Thanks everyone for chiming in. I used Mari a long time ago in production, think it was version 1 or something, and really liked it. But as others have pointed out, it is too expensive to buy for personal use.
It kinda spoiled me, though, so when I tried out Substance painter, it didnt seem sexy enough.
Thanks jack, for always keeping an eye on forums! Will be looking forward to see what game studios are doing with Mari. It is the best texturing software I have used, so would love to get to work with it again.
Mashru, how is it you dont feel ddo can help you get the results you want? I love its ability to easily keep track of masks and tweak spec, gloss and nrm values easily. Even though I mostly paint in stuff myself, like without it. So I use ddo more as a layer manager than a preset machine
I am still mostly grinding it out with photoshop need to do more tuts to kick up my 3d coat skill, but mari is entirely out of the question given the price.
Substance painter I am on the fence about seen a ton of good videos but not tried it for myself.
I just use dDo with a ton of hand-placed photo sourced detail, so it's not purely procedural. The next version of dDo looks a *lot* better, almost hand made, so that'll be definitely something to look into.
I just use dDo with a ton of hand-placed photo sourced detail, so it's not purely procedural. The next version of dDo looks a *lot* better, almost hand made, so that'll be definitely something to look into.
What do you think is putting the suite so much further ahead of ddo?
Well just the actual details generated are much better and it seems to blend in photos like an artist would. It's procedural but doesn't look procedural - just check out their workflow video here:
It's also PBR which is mega nifty for all these cool newfangled engines like UE4.
Oh, and Teddy, who made that video is a programmer and not an artist, so you can see how awesome results one can generate with it even when not being a pro at texturing.
Im actually in the beta, just havent gotten around to try it enough yet, unfortunately
The PBR confused me a little when plugging stuff unti UE4, honestly. But I probably just have to try it more than once
Yeah, I am super lucky!
Im trying to learn MODO at the time as well, so as soon as I figure out how to actually export this mesh without crashing, I will be able to do some stuff in the suite
The biggest problem I run in to is that I use adjustment layers A LOT in Photoshop. Solid color adjustment layers with masks are the meat and potatoes of my work flow. Unfortunately, nothing other than Photoshop supports them.
Yep, if anything Mari has a more powerful adjustment and masking system than even Photoshop,
Mari supports solid color layers and has very advanced Masking that enables you to use a full stack of layers as a mask. The easiest way to imagine this is your masks in PS having their own layer system.
Thanks everyone for tuning in, and especially you Jack!
Thats very very interesting! My workflow is the same as Dustin's, so that is the same problem I have had, as clients wants PSDs delivered.
Maybe I shoould read up more on Mari after all.
When I worked with it at an earlier studio, I completely loved Mari, but that was for film work. I didnt have have much luck when I tried doing a game asset in my sparetime. But again, that was version 1 or 1.2 or soemthing along those lines, and havent been at a studio with mari since. But I can see that 2.6 should have shaders closer to game engine results? :O
That would be absolutely perfect!
Can anyone confirm this? Is masking and stuff picking up the correct values from normal map as well (like cavity etc)?
Photoshop is really close on this one. You can import 3d files and paint directly on the surface of the 3d mesh - you just can't paint to layer masks - doh!
My problem is trying to explain myself to the tech guy in support. I think at Photoshop they imagine that all artists are using the 3d feature to place objects in a 2d scene? (only explanation I can think of) Not sure why I can't communicate the idea better.
I'm with Spoon and dustinbrown on this one. Maybe you guys can chime in. I think my own level of trying to explain the situation has hit a roadblock. If more artists get on board maybe Photoshop will be the best solution. The way I see it they are the root of the problem anyways. Industry standard and all that - and everyone wants PSD.
Have you tried using the 3d layer as a clipping mask and treating the layers as a projection, merging down to the 3d model when its complete?
