KEEPKWAN, AI creator

What was discussed

We started by asking where the name came from. He spent 12 years on the business strategy team at SM C&C, mapping out mid- and long-term plans, and first ran into AI in 2021. It fascinated him, but it also brought a fear that one day it might swallow him whole. So he took his own name, Yoon Seok-kwan, and put "Keep" in front of it because he wanted to keep hold of himself. KEEPKWAN. He said he carries the name around like a manifesto.

The moment AI started to feel like his own field went like this. He was researching where the company's revenue would come from five and ten years out, got curious enough to try it himself, and began working after hours and on weekends. He fed in a novel he had written in classical Chinese back in high school and turned it into images. This was the Midjourney beta era, so the results were crude and grotesque. Even so, the sheer fact that text could come back as something with color in it floored him, and that was when he became convinced this would be a serious weapon in the content business.

Quitting was not a snap decision either. At first he thought it might work as a new business line inside the company, but once he actually started, it was fun, and watching things take shape one by one taught him what freedom felt like. Even after deciding to leave the company he had been at for 12 years, he could not do it right away. He gave himself a year to test whether he could hold up, and only then walked out.

He kept coming back to the point that making things and selling them are two different jobs. Asked why the collaborations kept coming from Disney, the Ministry of Foreign Affairs, the Korea Heritage Service and CJ, he pointed to clear storytelling, a concept, and a world of his own. The big thing, he said, was anchoring himself in Korean traditional culture and carving out a position as the person who blends the traditional with the contemporary.

The work does not start with an AI tool. It starts with a sketch. He checks the historical sources first, and only then moves on to generation. In his university course, the first six weeks are pencil and paper only, so students learn to find material in ordinary life before they touch the technology. He sets his themes a year at a time, too. Last year's was "Wearing": he made an AI hanbok that rendered the dancheong patterns of Gyeongbokgung in neon color and staged a runway show at Gwanghwamun Square. This year's is "Scent", showing what a smell looks like as an image. Next year's is "Food", with the goal of creating a new taste offline.

The line that stayed with us most was that AI does not make images for you, it makes time. He is spending the time he has bought that way on learning pottery and bonsai. His advice was simple. Keep your own voice, do not just settle for whatever AI hands back, and keep asking what you can make with this tool.

In the group conversation that followed, we collected the moments when an AI result makes you lose interest. Half-convincing photorealism that drops into the uncanny valley. Consistency breaking down as repeated edits erase the face and the essence of the thing. Sloppy finishing, like a mangled font. The fixes people offered all came down to putting a human hand back in: photos you shot yourself, a personal touch, references you have verified, a consistent character guide. Someone noted that review and narrative structure still need about 30 percent human work. In the end everyone landed in the same place. The person who makes the story does not get replaced.