Learning current-gen workflow

19ulrich90

Hi Polycount community!

I'm a multimedia student, interested mostly in 3d environment art. Currently I have pretty basic understanding of
the process. This is render from last year's school project. Modeling, rendering done in Cinema 4D, textures in Photoshop.



This year I'm intending to acquire knowledge of current-gen workflow. For modeling I'll be using Maya and Zbrush, texturing Substance Designer, Painter/Quixel and bringing the models to Unreal Engine 4/Marmoset.

So I'm starting with prop creation, and will be posting questions along my process. I'm quite new at this forum, so excuse me if the questions have been answerd multiple times, but the number of topics here is quite overwhelming, and I'm trying to catch up.

REFERENCES

MODELING

Currently I'm at low poly stage with modeling the shamshir. This is a render with smoth preview on. I know i have to improve my hardsurface modeling quite a lot.







So my current questions are:

1. Is it necessary to split model in parts like I did or is it preferable to combine meshes if possible. I'm asking this in regard to high poly, normal maps and AO baking later, which I haven't done before and don't quite understand yet.

2. Should I remove polygons that are not seen, like the inner part of cross guard? Or should I leave just enough geometry to hold the edge?

Any other comments would be appreciated.

musashidan

1- if elements of the model are going to interfere with the bake(for instance the hilt and the blade) it's best to split them up.

2- you've answered your own question here:if the geo won't be seen then it's wasted(on the low-poly. High poly doesn't matter as its sole purpose is to provide a nice bake)

if the edges are needed to retain the required forms then keep them.

Also, I would advise perusing the stikies here, there is a wealth of useful info.

19ulrich90

MUSASHIDAN 

Thanks! Yes, I'll go through all the stickies here in technical talk and then continue with the sword.

19ulrich90

So I'm slowly making progress. I upgraded high poly model a little, added some larger details on wooden grip in Zbrush and but am having some problems with UV mapping/normal maps baking.
Front and back high poly renders:

So I'm not sure about couple of things.  I followed some tutorial where the guy mirrored geometry across axis, left UVs of both meshes overlaping, baked normal maps in XNormals and normals came out alright on both sides. But I also read that UVs shouldnt be overlaping during baking.

So I did the same the same as the guy from tutorial for marked geometry and UVs, but the result doesnt look ok to me. Here is UV layout and the back side of guard with mirrored geometry.


Result:



Even if I reverse normals it doesn't look right to me.

19ulrich90

Ok, so I made a video showing my process. Can anybody tell what I'm doing wrong?

https://drive.google.com/file/d/0B9kn6sD95ZWycVNsaG5ZazQxdTg/view?usp=sharing

Stranger

Well, try to use the baker built into Maya. 


Also, if you dont mind just link a zip of the models. il bake them on my side to see if i have issues. 

Obscura

Stranger said:

Well, try to use the baker built into Maya. 


.

This is a bad suggestion. A lot of engines use Mikk tangent space, which you can be synced to with an Xnormal bake, but not with a Maya bake

Stranger

Obscura said:

Stranger said:

Well, try to use the baker built into Maya. 


.

This is a bad suggestion. A lot of engines use Mikk tangent space, which you can be synced to with an Xnormal bake, but not with a Maya bake

http://www.handplane3d.com/ This should work and its freeware now.

JedTheKrampus

Handplane doesn't support Mikktspace.

Stranger

JedTheKrampus said:

Handplane doesn't support Mikktspace.

Really? Ah gotcha. i thought it would sense it would sync with unreal. thanks for telling me though. 

19ulrich90

Thanks for replies! I tried with Maya baker, but it comes out the same as with xNormals. 

I also tried moving mirrored UV shell out of the 0-1 UV square before baking as suggested here http://wiki.polycount.com/wiki/Normal_Map_Modeling#Mirroring , but the result is the same.

So here are the low and high poly OBJs, i would appreciate if you tried on your side @STRANGER

Stranger

Man, i have no whats going on to your side. my bakes where "okay".




Hopefully someone else can figure out whats going on.

Obscura

I wouldn't say fine. Its gamma corrected (at least it looks like), and also it suffers from really intensive gradients, which will cause issues in the most of the engines.