3d painting on a model in photoshop does not use the PSD where the 3d Layer is for the texture. The texture is a separate temp file, that has nothing to do with the PSD where the 3d layer is, other than being referenced by it. Any layers you create above the 3d layer are 2d layers, and will not be on the model/texture unless you merge down to the 3d layer.
You can create 2d layers and use the 3d layer as a clipping mask, and should be able to merge-down once you're happy with the results.
Or, in the temp psb file that is your diffuse texture, create your rust mask, and select that layer & mask then close the psb (and save). When you paint on the model, it should paint on the mask - unless of course... this is the problem that you are describing
If you can I'd also recommend trying out Mari, its SO GOOD! Started using it today after some recommendation (mainly Mudbox user for texturing and sometimes ZB polypaint) and its awesome
How's mari on the learning curve? I got it last week and between learning that and substance painter/designer I haven't a clue where to go and which is best to learn.
In Substance Painter, you can paint in the 3D view on all the channels, Albedo, Roughness, Metalic, Normal. You can paint on the 2D view as well on the 3D view, and when painting on the 2D view, you can paint seamlessly if you want, so there's that. It's got layers and stuff, and you can do stuff like change your mesh if you suddenly decide that you want to change some uv's or something, and it'll do it's best to redo all the work you've done so far. Problem is, it's in beta, and it shows sometimes.
How's mari on the learning curve? I got it last week and between learning that and substance painter/designer I haven't a clue where to go and which is best to learn.
From what I've seen from mari it looks amazing.
I used Mari in production once, and I would say it is very easy to learn, and a powerful beast. The only reason I am not considering it, is because I cannot afford it as a single freelancer atm
Ispheria
Yeah, I bought substance Painter on steamsale, and it definitely going in the right direction. At the moment it just lacks too many features for me to completely "convert".
I don't know if you are familiar with Blender. I just mention it because it is totally free. I found a hacked workflow which isn't very elegant but works.
1. Create all of your solid color layers as rgba .png
2. Stack the textures in Blender in the same material. Set render mode in 3d window to "glsl" and material set to "shadeless."
3. Paint to the alpha channel of each layer - ta da! 3d painting on a solid color with adjustment layer.
4. Open the .png in 2d editor and assign them to layers using alpha for mask. This is the tedious part of the workflow. All I need is someone who can make a javascript in Photoshop for this. It would be nice actually to have a script going both ways from .psd > .png and from .png > .psd.
5. Blender is missing things like selection tools, gradients, fills and just about everything else in a 2d paint app - so you basically need to be good at hand painting. Sometimes limits force you to be more creative.
6. There is a plugin which allows you to select texture layers for painting in Blender which will help with this workflow.
I don't recommend this workflow for the faint of heart - the best point is that it will be totally free. *also Blender is lightning fast and will run on any platform. You could get some stripped down version of Linux and be running this on a toaster.
You might have "smooth stroke" switched on? Not sure unless I could see it. Might depend on your mesh + texture size as well. If you have a large mesh turning off "double sided" faces will have make a big difference.
More people should check out the trial version of MARI. It wasn't always well suited for game assets (tangent space normal issues, non-square textures etc), but they've fixed a lot of this stuff or introduced workarounds. Painting and seeing the real-time result while you still move your brush makes it so much easier to work with PBR, no more guesswork, and it's also probably the least destructive workflow you could think of.
The fact that it's so scalable that you could theoretically paint your Zbrush sculpts with thousands of textures tiles and then bake everything over to your lowpoly is just gravy.
Well there is a workaround, simply scale your UVs so they fill the lower half of UDIM 1001 (between 0-1). Then do your texturing, and once you export the textures, open them in Photoshop and crop.
If you use UVW xform modifiers and manual canvas size etc in Photoshop, then it's 100% accurate and there is no fiddling involved.
Okay, I think I have found a new approach in Photoshop.
Create the solid color layer and then control it with a clipping mask layer. The clip mask can be painted to directly from the 3d view. This is the best solution I've found so far.
I can paint directly on the 3d surface and the method of hiding/revealing layers is non-destructive.
