Mapping biomes via probability distribution. Now far, far more better than filling Biome Map Ranges by hand.
I generate the world via following algorithm:
Because 3D functions are slow as hell (traversing the ANL graph for each voxel), the system is now completely 2D!
We have ANL Graph node for the world height map in range -1...1 (with some exceptions, I do not clamp heights)
We also have ANL Graph node for the biomes map. This function can be in any range. I use range -1 to 1 for voronoi cells, but other ranges are also used, as shown in the table below. The system is now overflexible, because any kinds of biomes can be created and distributed by probability.
Now it's time to start implementing the Material Layers subsystem. The layers you seen in first screen shots were just based on the original 3D slow-as-hell function, but now I have different approach: Defining what earth layers a biome should have, and setting per-layer 2D ANL Graph Nodes. So now, for entire world column of voxels, we sample fractals only [heightmap+biomemap + sum(layers)] times. Far, FAR faster than before.
However, heightmap based terrain have all of its limitations, no overhangs, cliffs and caves.... but the solution for them is: I will simulate the process.
The game first generates a region: 32x32 chunks (512x512 meters), then generates low-resolution biome map (128x128) and then for each cell of this map it decides what will be spawned and what not and will simulation algorithms run.
Simulation algorithms will drill caves, build dungeons and cities, even mountains and hills. A lot of imagination here. I am researching about dungeon generation, castle architecture, city - appartment design and what not.
Crafterria is a project with way too much effort put into it, but what will happen - only the time will tell.