Stranger

Obscura said:

I wouldn't say fine. Its gamma corrected (at least it looks like), and also it suffers from really intensive gradients, which will cause issues in the most of the engines.

Oh yeah. ( I dont know what i did but every bake in Modo looks like its gamma is raised.)  theres also some hard edge issues. but the mirror does not looks like the normal's are flipped.

19ulrich90

Thx for your effort! Yes, the hard edge issues don't bother me at the moment, just this flipped normals. Here is only bake from xNormals. I don't think the problem is in bake, as it looks similar to yours. How does it look if you apply it to the model?

Obscura

I checked your mesh, and one big issue is the overlapping of the two "cross shaped" sides. You have different smoothing on the two sides -  one side is smoothed and the another use hard edges, so basically they need different shading/normalmap. 2 options here... don't overlap them, or put that hard edge to the another side too. Also you basically cannot bake with overlappings. You need to offset one to UV space 1-2 from 0-1 (or to any outer space). You are also wasting a lot of geometry, but thats an another thing.

Stranger

Other then the hard edge, and the gradients 

Obscura

Here we go... 

Obscura

Result with multiple smoothing groups

Obscura

Stranger said:

Other then the hard edge, and the gradients 

Lol, it just gets worse. Now you got some triangulation mismatch errors also. The dark X shaped errors. The main issues are coming from the bad UVs/smoothing. I would suggest to check out this thread:
http://polycount.com/discussion/107196/youre-making-me-hard-making-sense-of-hard-edges-uvs-normal-maps-and-vertex-counts/p1

Stranger

Obscura said:

Stranger said:

Other then the hard edge, and the gradients 

Lol, it just gets worse. Now you got some triangulation mismatch errors also. The dark X shaped errors. The main issues are coming from the bad UVs/smoothing. I would suggest to check out this thread:
http://polycount.com/discussion/107196/youre-making-me-hard-making-sense-of-hard-edges-uvs-normal-maps-and-vertex-counts/p1

Not my mesh. Not my bakes.

19ulrich90

Thx @Obscura , sorry I see now that I posted wrong lowpoly obj with that one hardened edge. But as I have shown in the video I softened whole mesh (which is the same in Maya as applying one smoothing group in 3dsmax) before exporting, and it also shouldn't have been the cause of flipped normals on the back side, since @Stranger achieved same look on both sides. I also mentioned in few posts above that moving UV shell of the mirrored side out of the 0 1 UV space didn't help. Seems like there's problem with my Maya displaying that back side. Maybe I'll even try it in 3dsmax. Also @Obscura , about geometry, in which parts am I wasting it? I'm opened for critique about that  

Obscura

Like this. Here is a comparison. But it could be optimized even further without loosing quality. Also one thing to not forget, this thing wouldn't get more than 500 pixels on the screen, even in an fps game. In a tps or something that isn't fps, its like between 50-70...

Obscura

19ulrich90 said:

Thx @Obscura , sorry I see now that I posted wrong lowpoly obj with that one hardened edge. But as I have shown in the video I softened whole mesh (which is the same in Maya as applying one smoothing group in 3dsmax) before exporting, and it also shouldn't have been the cause of flipped normals on the back side, since @Stranger achieved same look on both sides. I also mentioned in few posts above that moving UV shell of the mirrored side out of the 0 1 UV space didn't help. Seems like there's problem with my Maya displaying that back side. Maybe I'll even try it in 3dsmax. Also @Obscura , about geometry, in which parts am I wasting it? I'm opened for critique about that  

Well. Basically everything was done badly with your mesh, even if we don't care about the different smoothing on the 2 sides. 
errors:
1 - When you use mirrored or overlapped uvs, you need to offset the uvs.
2 - If you want the mirrored/overlapped uvs with one smoothing group, then the uv need to be split along the symmetry plane. Your setup can work only with hard edges
3 - One smoothing group can only work with 100% synced normals, but even then its not recommended, so your smoothing is also bad. If you want the one smoothing group, then add some chamfer to reduce the gradients.
4 - You didn't use edge padding - which is not the source of your issues, but its still a thing.