Mari looks amazing - I look forward to working with it in the future when WETA gives me a call
Previously I'd use Zbrush's projection painting while working on a texture. I'll still do that from time to time but have started using Photoshop's 3d painting for projection stuff lately.
I'm familiar with touching up seams in Photoshop's 3D mode so your workflow definitely interests me, moose. Could you provide a quick walk through of how you go about using or for what parts of the process you use the programs while making a texture?
I'm a bit tired so hopefully what I meant to say came across clear. :icon15:
I am using 3d coat and photoshop at work for texturing. Maybe there is no masking in coat like in ps but you can always use linked layers. Its quite useful from time to time and it is a sort of masking. Also coat works very good with ps. You can move fast between these apps. Also coat is great for normals painting. You can use linked layers to achieve layer brush style
Mari seems to be having some really sexy projection settings. Edge Mask, Ambient Occ mask etc seem really really nice. Does any of you mari guys know, if that is purely taking geo into consideration, or if it works with normal maps as well? As in, could it be used on a 1 polygon plane with a normal map?
The more I see and read from Mari, the better it seems. Im just in doubt if it is worth the big investment
It's only taking geo into consideration. However, there is a channel mask option, so you can create a channel and have all kinds of textures in there for projection masking, such as cavity, micro AO, and whatnot.
There are a few caveats to texturing game art lowpolies with MARI. First, it uses a rather exotic way to calculate tangent space, so your normal maps will look bad in it. This can be helped if you bake object space normals, load them in mari, and then apply a world to tangent adjustment layer on top. However, baking proper object space maps can be tricky in itself if you're not used to it (for example, if you have repeating elements that share the same UV space, they have to have the same world space orientation during baking)
Then, textures have to be square, but it can be helped by adjusting the UVs and cropping the exported output.
Can't think of any more at the moment, but I'm mainly texturing highpolies anyway.
All our character artists here at work are texturing highpolies and bake the result over to the in-game mesh (with LODs) and I'd wager it's the same for most AAA studios.
Much like Substance Designer, Substance Painter seems to be focused primarily on producing post-effects like wear and debri.
I tried using it to create a hand-painted texture, and was not at all impressed with the result.
I'm still working on my exact process, but at the moment I'm using SD to bake out my normal map, then creating a combination of AO and Curvature to use as a base texture. It not only lets me know where the UV space is, but also shows all the edges and shadows.
I then take that AO/Curve map and import it into Artrage as a backdrop layer and paint over the top of it.
I also open up Marmoset, load my mesh and the Artrage texture into that, and occasionally save the texture in Artrage and flip over to Marmoset to check out exactly how it's turning out in 3D (Since Marmoset is not only an awesome game-quality render, but also updates in real time as changes are made to the texture).
If I had a second monitor it'd work even better, as I could have Artrage open on one and Marmoset on the other.
After I finish it in Artrage, I take it back into Substance Desiger/Painter to add some post effects if necessary.
On the whole, I just cannot get good results trying to make hand painted textures in 3D painters.
Do you mean handpainting from scratch, or handpaint-style polypaint/AO/curvature baking? If it's the former, I don't see what's stopping you from achieving the same result as in photoshop, at least with MARI. I have only briefly used substance painter and wasn't impressed due to the 2048 max texture size and not being able to put substances on layers and paint into masks.
Do you mean handpainting from scratch, or handpaint-style polypaint/AO/curvature baking? If it's the former, I don't see what's stopping you from achieving the same result as in photoshop, at least with MARI. I have only briefly used substance painter and wasn't impressed due to the 2048 max texture size and not being able to put substances on layers and paint into masks.
no all the baking you seem to be doing by hand. that would drive me nuts
no all the baking you seem to be doing by hand. that would drive me nuts
Thats how I read it as well. Do we misunderstand you?
