Alright, so I've been studying games like Xenoblade Chronicles X (and recently, Zelda: BotW) in terms of attempting to understand how they might have both designed and actually constructed their open-world land formations in their 3D modeler applications and I have come to the conclusion that I really have absolutely NO idea how they're able to visualize these huge 3D models as they're constructing them, much less how they somehow are able to keep them looking seamless when attaching them to one another.
For those that are 'in the know' in AAA studios, my question is about workflow -- it's not engine-specific (just keep in mind the limits of Unity in terms of not being able to stream worlds in-editor from disk to memory). I'm stumbling hard on what method most AAA studios use to keep pivot-point placement in-editor both accurate/seamless (i.e. Unity) while also being able to freely model/sculpt each mass of land in Zbrush (or whatever) and still visualize that seamlessness? Is there a guy whose job it is to grab these masses of land and try to line them up AFTER they're sculpted, or is there some better or more-common way this scenario is approached in most studios?
PS:
I know some of this workflow stuff could be under NDA or even just contractual silence, but I'd be satisfied with at least the "what" is used to do the design, sculpting, and seamlessness placement, even if the "how" it's done exactly cannot be disclosed.
Replies
Unity has a terrain system included, I recommend taking a good look at how it works.
Once you understand how it works, you can tweak it with tools from the Asset Store to extend the functionality where it has weaknesses. I put a bunch of info in my sketchbook thread about how I've been using it.
Sure, I probably should have mentioned it, but I've figured out that much. Terrain heightmaps are great -- except when they aren't.
There are things heightmaps simply can't do (such as almost everything in the below two images), and in games like LoZ: BotW and Xenoblade Chronicles X, their "terrains" appear to actually be sculpted meshes, and I can understand why. Being able to sculpt these types of formations seems necessary in these cases, especially when considering the skylines. In Xenoblade Chronicles X, for example, you can clearly see a mesh around the central hub area that was sculpted as a seemingly continuous mesh (the blue splash effect around the ship) or other such land formations:
In terms of designing huge cliffs and overhangs across wide-open mountain ranges across an entire skyline, which is quite common in games like Xenoblade Chronicles X and LoZ BotW, blending such things into nice worlds with interesting shapes, all the while keeping an eye on the skyline and line-of-sight the player has in much of these locations is paramount to making the world feel complete as well as breathtaking. I feel there must be some way that studios are able to visualize these skylines while both designing and sculpting without losing the seamlessness.
The bottom image in the lower-left toward the spider crab thing, take a look at the ridge leading up from the crab's leg leading up to the guy's right arm -- that whole ridge must have been sculpted.
Sure, there's a terrain engine there for minor bumps/hills, but to be able to sculpt something of that magnitude so seamlessly -- there's GOT to be a trick to it since the artists would still have to see what it looks like against the rest of the skyline, which also was sculpted.
If you don't believe me, get on the ground in the game and walk around trying to find cracks in the world geometry when those hills turn to mountains. There isn't any. Additionally, the textures seem to be to-scale also, meaning that nobody could have UV-Unwrapped those in any traditional way while still maintaining their seamlessness (at least as far as I can tell -- though I'm no expert). Bottom line is, they *had* to have some way to easily visualize the surrounding structures as well as the skyline while sculpting in order to know where to put the negative space -- not to mention being able to keep track of the positive space from various angles.
2 racing games, approx. 8km x 8km map size
- the terrain is sculpted as one big piece then chopped up into sections for artists to work on the smaller pieces
- I brought the smaller pieces that were adjacent to each other into a file. I would then use a script that automatically blended out lighting seams of adjacent pieces. I would also automatically fix any small geometry errors that crept into the seams.
- I re-exported the smaller pieces with fixed seams back into the artist's files
- I used vertex alpha to hide seams between the main terrain and extra pieces that sat on the terrain
- I also used vertex alpha to blend some floating shadow geometry into vertex lighting
1 mobile racing game
- terrain pieces are random and snap together in any order, massive number of combinations
- can't automatically blend anything
- I wrote a tool to automatically blend out any seams in real-time
As for the above images, it's not too difficult to write tools that automatically align each model to sit near-perfectly on the surface. With a bit of vertex alpha blending it can blend seamlessly.
From another angle, you can clearly see the seams in Xenoblade so it's just a model intersecting the terrain.
Sorry for the long delay to reply to your post, but I wanted to say thank you for demystifying that for me a little more. I apparently got fooled in the areas I was playing because of some nice shader tricks. Looking at it more, you absolutely have a point -- they are just intersections in the ground.
Two questions I had regarding the racing game technique:
For the last game I worked on, I wrote a shader that covered up the seams at the cost of some blurring. Game view is on the left and I representative of what player would see.
https://youtu.be/f-a3dJTZsqM
2. Depends. One game had the terrain manually sliced up into sizes that seemed reasonable for each artist.
On the mobile game we used null objects on the edges of each terrain chunk to snap it into place with the previous chunk. Each chunk was modelled independently of the other pieces. Lots of ways to do it.
I can't think of any reason vertical terrain would be different than horizontal terrain.