allows visitors to see the scale and detail of the composition with a clarity closer to its original visual impact. The work takes place in the digital image, avoiding direct intervention on the painted wall.

These projects share a practical advantage with the reconstructed calligraphy: they offer ways to view and debate an altered appearance while keeping the surviving physical object untouched.

Classic martial arts films enter the same debate

The Kung Fu Movie Heritage Project applies AI tools to another part of China's cultural memory. Companies involved plan to digitally restore one hundred classic martial arts films, including Bruce Lee and Jackie Chan staples as well as later hits.

The initiative promises cleaner images and sound while keeping the stories intact. It has also inspired a fully animated reinterpretation of A Better Tomorrow using AI tools throughout the production pipeline.

Tian Ming, chair of Canxing Media and a leader in the initiative, described the relationship between technology and artistic decision making:

AI is the brush, but creativity is the soul," Tian said. "Classic kung fu films embody China's spiritual backbone.

The film projects place technical restoration and new creative interpretation alongside each other. Cleaning an existing film, creating an animated version and completing missing calligraphy each involve different choices about what audiences will see.

Keeping the surviving work visible

Wang Xianzhi's reconstructed scroll makes transparency central to the viewing experience. The new characters are a continuation based on the surviving style, and clear labeling lets audiences follow where the historical material ends and the reconstruction begins.

Generative systems can favor neat symmetry or repeat shapes that a master calligrapher would vary. Safeguards include limiting training material to authentic fragments, applying rules about stroke order and comparing results with verified examples.

Calligraphers, historians and conservators bring different forms of judgment to that review. Their checks can examine the shape of individual strokes, the historical fit of proposed forms and the relationship between the reconstruction and the surviving object.

Responsible digital restoration can also make its decisions easier for others to inspect through practical measures:

• Identify reconstructions and record the team, date and tools used.
• Publish views of surviving fragments alongside proposed completions and the full scroll.
• Use metadata and visual watermarks to identify generated characters.
• Invite independent experts to review the work and retain a record of revisions.
• Share data and code where possible so other teams can test the methods.
• Release alternative versions when historical uncertainty allows more than one reading.
• Clarify rights and licensing for education and research.

These measures give readers a way to understand the choices behind a finished image. They also allow later research to revise a reconstruction while preserving a record of earlier decisions.

A new way to read the poem and the hand

Luoshen Fu brings literary meaning and physical gesture together. Chinese calligraphy treats writing as an art in which breath, wrist movement and ink leave a record of the mind. Wang's interpretation gives Cao Zhi's poem both literary elegance and personal feeling.

A complete digital scroll offers schools and museums a way to present the poem's full narrative while explaining the surviving evidence. Interactive displays can move from the sweep of the whole composition to the thickness of a single stroke.

Readers can encounter the poem's longing and grace while learning how researchers infer missing forms. Documenting methods, inviting critical review and updating results can make the reconstruction a resource for study as well as a tribute to Wang's art.

The Essentials

• Peking University's Centre for Chinese Font Design and Research completed a digital rendition of Wang Xianzhi's Luoshen Fu.
• The scroll contains 919 characters and draws on about 250 surviving authentic characters.
• Cao Zhi wrote the poem around AD 222; Wang created his calligraphic rendition in the fourth century.
• Researchers combined font design, AI character generation and expert review to recreate missing writing and its visual rhythm.
• Ithaca, Aeneas and work on the Herculaneum scrolls show other uses of AI in studying historical texts.
• Chinese projects also use AI to recolor paintings and murals and restore classic martial arts films.
• Clear labels, documented decisions and independent review help audiences understand reconstructed cultural works.