The main issue with feature models used for cliffs / craters / whatever is blending with terrain.
We will still need to model obscured (by model) parts using current method and we will probably need some shaders to help in creating less sharp texture transitions.
Maybe i-NoD has some opinions on this topic, since he is known from working on rendering engine and related things.
Nadszedł już czas, najwyższy czas, nienawiść zniszczyć w sobie.
The time has come, the high time, to destroy hatred in oneself.
Would it be possible to show a reflection of the sky and nearby terrain in water?
I think it would make water areas look more realistic, but not sure what effect it would have on performance for older computers (so maybe also an option to turn it off).
Also, Rman Virgil posted a pic elsewhere showing an unreleased 1.x build that used a different 3D engine - one of the pics had much deeper terrain under water and that added to the realism too. With deeper terrain under water, you could have layers of 80% transparent black at various heights to basically make the terrain be less visible as the water gets deeper for even more realism.
"Dedicated to discovering Warzone artefacts, and sharing them freely for the benefit of the community."
-- https://warzone.atlassian.net/wiki/display/GO
Just to clarify:
The picture I posted earlier is NOT an ingame pic, it is rendered in blender with ambient occlusion and soft shadows. However, the terrain is a simple heightmap grid like in warzone. It only has higher mesh density.
Anyway, I didn't say I'm against the "geometry tile" idea, many games use it successfully.
Yes, that is correct.
What some engines use is a similar concept to 2d tile-based rts:es like the westwood era c&c games. Essentially you produce a set of tiling geometry chunks that can be assembled to form a landscape.
The problem is: you need a lot of chunk types to produce a good looking landscape and you will still be more constrained when it comes to freeform landscape design. Afaik starcraft 2 does this. The graphics are awesome, but you have only 2 height levels and cliffs always have a pre-determined shape. C&C3 seems to use a combination of height-based terrain and feature tiles which is quite flexible and probably what we want.
Tile-based terrain is usually superior for urban terrain, allowing for sharp angles and interesting shapes. On the other hand, when it comes to large open landscapes, high-resolution heightmap terrain is what we want (just look at supreme commander).
I've not played other RTS of this type (except for Earth 2150) so it's hard for me to compare.
As for more detailed height maps...
What about setting arbitrary points for whole map instead of forcing it to waste memory for not needed points (as I doubt if map designers will want to manipulate each pixel explicitly)?
But I'm not sure if it wouldn't need to be translated into "regular grid" anyway, or at least sorted somehow, so VBO for terrain could be created easily.
Nadszedł już czas, najwyższy czas, nienawiść zniszczyć w sobie.
The time has come, the high time, to destroy hatred in oneself.
Well, you see, when you use a regular square grid of points, each vertex position will be implicitly known by knowing the index of that vertex, while with a non-uniform "pointcloud" you need to define both the exact xyz coordinate and the connectivity to its neighbours for triangulation. If we also take the possibility of dynamic mesh tesselation into account, you will see why a heightmap is a VERY attractive solution to any programmer.
Also, as long as you use brushes for making maps, the number of points shouldn't be a problem
It may sound now like I am a very strong advocate of heightmaps, and therefore I will acknowledge that they have at least two obvious weaknesses:
-They simply cannot produce cliff angles steeper than 90 degrees
-As cliff angles approach 90 degree angle, texture stretching becomes painfully obvious
