This is a really interesting topic and improving the workflow for generating large worlds with terrains, roads and buildings was the main source of inspiration for CityGen3D, which is a scene generation tool I’m working on.
So I’ll be interested to hear people’s thoughts on this too, as its something I’m essentially trying to help with.
The approach I’m taking in CityGen is to make everything data driven. So you change your scene by changing rules and conditions and have Unity generate your actual world for you.
In other words, the data that defines your scene and the scene itself are two distinct parts and the iterative process occurs from changing the input (via a UI) and not with the scene itself.
So as a designer, you can play around with different data sets and inputs and not worry about the actual scene geometry which can be rebuilt at any time autoatically from your data.
For example, if you had a road a mile long in your scene and suddenly decided it should be a six lane highway, this could be a real pain if you had to re-adjust your terrain and road geometry manually.
With CityGen, the idea is that you would just click the road on a map, change lanes to “6” and have it rebuild the scene for you.
Same with different biomes. What if the beach was a swamp, or the meadow was actually a pond? Instead of manually making those adjustments, changing textures and placing vegetation, you define what each biome should look like via a UI, and simply select it for particluar areas of the map via a drop-down.
Initially I’m using real world data as the input, but the idea is that you would be able to design your own map from scratch within Unity eventually. So you could draw your own road networks in 2D to create your own fictional city.
Another advantage of this data driven approach is that when you get round to implementing AI (such as road traffic), you already have a data set to use that doesn’t have to be defined over the top of your geometry.
CityGen is still work-in-progress ( [RELEASED] CityGen3D | Procedural city generation from map data ), so I’m very interested to hear about people’s current workflow headaches when building big cities & landscapes, and how they would like to see them get fixed.