Environment art workflow questions: texel density importance, when to bake highpoly

19ulrich90

Hi, I'm making this tavern environment, first time doing something this big, learning alot of new stuff, having a lot of questions.

So I set texel density of my environment to be 4k per 3x3m. Vertical pillar (3m) is UVed, textured to that size.



My question is, how would you UV a pillar that is going to be placed horizontaly beneath the window, as it is longer than 3m (about 5m). I understand that I could just tile the texture on a low poly model of a pillar, but what if i want to model high poly and than bake normal map, like I did with the vertical pillar. Am I just nitpicking and there woudn't be noticable difference between 4k texture on 3m and 5m long pillar?

oglu

i would just stretch the pillar... think it should look not to bad...
if its not looking good... insert some metal parts... to hide the seams...

19ulrich90

Yeah, I tought about stretching it or just making a 4k texture with wood pattern repeating a bit more than on a 3m pillar. Metal parts wouldn't fit the design though. Just wanted to ask how others deal with this as it is impossible to have all the assets with same exact texel density.

rollin

Imo it's nice to build all huge main textures on a tileable basis. So also your normalmaps and whatnot should work that way so you can just tile it along any length. Also texel density is view distance dependent. If you have assets  where you're not able to get closer with the camera (maybe at the roof of the tavern) you can use a lower texel density. 
If you have a single big objects with an unique texture you can spread it over so many texture tiles until you reach the texel density you need.

Deadly Nightshade

You shouldn't use uniform texel density across the entire level!!
Texture memory is expensive, which is why you should have a higher TD on things that are highly visible to the player, and lower TD on things that are not so visible to the player. One good example of how this is practised is to play Rage and study the quality of the environment/level textures. 

musashidan

Deadly Nightshade said:

You shouldn't use uniform texel density across the entire level!!
Texture memory is expensive, which is why you should have a higher TD on things that are highly visible to the player, and lower TD on things that are not so visible to the player. One good example of how this is practised is to play Rage and study the quality of the environment/level textures. 

Texel density can stay the same but texel resolution will be varied. So for instance a 2m tall door and a telephone will have the same texel density but different texel resolution. Once global texel scale for the level is set then you can use tools to normalise the TD between assets. I'm sure you obviously know this(because of who you are and what you do) but I thought you might have just confused the 2?

19ulrich90

@rollin ok, I'll show my process so we get this clear:D. I may be doing this the nonconventional way that's why I'm asking. Lets say i'm making a "pillar kit" with one 3m and one 5m long pillar. I make high poly versions of both in Zbrush first, decimate both and then UV low poly versions of both in Maya. Here's how I layed out the UVs, texel density is obviously not the same, texture resolution is. 




I then bring the kit in Substance Painter and bake the normal maps from high poly kit (high poly is just a bit od trim dynamic brush on the edges for the masking generators), add textures (I masked out sperate pillars so the 5m one can have more fibre repetition etc.). Please don't mind the sculpting and texturing skills, I'm still learning and its just for demonstration purpose





I brought it in Unreal and it's looking pretty good I guess considering texel density not being the same?



@Deadly Nightshade That makes sense, yeah, about assets not being the same TD from optimization's point of view. But in my case those pillars will be seen from same distance.  Rage is still waiting on my Steam wishlist

@musashidan I don't think he confused the 2. If the player will never be close to the model enough it can have lower TD since it wont be noticable from distance. 

So I'm looking at this example from game Kingdom come: Deliverance. Are all those planks and pillars made with same TD in mind?

Is the workflow in games like this the same to mine, where high poly is firstly made and than baked to low poly or are those bridge pillars just low poly models with bevel on the edges, few edge loops for slight model deformation and tiling textures? 

Quack!

In a modern work flow we are depending less and less on high poly unique baked assets and more on high tri count low polys with material blending and custom normals.  In this case I wouldn't bake a unique normal map and instead use a tiling texture and some edge bevels with custom normals.  Most assets like this, even in the past, would not be baked down from a high poly, major time waster.

As far as texel density goes, you should be very anal about getting them as close as possible.  They will never match 100% and that is fine, but you should always keep it in mind, two objects with varying levels of TD will be noticed by players very fast.  With that said, we are approaching texture resolutions that are so relatively high, that two objects with different TD's won't necessarily look very different.

19ulrich90

@Quack! Thanks, for the explanation. I tought custom normals are used more for hardsurface modeling though? I'm following David Lesperance's tutorial "Intro to Game Art" and he's pretty much modeling everything in highpoly first in 3dsmax and 3Dcoat, remeshing in Zbrush and baking normals on low poly when it comes to more organic models. It may be because he's creating more complex modular meshes composed of multiple parts.

Quack!

19ulrich90 said:

@Quack! Thanks, for the explanation. I tought custom normals are used more for hardsurface modeling though? I'm following David Lesperance's tutorial "Intro to Game Art" and he's pretty much modeling everything in highpoly first in 3dsmax and 3Dcoat, remeshing in Zbrush and baking normals on low poly when it comes to more organic models. It may be because he's creating more complex modular meshes composed of multiple parts.