As for Mari painting highpolies, am I the only one who bakes to the low poly and paint that one? Have I completely misunderstood the workflow? :O
And how does that work in Mari? I thought Mari only had PTEX and not vertex colors. How is that baked to the low poly? Im totally new to this perspective
This is what I am probably misunderstanding:
"I then take that AO/Curve map and import it into Artrage as a backdrop layer and paint over the top of it."
I understand it as him using the bakes as "reference", a layer in the background, and paints everything on top of that one. I dont know artrage, so I might miss something in the terminology.
This is what I am probably misunderstanding:
"I then take that AO/Curve map and import it into Artrage as a backdrop layer and paint over the top of it."
I understand it as him using the bakes as "reference", a layer in the background, and paints everything on top of that one. I dont know artrage, so I might miss something in the terminology.
No, you have it right.
People need to start quoting here, 'cause I have no idea who's talking to who.
Thanks for confirming, Grim.
In that case, you kinda are baking by hand, though. Right? If so, why? :O
I don't know what you mean when you say "by hand".
Like you said, I'm only really using the bake as a reference to help me better grasp the exact shape while painting the diffuse.
It's not like I'm painting out a normal/specular map or something.
Do you mean handpainting from scratch, or handpaint-style polypaint/AO/curvature baking? If it's the former, I don't see what's stopping you from achieving the same result as in photoshop, at least with MARI. I have only briefly used substance painter and wasn't impressed due to the 2048 max texture size and not being able to put substances on layers and paint into masks.
You must have used an early beta, you can do all that now
4k support is coming next month.
Thanks for the tuning in, Jerc!
I bought both SD and SP, but still waiting for a few more feature before it can replace my current workflow. Nice to see it getting closer and closer
unless a client wants me to bake in maya or max, i just export my data and hit bake and it will bake 1 or 100 charactzer with 1 or 100 subsets it doesn't matter it will bake anything i throw at it without ever touching xnormal. as far as i know SD doesn't support automation. Thats why i asked. i don't want to fiddle with highpolies in max or anything just set up the cage and bake, no matter how much data i need to bake.
You can automate everything in SD. Actually not in SD directly, but there are command line tools for the bakers, so you can call them from any software or just through a script and bake as many meshes as needed at once.
Thats how I read it as well. Do we misunderstand you?
As for Mari painting highpolies, am I the only one who bakes to the low poly and paint that one? Have I completely misunderstood the workflow? :O
And how does that work in Mari? I thought Mari only had PTEX and not vertex colors. How is that baked to the low poly? Im totally new to this perspective
For character work, what I think our guys do is take the lowest subdivision level, unwrap it max or maya, and bring it back into zbrush. Then they switch to higher subd levels and use project to conform it to the high-res shape. It looks the same as the sculpt now, but has UVs!
Or they use a decimated mesh.
PTEX is a possibility but it has too many downsides. You get ugly filtering in MARI which makes it look like there are seams everywhere once it switches to higher mip levels. Tangent space normals don't work with it. 2D based effects like gaussian blur don't work with it. Once imported, you can't change the mesh anymore or the textures will become all messed up. Also, there is no subdivision implementation in MARI, which would alleviate at least some of these issues.
For hardsurface work, if you use traditional turbosmooth with support loops, it can be very hard to unwrap the basemesh and have undistorted UVs. If you use double TS/opensubdiv/TS pro, it's actually very easy. If you look in the misc work section on my portfolio, I've done the diving helmet in this fashion and will do the same with the metroid ship.
The advantages of the approach are:
- You can very easily apply cloth detail without the need of sculpting, like vertical lines in a shirt that flow along with the folds, just unwrap is straight and apply a displacement map
- You don't have to bother with so much micro detail in the sculpt as you can use tangent space normal maps or micro displacement maps in MARI, and later bake them over to the in-game mesh
- You don't need to stress out so much over triangle count and skinning, as you can simply change the low-res mesh as needed and bake everything over. Also useful for cutscene quality models vs LOD3 models.
