Wu Aihua was not designed as a quick content experiment. The director initially treated her as a "touchstone", a 10 — to 20-second test to explore a new idea. But once the demo appeared on RedNote in September last year, messages began flooding in, urging him to expand Wu Aihua's world.

"I kept revising her," he says. "She was finished very early, but I kept optimizing. Adding even 30 more seconds is exponentially more work than people think."

At the core of those revisions was character. The director insists his virtual projects must be story-driven, not aesthetic showcases. "I don't want Wu Aihua's music videos to belong to a single style," he says. "They need a world. Who is she? What kind of world does she live in?

"Current AI models still struggle with complex fight choreography," the director notes, adding that he plans to incorporate more AI-generated characters into the stories of Wu Aihua's lore. "If you push too hard, the emotion collapses."

Instead, he turned to pose-based action logic, inspired by Jackie Chan, Donnie Yen, and Bruce Lee films, as well as by Japanese animation such as Dragon Ball. Each strike becomes a pose; each pose, a storyboarded frame. The result is movement that feels continuous without needing hyper-realistic combat.

Wu Zhiqi is careful to demystify AI's role in the process. "AI isn't magic," he says. "It's difficult. You need judgment, taste, and knowledge. Otherwise, nothing works."

Trained in graphic design and deeply immersed in different music genres, Wu approaches AI as an extension of his aesthetic judgment. "If I didn't have my own sense of beauty, I couldn't have created Wu Aihua," he says.

He rejects the idea that AI replaces creators. Instead, it forces them to learn more: storytelling, editing, rhythm, and pacing.

"To hold someone's attention for over three minutes — whether they love it or hate it — you have to understand cinema," he says, citing Stephen Chow films as a key influence.

Looking ahead, the Wu Aihua sensation is only beginning. The director is currently experimenting with multilingual tracks, including a bold collision of Hakka dialect and Spanish, and even African languages.

Some of these experiments are very challenging. "AI can't accurately reproduce Hakka," he says. "I've spent over a month trying to re-create its rhythm using English. It's extremely hard."

The goal, however, remains constant: to build Wu Aihua not as a gimmick but as a living IP — one that could eventually include friends, enemies and an expanding narrative universe across animation, music and visual storytelling.

"Everyone is making virtual humans now," he says. "That's boring. If the world isn't different, why should anyone care?"