That workflow works for a certain types of assets and should be learned.  However, it is VERY time consuming. So in conjunction with learning the high-low-bake, you should also learn the tiled_material-low-custom_normals workflow.
I use custom normals on every object I make nowadays and rarely bake high polies, I just throw a shit ton of tris (within reason) at everything. For example, a 30-50k (with exceptions) tri first person weapon doesn't need a high poly anymore. And soon not much will.

If it was up to me we all would stop baking high poly models right now so that companies can make games with fewer people and not have to layoff constantly. But that is for next generation, right now though we should still bake high polys for objects that need them, which is becoming fewer and fewer objects with our increasing triangle and material limits.

Goobatastic

I have to agree  with Quack isnt worth baking down a wooden beam might aswell use a tiling texture with a detail normal map which can fade in from a certain distance.

musashidan

19ulrich90 said:

@musashidan I don't think he confused the 2. If the player will never be close to the model enough it can have lower TD since it wont be noticable from distance. 

As I said above, usually a texel to real-world measurement ratio is decided upon for the game or level to aid in TD consistency across assets. So to take my example of the 2m tall door and the telephone.(if both are uniquely UVd) They will both have equal(or very close to equal) TD but the door might use a 1Kx512 map and the phone might use a 256x256 map. Both have equal TD but very different resolution.

If you normalise your TD between already UVd assets you can get a good idea of what resolution they need to be by seeing how the UVs fit onto 0-1.

If the TD is all over the place between assets then the level can quickly become a mess if many artists are authoring assets independently with no set TD ratio.

musashidan

19ulrich90 said:

@Quack! Thanks, for the explanation. I tought custom normals are used more for hardsurface modeling though? I'm following David Lesperance's tutorial "Intro to Game Art" and he's pretty much modeling everything in highpoly first in 3dsmax and 3Dcoat, remeshing in Zbrush and baking normals on low poly when it comes to more organic models. It may be because he's creating more complex modular meshes composed of multiple parts.

Is this an old tutorial you're watching because David Les has repeatedly stated that he avoids baking altogether and uses a chamfered workflow(as @quack described) almost exclusively.

19ulrich90

musashidan said:

19ulrich90 said:

@musashidan I don't think he confused the 2. If the player will never be close to the model enough it can have lower TD since it wont be noticable from distance. 

As I said above, usually a texel to real-world measurement ratio is decided upon for the game or level to aid in TD consistency across assets. So to take my example of the 2m tall door and the telephone.(if both are uniquely UVd) They will both have equal(or very close to equal) TD but the door might use a 1Kx512 map and the phone might use a 256x256 map. Both have equal TD but very different resolution.

If you normalise your TD between already UVd assets you can get a good idea of what resolution they need to be by seeing how the UVs fit onto 0-1.

If the TD is all over the place between assets then the level can quickly become a mess if many artists are authoring assets independently with no set TD ratio.

I get what you are saying, for the object that are the same distance to the player the TD must ofcourse be as exact as possible, if its a smaller object it will use lower texture resolution. Lets say TD is 1k per 1x1m. 1x1m wall will have 1k texture, 2x2m wall will have 2k texture, different resolutions, TD is the same. But if i understood @Deadly Nightshade correctly, he was talking about objects that are further from the player. 1x1m wall that is far away from the player and will never be seen up close, doesn't need 1k TD per 1x1m.

19ulrich90

musashidan said:

19ulrich90 said:

@Quack! Thanks, for the explanation. I tought custom normals are used more for hardsurface modeling though? I'm following David Lesperance's tutorial "Intro to Game Art" and he's pretty much modeling everything in highpoly first in 3dsmax and 3Dcoat, remeshing in Zbrush and baking normals on low poly when it comes to more organic models. It may be because he's creating more complex modular meshes composed of multiple parts.

Is this an old tutorial you're watching because David Les has repeatedly stated that he avoids baking altogether and uses a chamfered workflow(as @quack described) almost exclusively.

No, its his new tutorial from gumroad. Here's a screeshot.

He was probably talking about hardsurface modeling, here he's creating more destructive environment that's why he's baking normal maps. 

As I said I'm a beginner, so I'm a bit confused to when only use modeling and when baking.
EDIT: I changed the topic's title since other questions have come up.

musashidan

A lot of this damage can now be painted directly to the normal channel in programs like Sub Painter. So you could set up a brush to mimic ZB's trim dynamic for instance, or a damaged wood brush. And damage can also be directly modeled to these mid-poly assets(or automated using an edge-damage script)This and chamfering/vertex normal editing can get pretty close to a baked ZB sculpt. 

As with a lot of techniques with this stuff it really comes down to the asset in question. So it is indeed a great idea to master all the techniques so you can apply them at will per asset. 

So some relatively unimportant planks won't get a high-poly bake but a 'hero' pillar with intricate carvings would. And again, even the pillar could be normal-painted using custom stamps/brushes.