- For hardsurface stuff, you can bake the textures to versions of the mesh that need different topology. For example, the weapon in a first person shooter needs a ton of detail at the sights or other parts closer to the camera, but little near the muzzle, and the side facing away from the camera might be missing. Whereas the third person weapons you pick up from the ground need to have consistent detail all around, but an overall much lower polycount
- You can also do kitbashing with a library of textured highpoly parts.
It may sound like a lot of extra work, but you can mostly use the UVed highpoly basemesh as the lowpoly with some adjustments.
Replies
It kinda spoiled me, though, so when I tried out Substance painter, it didnt seem sexy enough.
Thanks jack, for always keeping an eye on forums! Will be looking forward to see what game studios are doing with Mari. It is the best texturing software I have used, so would love to get to work with it again.
Mashru, how is it you dont feel ddo can help you get the results you want? I love its ability to easily keep track of masks and tweak spec, gloss and nrm values easily. Even though I mostly paint in stuff myself, like without it. So I use ddo more as a layer manager than a preset machine
Substance painter I am on the fence about seen a ton of good videos but not tried it for myself.
What do you think is putting the suite so much further ahead of ddo?
https://www.youtube.com/watch?v=YyJJAp17K-Y
It's also PBR which is mega nifty for all these cool newfangled engines like UE4.
Oh, and Teddy, who made that video is a programmer and not an artist, so you can see how awesome results one can generate with it even when not being a pro at texturing.
The PBR confused me a little when plugging stuff unti UE4, honestly. But I probably just have to try it more than once
I'd definitely give it a whirl and see what you can generate with it!
Im trying to learn MODO at the time as well, so as soon as I figure out how to actually export this mesh without crashing, I will be able to do some stuff in the suite
afaik Mari supports adjustment layers?
Mari supports solid color layers and has very advanced Masking that enables you to use a full stack of layers as a mask. The easiest way to imagine this is your masks in PS having their own layer system.
Thats very very interesting! My workflow is the same as Dustin's, so that is the same problem I have had, as clients wants PSDs delivered.
Maybe I shoould read up more on Mari after all.
When I worked with it at an earlier studio, I completely loved Mari, but that was for film work. I didnt have have much luck when I tried doing a game asset in my sparetime. But again, that was version 1 or 1.2 or soemthing along those lines, and havent been at a studio with mari since. But I can see that 2.6 should have shaders closer to game engine results? :O
That would be absolutely perfect!
Can anyone confirm this? Is masking and stuff picking up the correct values from normal map as well (like cavity etc)?
Photoshop is really close on this one. You can import 3d files and paint directly on the surface of the 3d mesh - you just can't paint to layer masks - doh!
My problem is trying to explain myself to the tech guy in support. I think at Photoshop they imagine that all artists are using the 3d feature to place objects in a 2d scene? (only explanation I can think of) Not sure why I can't communicate the idea better.
I'm with Spoon and dustinbrown on this one. Maybe you guys can chime in. I think my own level of trying to explain the situation has hit a roadblock. If more artists get on board maybe Photoshop will be the best solution. The way I see it they are the root of the problem anyways. Industry standard and all that - and everyone wants PSD.
LINK:
http://feedback.photoshop.com/photoshop_family/topics/photoshop_layer_mask_on_3d_model
3d painting on a model in photoshop does not use the PSD where the 3d Layer is for the texture. The texture is a separate temp file, that has nothing to do with the PSD where the 3d layer is, other than being referenced by it. Any layers you create above the 3d layer are 2d layers, and will not be on the model/texture unless you merge down to the 3d layer.
You can create 2d layers and use the 3d layer as a clipping mask, and should be able to merge-down once you're happy with the results.
Or, in the temp psb file that is your diffuse texture, create your rust mask, and select that layer & mask then close the psb (and save). When you paint on the model, it should paint on the mask - unless of course... this is the problem that you are describing
^ this is what I am saying. I can paint on the solid color layer - not on the layer mask.
