It all began in the car. My daughter, who is ten, and I were imagining a game: what it would be like, what story it would tell and what its characters would do. One of those conversations where one idea leads to another and the journey fills with possibilities.

At home, we opened Claude Code to turn that conversation into a video game. We reached a playable version. Our direct involvement was about two hours; the AI tools spent approximately five hours executing the work. These are estimates from this family experience, with these tools and this scope, not a promise of how long any game takes to create.

An illustrated autumn forest with a path, mushrooms and a fallen log.
The forest in the prologue of Dragens hemmelighet. The settings look like an illustrated storybook.

It is called Dragens hemmelighet, The Dragon’s Secret. A girl and her father travel through forests, rivers, ruins and caves. The dialogue is in Norwegian, and the girl discovers a special connection with water and animals.

We wanted cooperation to mean something: the father can lift her up or hold an obstacle while she passes; she creates columns of water so he can cross a river, and he then prepares the way for her. Each needs the other.

The father, wearing a mustard shirt, a backpack and a red scarf.
The father, one of the two main characters.
The daughter, wearing a purple cape, with a spiral of light above her hand.
The daughter. Cooperation between father and daughter is part of the game’s actions.

There is also a dragon, Slem, and a family secret. The story is about fear, caring for others and understanding what lies behind anger. I would rather not reveal the ending.

Slem, a blue-green dragon with large wings, pale horns and a spiral on his shoulder.
Slem. His appearance had to fit the same world as the characters and settings.

Claude Code was building the screens, controls and movement. We explained what we wanted, tried it out and asked for changes. But the first images did not convince us. We asked ChatGPT for drawings and liked the results much more: expressive characters and a storybook look.

The combination worked, although we were acting as messengers. Each request had to be passed on, the image collected and delivered to the other side. If a pose was missing or something did not fit, we started again. Every improvement could mean another trip between conversations.

To automate this, we moved to Codex with the OpenAI image generation tool. Claude Code and Codex shared a folder on our Mac, organised as a mailbox for requests and deliveries. The process was simple:

  1. Claude Code left a request with references and requirements: for example, separate character poses on a transparent background.
  2. Codex picked up the request, generated the image and performed technical checks on dimensions, transparency and pose separation.
  3. It saved the complete delivery and, when needed, a map of the positions so it could be animated.
  4. Claude Code incorporated the files, ran its checks and confirmed integration. A correction led to another request.

Each request had an identifier to avoid duplicates. Incomplete files did not count as deliveries. Before relying on the process, we checked that a message was received and then that an actual image was delivered.

Sharing files on the computer does not mean the models ran locally. We had organised the exchange of work between tools. That let us give more attention to the story and to trying out what we wanted to improve.

Eight consecutive drawings of the daughter walking, with the same clothes and hairstyle.
An image for a game can contain many poses. This sheet provides the daughter’s walk cycle.

The technical checks did not replace our visual review. A character had to remain recognisable while moving; even the scarf had to disappear from the father’s neck and appear on the daughter’s wrist when he handed it to her. A beautiful image might not work for that scene.

A 13-second sample assembled for this article using game assets: walking together, Slem’s flight and the trout’s jump. This is an animation demonstration, silent; it is not a recording of gameplay.

We added sounds and small details: moving squirrels, falling leaves and a jumping trout. Once you can play, you start noticing everything that could bring the world more life.

Eight poses of a red squirrel: sitting, eating, looking, jumping, landing and running.
One of the game’s squirrels. Environmental details also needed their own drawings and movements.

AI carried out much of the work between our reviews. We contributed the decisions, corrections and judgement about whether it resembled the game we had imagined. The result was an adventure we could play together.

A blue river between rocky banks, surrounded by autumn trees and mountains at sunset.
The river, one of the settings where the two main characters’ abilities complement each other.

I wanted to share this in Juss og KI because learning to use AI can also mean organising how tools hand work to one another. A shared folder and clear rules helped us stop carrying every request ourselves.

But what matters most to me is still having done it with my daughter. Taking part in creating a world and then exploring it together. Technology shortened the path from imagining it to playing it.

We stepped out of the car with a story in our heads. A few hours later, we could already step into it.

Text by Roberto Ortega, prepared with AI editorial assistance. The Norwegian and English editions are editorial adaptations of the Spanish original.