* Its not the fault of Maya, its the fault of how you did things. If you have proper setup, then any baker can give acceptable result.

19ulrich90

Thx! I'll remeber it for next models.

Now I just hope someone else comes up with the solution for that mirrored normal, so I can finally move on  

edit: didn't see your 2. reply when posting this

Obscura

I explained to you whats wrong and how to fix right? It baked just fine at me with mirrored uvs with both one or multiple smoothing groups, after I fixed the mistakes.

19ulrich90

I edited previous post. I will go through the errors you listed, thx.

19ulrich90

OK, focusing just on cross shape, I'm not sure if I solved my main problem with flipped normal, but this is how it comes out looking the same for both sides.
Instead of mirroring geometry, I duplicated it and rotated it around, and offset the duplicated UVs. I did everything the same just instead mirroring, I duplicated and rotated it.

In Maya



In Marmoset





@Obscura I will go through and also try to solve other errors you listed.

19ulrich90

So here are the two setups, one with softened edges/1 smoothing group, and one with hardened edges around UV shells/multiple smoothing groups, both offset UVs. I didn't want to split the marked UVs though, as it would create symetrical texture later when adding for example edge wear.


1SG bake


MultipleSG bake


I also deleted inner loop of faces as it woudn't be seen from any perspective. Renders in Maya look pretty identical from distance, from up close multiple smoothig groups setup is maybe looking a bit cleaner.



Added rar with OBJs and bakes if anyone would want to have a look.

defragger

unless you are not using the rest of the UV space you should do at least one more UV seam where I have drawn the red line. You could rearrange the UV shells and get a lot more texture-res and less wasted area. Also, use polygons where they are needed (like mentioned earlier by OBSCURA).

Using SM-groups by UV-shells is generally a good idea for hard-surface meshes. Your normal bakes are looking way better now.

19ulrich90

Thanks @defragger  ! Yes I know, UVs aren't arranged optimaly at the moment, as I'm just focusing on learning normal maps baking . I'll be using one uv space for whole sword at the end. Ok, I'll also optimize the mesh.

So basically, I should be paying attention that the bake doesn't have as much contrast (or gradient?) as in 1SG case (even though in this case render comes out fine)?

defragger

The thing is, everything works together. I would not suggest learning just one step at a time. Its a whole workflow and should be learned as such, because the quality and setup of your low- and high-poly meshes and UV mapping have a huge impact on your bakes. You will see when you make more complex meshes with heavily optimized low-poly and minimum of wasted UV space. It will be a bit more complicated. So always see it as a whole. That will also answer your question about the gradient in case of 1SG.

19ulrich90

Ok, I'm back after some break from 3d because ...life .

So I made UV layout for whole sword which looks like this.



I'm concerned about the blade though, as it will have lower texel density because its dimensions. Is it also ok to organise UVs like I did, and then crop the textures for example from 4096x4096 to 2048x4096?


Another weird problem that I'm having is normals on the grip. When I tried to bake normals for al the pieces at the same time in xNormals it gave me error about grip cage even though I didn't use it. So I baked normals for each piece one at the time and combined them in Photoshop. If I apply those combined normals to seperate grip piece of the sword it looks fine, but when I combine the pieces of the sword in one mesh normals of the grip are offset. 



EDIT: About the offset normals, looks like I've been messing with UV's after exporting seperate grip piece and before combining all the pieces, my bad. 

Synaesthesia

Personally, I'd go non-square with your UVs and set them up as 2k by 4k.

19ulrich90

Thx for your reply @Synaesthesia . What about texel density? Will the difference between blade and other parts be to visible if I leave it like that? Here is a snapshot of model with checkerboard.

JedTheKrampus

I'd also split the blade up into two or three sections

19ulrich90

Normals now looking ok, new UV layout with better texel density distribution, ID map with minor errors (can those be fully avoided, or is manual cleaning in photoshop usually required?)

codyaq2

I would scale the UVS 2 high to 1 wide, then squash them back to 1 to 1.
When you render out the maps at 2-1 it will retain perfect uvs.

19ulrich90

Getting there with Substance Painter Rust and scratches need to be added