Kazuki Takakura: the artist who left the code to AI

Who is Kazuki Takakura?

Kazuki Takakura is a Japanese artist, born in 1987 in Yamanashi Prefecture, with a graduate degree from Tokyo Zokei University². He works with pixel art (drawings made dot by dot, like in old console games), games, virtual reality and AI, and has exhibited in Japan, Taipei, New York and Mexico².

He has been drawing, making music and inventing rules and worlds for a long time¹. Coding, no. According to him, he has no programming knowledge and, even with the game finished, still finds it strange that writing text makes things move on the screen¹.

Until then, he had only made games as part of a team¹. Every change depended on money and on other people's approval. What he wanted, he says, was to make games with the same freedom he has when he makes a drawing¹.

How do you play Onikappa?

Onikappa is a short action game seen from the side, in which the screen scrolls as the character moves¹. The player switches between two creatures from Japanese folklore: an ogre (oni) and a kappa, a water creature¹. Each has different abilities, and the game has shooting, in the style of Mega Man, a console classic¹.

The level takes place inside the body of a giant of knowledge, the "Descartes Machine", a robot inspired by Western philosophy². The player explores, faces enemies and reaches the boss. Collect the four philosophy plates scattered along the way, and the boss gets weaker¹.

The full name is "Onikappa to Chi no Kyojin - 001_Descartes Shinkami", something like "Ogre, Kappa and the Giant of Knowledge". It's the first level of a series, released on its own; the next ones are expected to come out as separate games².

How did he make the game with AI?

AI wrote all of the code; the rest was his. Takakura used two AIs that work inside the project folder, Claude Code and OpenAI Codex¹. He describes the split as a small team¹:

According to the report by the Japanese site KAI-YOU, he made the graphics, the music, the rules, the tuning and the testing, and did not use AI to generate images or sounds². The game's itch.io page says the same: the AI disclosure marks only the code³.

The idea he is proudest of came from the artist's side. Instead of marking, one by one, the walls and floors the character steps on, he paints those areas with colors on an image, and the game reads the painting¹. He calls it a "collision paint system" and says it was the project's biggest invention¹.

Why didn't he let AI decide the game?

Because, for him, what makes a game fun depends on the body. The feel of the controls, the difficulty and what is fun are decided by playing, and AI has no body to feel that¹. Left on its own, he says, AI tends to produce an expression that represents the average of the audience¹.

Our reading: it's the same division of labor seen in other cases, like Curve Clash: AI does the technical part, and the person making the game keeps deciding what it is. When you don't decide, AI fills the gap, and it fills it with the most common answer. This also shows up in what vibe coding is.

When and where did the game come out?

On May 17, 2026, on itch.io, a store for independent games, playable in the browser². The final version had been finished one day earlier, on May 16¹. In October 2026, the itch.io page listed the game as free, with a button to support the author³. It's also on the website of his studio, STUDIO TOKOYO⁴.

The game was shown for the public to play at the exhibition "NEW MEDIA / NEW TECHNOLOGY / NEW ARTIST", at Artist Cafe Fukuoka, from May 23 to June 7, 2026². In July, Takakura published a long account of how he made the game; the first part is free, and the rest is paid¹.

By the numbers

What does he advise people who are starting out?

Start small, in a kind of game you already like, according to him¹:

• One level only. A small game gets finished; a big one stalls on the art and on complicated rules.
• Simple rules, like those of an 8-bit console game. A beautiful 3D game is hard to finish with AI.
• A game that runs in the browser, which is quick to test.
• Avoid games that depend on exact timing, like rhythm and fighting games.
• Keep the feel of the controls for yourself. It's the part AI can't feel.

He also points out that a game isn't good just because it's beautiful and expensive to make¹. To get to that first small level, see how to start making a game.

The other side

Takakura is not a beginner at games: he's a professional artist with years of exhibitions, who had already made games as part of a team and already knew how to draw and compose. AI took the coding off his hands, not the rest of the work. His account doesn't say how long the game took, how many people played it or whether it made any money, and he says he's still looking for a company to distribute the next ones. The story proves you can make and release a game without coding, not that it's easy or that it makes money.

What does this story change for people who don't code?

Our reading: Takakura's story shows that people who already have a talent outside programming, like drawing, writing or composing, can now carry that talent all the way to a complete game. The barrier that used to require a programmer on the team is now handled by AI. The part that gives the game its identity stays with you.

To see other cases, of people who already coded and people who didn't, see who has made a game without being a programmer.