How to approach baking clothes
Hi
I was wondering if some experts could help me out with something... What's the best way of approaching the baking of clothing? Is thickness added to the low poly meshes first? I figured it would need to be added afterwards, but when I tried that, there were Normal issues around the edges. I'm guessing the right approach would be to add a separate Smoothing Group to those outer rings that run through the middle and give them their own UVs...?
Also, for things like sleeve pieces, plackets, pockets etc. how are they handled? Are they merged with the main piece, like, say... a shirt...? I currently have a shirt with plackets and rolled sleeves that are part of my high and low polys. However, Substance Painter is projecting those elements below/onto the surface as well. Do those things need to be separate? And if so, wouldn't merging them back afterwards cause similar problems as adding thickness at the end? I imagine the Normals etc would change. If they don't need to be separated, how can the projection to the surface of the mesh be avoided? Ignore Backfaces is enabled by default for me... I've also tried different frontal and rear distances. Is this something only a cage could prevent?
Any suggestions would be greatly appreciated. :)
Replies
Unless you have special requirements for the game mesh/s then just tackle it like you would normally. I split the skin and costume meshes and then further split cloth and hardware to avoid baking artefacts. Your description is confusing maybe a coupe of screenshots would help to clarify your problem.
Sorry, this is what I'm referring to... In the first example,it's the edges/rings of sleeves from adding thickness:
And then the other issue is when parts of it, such as the collar/plackets, project onto the surface below:
I have no idea why you would need to bake the inside of a garment. Little bits like the first part of the sleeve ends, or collar area where you want to avoid seeing into the model are normal. I haven't had the problem you are experiencing with the fold edges. The only thing I can think of is to check and unify normals on the hi and lo poly. Like I said I have never done the whole inside of a garment, but if I had to I would keep the edges of the garment whole (cuffs, collars, open shirt bottom) and just cut out the inside and separate the low poly in pieces to avoid baking echoes by putting them into separate selection sets in substance.
As kanga mentioned, worry only about the outside of the garments. The inside of a garment (if at all visible) can just be duplicated of the outer, delete what you don't need, and pushed inwards with a push modifier (stacking UVs as well).
The thickness of a garment could also be done post-bake. The edge highlight of the garment thickness could then just be handled with weighted normals rather than trying to come up with a highpoly of a garment with thickness + inners, and relying on baked tangent normals.
Hmm... Okay. So, you're suggesting to tackle the problem areas with the sleeves and collar edges after baking by 'pushing' them away so they can't be seen using Weighted Normals...? From what I've gathered about Weighted Normals, as I don't think I've used them before, is that essentially chamfering the edges so it'll make them rounder...? Also, would you recommend a Push Modifier over a Shell?
To be honest, looking at the reference model I have, none of the clothing meshes have interiors. They've been closed off with geometry and everything's put together so you can't see inside. :)
Yeah alternatively, you could just simply shell yes, then UV the fresh "thickness" geometry to something else, since it'd be skewed/stretched from the existing textures.
You could either not worry about "weighted normals" and just give the thickness geometry a smoothing group split/hard edge, or have the entire thing one smoothing group and weight normals to get a nice edge highlight/shading result.
Yeah, I think I mentioned in the first post that's what I'd done - I'd given them their own smoothing group, but I split them so they had their own UV. I remember being taught, that if anything has it's own Smoothing Group, it needs its own UV...? Or is it the other way around - that a new UV needs its own Smoothing Group...? 🤔 But yeah, I may just have to try and smooth/curve that edge and have it so it's part of the main sleeve. Thanks for the suggestions. :)
That just leaves the other problem - where the underside of that collar is being projected onto the surface. What would you recommend there? I did try using a cage the other day, but I still had the same problem. Would those need to be split off as their own piece? But if so, won't welding them back afterwards cause problems?
Thanks again
I would say that your issues mostly come from thinking too much in terms of "how to bake this or that part". If you were to start thinking more about the actual model being worked on (ie the so-called "low" = the one that actually goes in the game), many of these issues will just solve themselves. Spend more time and energy on that, as opposed to considering the lowpoly meshing like just a technical step at the end. As indeed your screenshots seem to tell the story of a lowpoly model being treated as an afterhtought.
This can be seen on the shot showing the cuffs for instance, as the edge perimeter seems uneven and really not planned ahead - whereas you should know even before any modeling happens how many edges the perimeter of each limb is going to receive. 8, 16, 24, 32, whatever the count may be.
Or on the shot showing the collar, where there seems to be no matching at all between your low and what is being projected onto it.
First way, you got it correct. New smoothing group -> must have uv split. New uv island =/= new smoothing group (you can split uv islands without needing to give it a smoothing split).
It's tricky, but comes down to how much fidelity you'd like. You could simply bake down the collar to flat geo, so there isn't an actual problematic silhouette of the collar sticking up. Otherwise, it's a matter of having a hand-curated cage that doesn't intersect/overlap to surrounding areas.
Hmm. Okay. Personally, I would like a bit more fidelity to it rather than baking it straight to a flat plane. :) Is it possible to weld the piece back after the baking? If not, I suppose I could just have it as a separate piece and keep it 'floating', so it looks like it's attached. And then attach all the pieces together.
