A mechanical sea creature in flight, a black cat garbed in a kimono and butterfly flitting through a garden all leap out on the screen. Is there an upcoming ukiyo-e artist adding a 21st-century flair to the genre?

Nope. It’s all generative artificial intelligence. And it was fed 24,038 scans of actual pieces.

The Evo-Ukiyoe system, which was made publicly available on July 22, was developed by Tarin Clanuwat. Simply type in whatever comes to mind and the program will churn out a woodblock print.

“I would like the general public to come into contact with classics in their daily lives,” said Clanuwat, who is an information technology engineer and classical Japanese literature scholar.

Clanuwat aims to boost a modern appreciation for the historical artform with her project. Pitched by Sakana AI, a Japanese startup firm founded by former Google engineers, Evo-Ukiyoe was created by scanning those 24,038 ukiyo-e.

Real works were used in the hope of reproducing not only the iconic color combinations, but also the texture of “washi” paper.

The AI model is exclusively distributed to parties who use it for education, research and development. However, anyone can input prompts to try out the system for free on Sakana AI’s demo website.

Conventional AI models often fail to accurately recreate ukiyo-e’s distinctive style even if the prompt explicitly includes the term “ukiyo-e.” For example, for “a male in the ukiyo-e style,” one tool generated an illustration of a bearded man clad in a woman’s kimono.

This bothered Clanuwat, as did “too many colors” showing up the AI art. Given that ukiyo-e are largely defined by using woodblocks for reproductions, there would have been a limited number of hues available during its heyday.

Clanuwat decided to work on an AI model that could precisely generate ukiyo-e’s authentic color variations and textures.

She sought the cooperation of Ryo Akama, a literature professor at Ritsumeikan University and director of the institution’s Art Research Center. Clanuwat initially connected with Akama while researching.

Clanuwat arrived in Japan in 2008 as a Japanese government-sponsored international student with the aim of researching classical literature. She spent 10 years before finishing a doctoral program at Waseda University.

Drawing on computer programming skills she has refined since she was a fifth-grader in elementary school, Clanuwat built a professional career that includes time in Google’s AI research and development department.

Clanuwat joined Sakana AI in April this year where she devoted herself to creating an AI system that could instantly recognize simplified characters in classical publications.

Once she procured a museum’s-worth of digitized ukiyo-e pieces, she added captions to each one with the help of a generative AI program.

These learning materials were then input into Sakana AI's inhouse image generator developed that same month as additional training for the tool.

As a result, it successfully created an ukiyo-e of a crane beneath a snow-covered tree with the prompt, “a crane standing in a snowy garden.”

Limited hues and an uneven, irregular backdrop of dark blue checked off more requirements typical of woodblock prints.

Evo-Ukiyoe is also able to tap into the genre’s ambience in how it depicts humans.

After entering the prompt, “a man operating a laptop,” the model output a man sporting a topknot and clad in a kimono looking at a computer screen. His thick eyebrows, along with his long and narrow eyes, represent common attributes in ukiyo-e.

Some challenges remain. Evo-Ukiyoe still sometimes depicts men in women’s kimono.

“The AI model has yet to fully understand how to illustrate kimono and hairstyles,” said Clanuwat. “Painting humans is difficult.”

Her policy in her profession, which ended up leading to the completion of the AI ukiyo-e generator, has always been the same.

“My hope is bringing people living in present times closer to classics with the power of AI,” said Clanuwat. “Japan is home to hundreds of millions of carefully preserved classical materials. It would be a waste not making use of them.”

Akama, who provided the ukiyo-e image data, views the AI art generator positively.

“It may prove an excellent educational tool to deepen the understanding of the appeal of the works of Utagawa Hiroshige (1797-1858), if Hiroshige-like images created by AI are compared with real pieces by Hiroshige for discussions on their differences,” he said.

Generative AI’s potential is vast, but so are the real-world complexities it’s embroiled in. Environmental impact, copyright and identity theft connected to voter suppression in the United States are among them.

Many companies view generative AI as a helpful trend in Japan, with Yokosuka Mayor Katsuaki Kamiji even allowing the creation of a version of himself that speaks fluent English.

On the flip side, in another part of Kanagawa Prefecture, a Kawasaki man used it to create computer viruses before he was arrested.

Following the European Union and the United States, the Japanese government plans to submit a bill on regulating “high-risk” AI next year.