Current direction for Machi project is:
2D side scrolling semantic tile map like Terrarium with buildable/breakable tile system.
AI first living environment where all aspect of it is living and simulated. For instance, trees grow from seed. Water falls, soil produces vegetation.
A virtual space where AI agents interact.
Uses Gen AI to render semantic tiles in high fidelity.
Cross road of nature and artificial buildings.
Focus on editable worlds that are built by the agents.
After trying out Midjourney, alternate path emerges.
Perspective pixel art adventure type of environment like Final Fantasy 7.
Voxel semantic mapping of the scene from a single image.
Voxel semantic map being used as a control net to generate walkable pixel art scene.
Focus more heavy focus on story telling and visual development.
2D is probably the first steps since it is simpler but, perhaps we can go 3D soon after. Let’s keep myself busy.
Either way, control-net using semantic 2D/3D tile is still the same. Gaining that technology will be crucial.
Whether it be 2D or 3D really depends on what kind of experience we are planning to give to the user and the AI agents.
The limitation of 3D solution is that we will mostly be providing shallow voxel map which wouldn’t be as detailed as the 2D one which will give much more detailed annotations.
I think we should keep iterating on 2D world, but have a separate thread going to research on 3D pathways.
