I keep returning to Infinite Journeys, the montage of artists building scenes around a shared visual premise. It makes me imagine a future tool that could turn each small glimpse into a place you can enter.
That is an idea about where generation might go, not a description of a working system. A beautiful image and a navigable world have very different obligations.
An image only has to make sense from its chosen viewpoint. A staircase can disappear behind an object before its geometry becomes awkward. A window can suggest a room that has never been designed. Once the viewer moves, those convenient ambiguities become visible requirements.
The first requirement is consistency. Walking around a table should reveal another side of the same table. Returning to a doorway should not produce a different room merely because the view changed. The world needs a memory of relationships, whether those relationships are stored explicitly or maintained through another representation.
The second is behavior. A floor needs to support the player. A door may need to open. A distant building might be scenery, while a nearby handle must be reachable. Appearance alone does not specify those distinctions, and a general prompt rarely describes every interaction.
There is also an artistic problem. An image generator can provide a striking arrangement, but a space needs to manage a sequence of discoveries. What becomes visible first? Where can a visitor pause? Which route creates anticipation? A place has rhythm even when nobody has supplied a timeline.
I would begin much smaller than a generated universe. One room with a few consistent objects, one reliable way to move through it, and a clear boundary around what can be changed would already be interesting. The test would be whether the space remains understandable after the initial surprise.
The prospect still excites me. A sketch or short video could become the beginning of a spatial conversation: make this corridor narrower, let the light enter from the side, preserve the chair while changing the room around it. That requires persistent control, not simply repeated requests for a fresh image.
For now, I enjoy the distance between the montage and that imagined tool. It makes the engineering questions more precise. I don't just want more views of something beautiful. I want the views to belong to a place that can survive being explored.