How's mari on the learning curve? I got it last week and between learning that and substance painter/designer I haven't a clue where to go and which is best to learn.
From what I've seen from mari it looks amazing.
I used Mari in production once, and I would say it is very easy to learn, and a powerful beast. The only reason I am not considering it, is because I cannot afford it as a single freelancer atm
Yeah, I bought substance Painter on steamsale, and it definitely going in the right direction. At the moment it just lacks too many features for me to completely "convert".
1. Create all of your solid color layers as rgba .png
2. Stack the textures in Blender in the same material. Set render mode in 3d window to "glsl" and material set to "shadeless."
3. Paint to the alpha channel of each layer - ta da! 3d painting on a solid color with adjustment layer.
4. Open the .png in 2d editor and assign them to layers using alpha for mask. This is the tedious part of the workflow. All I need is someone who can make a javascript in Photoshop for this. It would be nice actually to have a script going both ways from .psd > .png and from .png > .psd.
5. Blender is missing things like selection tools, gradients, fills and just about everything else in a 2d paint app - so you basically need to be good at hand painting. Sometimes limits force you to be more creative.
6. There is a plugin which allows you to select texture layers for painting in Blender which will help with this workflow.
wiki.blender.org/index.php/Extensions:2.6/Py/Scripts/3D_interaction/Texture_paint_layers
I don't recommend this workflow for the faint of heart - the best point is that it will be totally free. *also Blender is lightning fast and will run on any platform. You could get some stripped down version of Linux and be running this on a toaster.
It's nice to know about this alternative.
I've tried blender but i get tons of brush lag, it would be fine if i didn't have that.
i'm not sure but the problem seems to stem from pressure sensitivity.
I didn't try working without pressure.
But I know my pc should be able to handle it.
The fact that it's so scalable that you could theoretically paint your Zbrush sculpts with thousands of textures tiles and then bake everything over to your lowpoly is just gravy.
They fixed this!? :O
That was an instant turndown for me, back when I tried it last time.
If you use UVW xform modifiers and manual canvas size etc in Photoshop, then it's 100% accurate and there is no fiddling involved.
This should give a good overview of Mari's advanced masking. It shows using mask groups (I.e layered masks) to get very fine grain control.
Create the solid color layer and then control it with a clipping mask layer. The clip mask can be painted to directly from the 3d view. This is the best solution I've found so far.
I can paint directly on the 3d surface and the method of hiding/revealing layers is non-destructive.
Mari looks amazing - I look forward to working with it in the future when WETA gives me a call
Here are the rest of the tutorials if you want to brush up for that call.
http://vimeo.com/channels/maritutorials/
I'm a bit tired so hopefully what I meant to say came across clear. :icon15:
The more I see and read from Mari, the better it seems. Im just in doubt if it is worth the big investment
There are a few caveats to texturing game art lowpolies with MARI. First, it uses a rather exotic way to calculate tangent space, so your normal maps will look bad in it. This can be helped if you bake object space normals, load them in mari, and then apply a world to tangent adjustment layer on top. However, baking proper object space maps can be tricky in itself if you're not used to it (for example, if you have repeating elements that share the same UV space, they have to have the same world space orientation during baking)
Then, textures have to be square, but it can be helped by adjusting the UVs and cropping the exported output.
Can't think of any more at the moment, but I'm mainly texturing highpolies anyway.
All our character artists here at work are texturing highpolies and bake the result over to the in-game mesh (with LODs) and I'd wager it's the same for most AAA studios.
I tried using it to create a hand-painted texture, and was not at all impressed with the result.
I'm still working on my exact process, but at the moment I'm using SD to bake out my normal map, then creating a combination of AO and Curvature to use as a base texture. It not only lets me know where the UV space is, but also shows all the edges and shadows.
I then take that AO/Curve map and import it into Artrage as a backdrop layer and paint over the top of it.
