The idea behind Untersea began while experimenting with Midjourney.
An image would show a place that seemed worth visiting. A street, a building, a landscape. Looking at it was not quite enough. We wanted to walk into it, open a door, and see what was around the corner.
But that particular place existed only in the image. There was no game to load and no world beyond the edge of the frame.
So we decided to try building one.
From an image to a world
We began by exploring research into generating 3D environments from images. The early results were exciting enough to build around, but working with them exposed limitations that were not obvious at first.
An impressive demonstration and a dependable game-development tool are different things.
That changed our approach. Rather than expecting one model to produce everything, we designed a process that brings different tools together. It starts with an image, written direction, and other references. Generation and review happen in stages, with opportunities to inspect the result and change it.
We call this our world compiler.
Keeping creative direction in the process
We have built a process that uses AI agents and generation tools to carry out parts of this work, with people reviewing and directing the results.
The important part is being able to keep working on what comes out. An initial result is rarely the end of an idea. You might want to change the layout, replace an asset, adjust the atmosphere, or reconsider what belongs in the scene.
Our aim is to make that direction accessible without requiring the person giving it to handle every technical step themselves. Someone should be able to describe what they envision and participate in refining it, even without a background in programming or 3D design.
We are still honing the process, with the goal of producing later episodes in a fraction of the time they would take using conventional workflows. That means improving the quality and consistency of the results, not just generating more material faster.
A place worth entering
In What makes a place worth exploring?, we wrote about places that feel as though they existed before the player arrived. Places that suggest lives, histories, and conflicts beyond what is immediately visible.
That is the other half of this project.
Being able to walk through an environment does not automatically make it interesting. There need to be reasons to explore it, people worth getting to know, and decisions that mean something.
Counterproof gives us a specific game in which to develop these ideas. The technology helps us build the world, but the experience of being there is what matters.
We started by wanting to step into an image. What we want to make is a game worth spending time in.