SubNautical - Terrain & Voxels under Unity

[AEon]

Noted that "-= Castle Does =-" has a section on the game SubNautical in development he is working on showing work with Terrain and Voxels for the underwater game using the Unity engine.

Since I have never seen much work done with Voxels, I found watching the pretty much daily episodes, Castle's daily workload (sped up), interesting to follow.

YouTube Playlist: SN

Should mentioning this be OK with Castle, I am sure he would be interested in some comments from the seasoned mappers her.

Hope the post here at LEM is OK... since I don't read the other forums .

[Castle]

Noted that "-= Castle Does =-" has a section on the game SubNautical in development he is working on showing work with Terrain and Voxels for the underwater game using the Unity engine.

Since I have never seen much work done with Voxels, I found watching the pretty much daily episodes, Castle's daily workload (sped up), interesting to follow.

YouTube Playlist: SN

Should mentioning this be OK with Castle, I am sure he would be interested in some comments from the seasoned mappers her.

Hope the post here at LEM is OK... since I don't read the other forums .

This is 1000% cool with me! Pimp it all out! attention attention!

IF you or anyone has comments or questions feel free to ask away.

[AEon]

Some thoughts... (more like bumbling along)

While watching it becomes clear that even though the voxels let you paint geometry so quickly, one does end up having to use basic shapes like spheres (and I think cubes) to start off, to then refine the look until is looks "natural" in some way.

Every time I thought, watching the videos, that something needs more work, you actually give it more work to make it look better.

It seems to me that voxel geometry cannot be copy/pasted?... i.e. you had a pretty cool tower column at the start of the "series"... then later had to redo it, but seemed to be struggling to reproduce the "happy accidents" from the start in the new version. So having a voxel shape library for things that worked elsewhere to use as a base somewhere else could help. Or for your octopus rock... to be able to save that out, to "re-import" it in another "batch" could very much help. (Unless I just did not understand the way Unity works.) Presently it seem all the "batches" are island of content, no way to move the content both in terrain or in voxels to move around, well... easily.

The other thing about height maps and voxel geometry is that you can get results fast, but it is actually difficult to come up with something that looks good. I am relatively sure after 3-4 games have used such an engine, that the players might actually have seen it all... since you can only create so many caves or cliffs or towers or bridges etc... I am probably just not creative enough, maybe.

[Castle]

Some thoughts... (more like bumbling along)

While watching it becomes clear that even though the voxels let you paint geometry so quickly, one does end up having to use basic shapes like spheres (and I think cubes) to start off, to then refine the look until is looks "natural" in some way.

Every time I thought, watching the videos, that something needs more work, you actually give it more work to make it look better.

It seems to me that voxel geometry cannot be copy/pasted?... i.e. you had a pretty cool tower column at the start of the "series"... then later had to redo it, but seemed to be struggling to reproduce the "happy accidents" from the start in the new version. So having a voxel shape library for things that worked elsewhere to use as a base somewhere else could help. Or for your octopus rock... to be able to save that out, to "re-import" it in another "batch" could very much help. (Unless I just did not understand the way Unity works.) Presently it seem all the "batches" are island of content, no way to move the content both in terrain or in voxels to move around, well... easily.

We have a copy paste system and instanced geometry available however the systems to make use of these functions are clunkier than rebuilding something from scratch. Geometry instancing is fundamentally different in practice from using static meshes. Using static meshes allows for me to take a few high detail large rock models and create a number of unique scenes by using them over and over again while the system we currently have works more like tiles where we reuse an area over and over again. I decided that given the choice I would want to use tiles sparingly as literally seeing the same area over and over again is not an ideal scenario for the players experience.

With that said, however, voxel level design is in its infancy. There are a number of systems that can be created that have yet to be done on all fronts. Thus ideas that you have to bring to the world of voxel level design are of high value. Voxel level design has brought many advancements to the table on things that I have never imagined would happen. Large communities of players working on single world in real time without the need for source control is remarkable. If you take a close look at the biggest minecraft servers or if you look at the systems that power Landmark they allow copy and pasting of voxels along with a plethra of other features. Anyway I feel I might have got a bit side tracked. Your question touches on a many topics in the field of voxel level design that I find very interesting.

The other thing about height maps and voxel geometry is that you can get results fast, but it is actually difficult to come up with something that looks good. I am relatively sure after 3-4 games have used such an engine, that the players might actually have seen it all... since you can only create so many caves or cliffs or towers or bridges etc... I am probably just not creative enough, maybe.

Yes and no, it really boils down to what you do with your toolset. I say this argument could just as easily be posed to a standard terrain and mesh method most games use today. When you factor in the voxels are still very much a growing art its a bit like kicking a feature set when its down.

After working with a system like the one in Subnautica I would strongly suggest moving to this method over the standard terrain and static mesh method. The biggest hindrance on the project hasn't been the geometry itself but rather figuring out what assets and features best compliment the project. DX11 tessellation when combined with even a mediocre material system is almost like adding voxels on to your voxels. The ability to smooth or sharpen geometry creates situations where you have what are called micro voxels. Micro voxels are a modern example of a happy accident that can be used to create various different things.