I also open up Marmoset, load my mesh and the Artrage texture into that, and occasionally save the texture in Artrage and flip over to Marmoset to check out exactly how it's turning out in 3D (Since Marmoset is not only an awesome game-quality render, but also updates in real time as changes are made to the texture).
If I had a second monitor it'd work even better, as I could have Artrage open on one and Marmoset on the other.
After I finish it in Artrage, I take it back into Substance Desiger/Painter to add some post effects if necessary.
On the whole, I just cannot get good results trying to make hand painted textures in 3D painters.
no all the baking you seem to be doing by hand. that would drive me nuts
Thats how I read it as well. Do we misunderstand you?
As for Mari painting highpolies, am I the only one who bakes to the low poly and paint that one? Have I completely misunderstood the workflow? :O
And how does that work in Mari? I thought Mari only had PTEX and not vertex colors. How is that baked to the low poly? Im totally new to this perspective
I'm completely lost now!
This is what I am probably misunderstanding:
"I then take that AO/Curve map and import it into Artrage as a backdrop layer and paint over the top of it."
I understand it as him using the bakes as "reference", a layer in the background, and paints everything on top of that one. I dont know artrage, so I might miss something in the terminology.
No, you have it right.
People need to start quoting here, 'cause I have no idea who's talking to who.
In that case, you kinda are baking by hand, though. Right? If so, why? :O
I don't know what you mean when you say "by hand".
Like you said, I'm only really using the bake as a reference to help me better grasp the exact shape while painting the diffuse.
It's not like I'm painting out a normal/specular map or something.
You must have used an early beta, you can do all that now
4k support is coming next month.
I bought both SD and SP, but still waiting for a few more feature before it can replace my current workflow. Nice to see it getting closer and closer
unless a client wants me to bake in maya or max, i just export my data and hit bake and it will bake 1 or 100 charactzer with 1 or 100 subsets it doesn't matter it will bake anything i throw at it without ever touching xnormal. as far as i know SD doesn't support automation. Thats why i asked. i don't want to fiddle with highpolies in max or anything just set up the cage and bake, no matter how much data i need to bake.
For character work, what I think our guys do is take the lowest subdivision level, unwrap it max or maya, and bring it back into zbrush. Then they switch to higher subd levels and use project to conform it to the high-res shape. It looks the same as the sculpt now, but has UVs!
Or they use a decimated mesh.
PTEX is a possibility but it has too many downsides. You get ugly filtering in MARI which makes it look like there are seams everywhere once it switches to higher mip levels. Tangent space normals don't work with it. 2D based effects like gaussian blur don't work with it. Once imported, you can't change the mesh anymore or the textures will become all messed up. Also, there is no subdivision implementation in MARI, which would alleviate at least some of these issues.
For hardsurface work, if you use traditional turbosmooth with support loops, it can be very hard to unwrap the basemesh and have undistorted UVs. If you use double TS/opensubdiv/TS pro, it's actually very easy. If you look in the misc work section on my portfolio, I've done the diving helmet in this fashion and will do the same with the metroid ship.
The advantages of the approach are:
- You can very easily apply cloth detail without the need of sculpting, like vertical lines in a shirt that flow along with the folds, just unwrap is straight and apply a displacement map
- You don't have to bother with so much micro detail in the sculpt as you can use tangent space normal maps or micro displacement maps in MARI, and later bake them over to the in-game mesh
- You don't need to stress out so much over triangle count and skinning, as you can simply change the low-res mesh as needed and bake everything over. Also useful for cutscene quality models vs LOD3 models.
- For hardsurface stuff, you can bake the textures to versions of the mesh that need different topology. For example, the weapon in a first person shooter needs a ton of detail at the sights or other parts closer to the camera, but little near the muzzle, and the side facing away from the camera might be missing. Whereas the third person weapons you pick up from the ground need to have consistent detail all around, but an overall much lower polycount
- You can also do kitbashing with a library of textured highpoly parts.
It may sound like a lot of extra work, but you can mostly use the UVed highpoly basemesh as the lowpoly with some adjustments.