However, one of the greatest joys we have all experienced when it comes to hightmap terrain has been the ability to freely paint various different materials on to the surface of ground without worry. This luxury seems to be absent from the world of voxels as it is much more difficult to freely paint materials on to its Three dimensional surface. Currently right now we know that we have the ability to save different variables into the voxel itself. These things can range from anything to what material this block is made out of to how smooth it should look. But in the end, painting materials right now means selecting each voxel and applying a different material to it.

I personally feel that the weaknesses of voxel level design are only due to its youth and once only a handful of additional hurdles are overcome it will quickly turn it from novelty to all encompassing.

[AEon]

I think the problem with my comments yesterday was, that for a brief instant I fell into the trap of thinking, "voxels are here ... now anything can be done with them"... but as you say, this is only the beginning. To me it might be the shock of transition between playing 550+ hours Skyrim to then end up with "blotchy rocks" (pretty ones though ). A very unfair comparison, since Skyrim is pretty much top of the game regarding meshes and artists used to them, and the voxels are pretty new, the tools are very much still in development... and "we all" as yet are coming to terms with that tech (well I am just observing).

It took me a while to rethink what you were doing and why it was being done. Technically terrain engines let you create a heightmap from a 2D image (did some of that in Far Cry for a beta map)... and back then you always had the issues of steep shifts in terrain, where you needed to use special vertical shaders to make the "cliffs" look good... usually leading to many visual bugs. From watching your work on SubNautica the sharp edges in terrain are already "fixed" a huge amount better. And then you have voxels to create overhangs and all the other hand crafted really nice natural-looking geometry. Letting you "paint geometry" over the terrain edges. So presently I see voxels as a really valuable and relatively easy to learn means to create really special looking landscapes... without having to resort to huge meshes. For those not adept at modelling (and I never was able to get anything properly created with gmax, that I could not have done so much more quickly in GTKRadiant) it really opens a door in regards to accessible creative map building. At least when regarding "advanced terrain" (as I feel inspired to call it).

I was wondering if it even possible to create "perfect" cubes using voxels? Or are the surfaces always bumpy sphere approximations of a flat surface and the cube corners in fact actually rounded off? Pardon the ignorance.

Well the way SubNautica uses height-mapped terrain to get the large scale landscapes created, to then refine those and give them really nice detail and an unusual look... finally to add meshed to add polygon-based detail is really effective. And the game, with lots of terrain-related exploration (swimming that is), certainly seems to lend itself to this tech approach.

It will be interesting to see how voxels will be able to encroach on meshes in the future, i.e. adding very detailed geometry that is more "angular" (for a better word).

Reminds me a bit of your UT4 discussion of the Epic level designers working using brushes instead of meshes to add detail... something I found very strange. Since Epic was among the first to pretty much ditch brushes and only use them for "vis blocking" and plaster everything else with detailed meshes. It made me think maybe they hired more than a few folks with Hammer/Radiant background .

I love your max work on the temp meshes in UT4, letting you quickly "bevel" (IIRC that was the term) geometry to make it look so much more detailed so quickly. Certainly something that is a nightmare in Radiant.

Anyway, inspiring and very interesting to watch your videos.
(Looking forward to how your UT4 mapping goes.)

Regarding a few ideas, probably not really creative but maybe it inspires in some way:

• Playing Solstheim (the DLC island) in Skyrim made me remember that hexagonal rock columns are actually seen in nature. A forest (with gaps between them) of them in different shapes and heights might be a bitch to do with voxels but could give a certain part of the underwater world a slightly techy look.
• About making bridges look different... in your special area (many videos back) you had very steep cliffs connecting them with "worm-like tubes"... these were arcs or bridges of some sort. I noted one bride that was "Y"-shaped (i.e. forked).... if you use bridges that fork more often e.g. 3 times this might create an unusual look. This might end up in voxel domes with the "tentacles" supporting the roof of the dome. Come to think of it a double-helix or even triple-helix shape (large radius though) might suggest something out of the ordinary, but will suggest it could be natural in some "alien" way.
• About the idea you developed of the "Adope Village" on a rock, those squarish-looking "houses". It is an interesting concept to develop further IMO. The idea in a nutshell: "add human/alien-created structures that are just slightly too regular to be natural, but sill keep the the player guessing"... using real-world architecture, like that of Brasilia but "naturalizing" it could turn out quite stunning... and the voxels would give it a new spin. To mention another cliche: like the very overgrown long lost ruins of Atlantis or some such.
• Since roundish is "natural" and "voxel-inherent" anything that is angular or at least looks so, would stand out in the SubNautica world. A simple example... a squarish entrance to one of your caves could suggest artificial... but the degree of natural could then be adapted to the needs of the game.

Hope a hexagonal forest makes it into the game ... if its not already been done by another artist in the team.

[Castle]

I think the problem with my comments yesterday was, that for a brief instant I fell into the trap of thinking, "voxels are here ... now anything can be done with them"... but as you say, this is only the beginning. To me it might be the shock of transition between playing 550+ hours Skyrim to then end up with "blotchy rocks" (pretty ones though ). A very unfair comparison, since Skyrim is pretty much top of the game regarding meshes and artists used to them, and the voxels are pretty new, the tools are very much still in development... and "we all" as yet are coming to terms with that tech (well I am just observing).

Yeah we haven't seen the full potential of voxels yet for sure.

It took me a while to rethink what you were doing and why it was being done. Technically terrain engines let you create a heightmap from a 2D image (did some of that in Far Cry for a beta map)... and back then you always had the issues of steep shifts in terrain, where you needed to use special vertical shaders to make the "cliffs" look good... usually leading to many visual bugs. From watching your work on SubNautica the sharp edges in terrain are already "fixed" a huge amount better. And then you have voxels to create overhangs and all the other hand crafted really nice natural-looking geometry. Letting you "paint geometry" over the terrain edges. So presently I see voxels as a really valuable and relatively easy to learn means to create really special looking landscapes... without having to resort to huge meshes. For those not adept at modelling (and I never was able to get anything properly created with gmax, that I could not have done so much more quickly in GTKRadiant) it really opens a door in regards to accessible creative map building. At least when regarding "advanced terrain" (as I feel inspired to call it).

It also allows for environments that can be more easily modified during game play which is a pretty interesting aspect to keep in mind too.

I was wondering if it even possible to create "perfect" cubes using voxels? Or are the surfaces always bumpy sphere approximations of a flat surface and the cube corners in fact actually rounded off? Pardon the ignorance.

One thing that has to be kept in mind are limitations with triplanar materials. Triplanar materials are fantastic at everything but hard edges I have found. So in a setting where you use them you need to have at least some kind of beveling going on. Perfect cubes are otherwise exactly what voxels are. 3 dimensional arrays to be more specific.

Well the way SubNautica uses height-mapped terrain to get the large scale landscapes created, to then refine those and give them really nice detail and an unusual look... finally to add meshed to add polygon-based detail is really effective. And the game, with lots of terrain-related exploration (swimming that is), certainly seems to lend itself to this tech approach.

It will be interesting to see how voxels will be able to encroach on meshes in the future, i.e. adding very detailed geometry that is more "angular" (for a better word).

I feel that DX11 tessellation combined with an advanced material system that allows both blending of materials, decals and the ability to paint height will be when voxel systems will have potential to surpass current methods.

Reminds me a bit of your UT4 discussion of the Epic level designers working using brushes instead of meshes to add detail... something I found very strange. Since Epic was among the first to pretty much ditch brushes and only use them for "vis blocking" and plaster everything else with detailed meshes. It made me think maybe they hired more than a few folks with Hammer/Radiant background .

This had me scratching my head too. But after a short while I think that UT3 has been seen as a hard lesson of what not to do. So they are now building level shells with all brushes more so than in UT3.

I love your max work on the temp meshes in UT4, letting you quickly "bevel" (IIRC that was the term) geometry to make it look so much more detailed so quickly. Certainly something that is a nightmare in Radiant.

Anyway, inspiring and very interesting to watch your videos.
(Looking forward to how your UT4 mapping goes.)

So far its going great! I have a new version ready to release but cant due to an ambient lighting bug. how annoying!

Regarding a few ideas, probably not really creative but maybe it inspires in some way:

• Playing Solstheim (the DLC island) in Skyrim made me remember that hexagonal rock columns are actually seen in nature. A forest (with gaps between them) of them in different shapes and heights might be a bitch to do with voxels but could give a certain part of the underwater world a slightly techy look.
• About making bridges look different... in your special area (many videos back) you had very steep cliffs connecting them with "worm-like tubes"... these were arcs or bridges of some sort. I noted one bride that was "Y"-shaped (i.e. forked).... if you use bridges that fork more often e.g. 3 times this might create an unusual look. This might end up in voxel domes with the "tentacles" supporting the roof of the dome. Come to think of it a double-helix or even triple-helix shape (large radius though) might suggest something out of the ordinary, but will suggest it could be natural in some "alien" way.
• About the idea you developed of the "Adope Village" on a rock, those squarish-looking "houses". It is an interesting concept to develop further IMO. The idea in a nutshell: "add human/alien-created structures that are just slightly too regular to be natural, but sill keep the the player guessing"... using real-world architecture, like that of Brasilia but "naturalizing" it could turn out quite stunning... and the voxels would give it a new spin. To mention another cliche: like the very overgrown long lost ruins of Atlantis or some such.
• Since roundish is "natural" and "voxel-inherent" anything that is angular or at least looks so, would stand out in the SubNautica world. A simple example... a squarish entrance to one of your caves could suggest artificial... but the degree of natural could then be adapted to the needs of the game.

Hope a hexagonal forest makes it into the game ... if its not already been done by another artist in the team.

There are all kinds of things to explore on all fronts so far. Ancient village or squared off caves that look less natural. I am all for adding things to the world that show a story but dont tell you too much all at once. This aspect is really enjoyable I find!