The original low poly was a retopped job from the high poly. I had the high poly parts separated and then used Maya's Quad Draw tool and tackled each part separately and then welded parts together afterwards. As it was a retop job, do the number of edges still need to be considered...? Usually, I keep things big and chunky to begin with so as not to overcomplicate things and confuse myself. Afterwards, I'll give the entire mesh a subdivision so it matches the high poly a bit better. Chances are, it looks a bit rushed/sloppy as I was just doing a quick test to demonstrate the issue I keep getting. :)
"As it was a retop job, do the number of edges still need to be considered...?"
Oh hell yeah. if you don't decide beforehand how many edges the cross section of each limb is supposed to have, you are guaranteed to end up with sloopy results imho. Heck, I could tell it was the case just be looking at that cuff screenshot, without the help of a wireframe.
Again - the low is not supposed to be some kind of afterthought. As density gets higher, it becomes more and more important for it to look good and tight in and of itself.
Of course I do understand that some clients may want to artdirect the smallest detail of a sculpt in the most anal way (sometimes for good reasons), which indeed makes the low feel like an extra tedious step after all that. But if you are working on your own stuff then you have all the freedom to work non-linearly, tackling both at the same time.
You might even get some faster and cleaner results by forcing yourself to built the whole asset using subdivision modeling (leveraging creasing, and so on), only using sculpting for the folds. And then the subdiv cage of the high can very easily become the starting point of your low - That way you don't have to do any tedious Quad Draw retopo anymore. But of course it's never that simple either.
Sorry, I'm completely stressed out of my head at the moment, so my brain is really struggling to process what you mean. I think I'm burning myself out. Or starting to. 😣 I would have liked to have mastered Laura Gallagher's method on Outgang for getting clean topology - https://www.youtube.com/watch?v=o_Q-N8CoyCU&ab_channel=Outgang - although the suggestion that no need for manual retopology is a joke. It's a bit misleading. Unless you want everything stitched together with that method, you HAVE to retop, like outlined in this guide here: https://www.artstation.com/artwork/zZOOL. I had to ditch that due to time issues and because I wasn't getting anywhere with it. So I had to manually retop over the high poly. But I'm really confused with what you're suggesting though when you say to build it using subdivision modelling. The entire garment was made in Marvelous Designer and then refined in ZBrush, so all the creases and folds are done. Laura's method would have enabled me to have given the high clean topology with UVs and then I could have lowered the subdivision level in ZBrush to get the low poly. Sadly that didn't happen. :-\
@DustyShinigami
pior knows his stuff. It just will take a lot more practice before you start to understand finer details.
WHen I was first learning baking it seemed impossible. He gave me some advice (and others as well) and it didnt exactly make sense. But after I kept working then I was better able to understand what they were talking about.
If somebody tells you a solution, you cant really understand it until you fully understand the problem. But the only way to understand the problem is just keep working. Failures are how you will learn to understand the problems, then these solutions people are offering will make a lot more sense. And that saves you hundreds of hours of time because you dont have to come up with solutions on your own at that point.
There is always 10,000 ways to do what you want to do. Every artist out there teaching their workflow has just found some way that works for them. You can try to follow it, but there is no rules, so you dont need to stress over anything. As long as you keep finishing art you'll be fine. But return to post like this from time to time and it will make more sense.
Heya -
I don't mean that no retopo at all is needed. What I mean is that assuming that everything can (or should) be fully zeedbrushed or fully MDeed is a mistake or at least an oversimplification that can cost hours and hours of wasted time.
Many parts of a garment can, indeed, be made using good old subdiv modeling. The collar that you are having trouble with is a perfect example of that : you could build it easily with some simple (but clean and accurate) geo ; then give it some thickness, and subdivide further and sculpt a few well placed folds where needed. And then use a version without the subdivision to get a headstart on your low. In some cases it can be faster than building it in MD and having to process/retopo the results. You'd do the sole of a shoe exactly that way, so why not do so with a collar too ? That's all I am saying.
The resources you linked are of course of great quality, but nothing guarantees you that these are the most efficient approaches (as a matter of fact both seem somwhat convoluted to me, at least in the way they are presented). And they are respectively 2 and 4 years old, and since then Blender has gained a lot of traction. While it cannot do cloth sim like MD does, it does however allow you to work on the high (subdiv and sculpt) and the low in a unified environement all at the same time. This alone unlocks a ton of potential.
Also if you find yourself retopoing many MD pieces by hand, you have some optimization to do in your workflow since a Quadremesher pass will let you do that near automatically, following the parts of the pattern.
And again at the end of the day no matter the approach, it's the low that matters. Don't get too worried about a hypothetical perfect way of doing things, and focus on cleaning up what you have at the moment. And on the next piece you'll review all the techniques you know and decide a bit more proactively what to use and why.
Now that I've had a break and tore myself away from what I was doing, what you said now makes sense. 😂 I've actually been doing subdiv modelling for accessories for a couple of characters, including the soles of some boots. It's helped speed things up so I've not had to retpo them all. I've made the low polys first and then increased the subdivisions/added a smooth in Maya/ZBrush to get the high polys.
But yeah, I'm not after a definitive way or ultimate workflow. :) I understand there are tonnes of ways of tackling things. I was just after suggestions on what I could potentially do, or how others would approach, baking clothes with thickness. Or if no thickness should be added until after. But I'm going to scrap having any thickness and hide any holes/gaps. I'll close the holes with some quads/tris. It won't matter if they look messy in bakes/textures as they won't be seen. :)
Also, I don't tend to bother with a Quad ZRemesher for clothing. I tend to prefer manually retopping them so I can definitely get the edge flows correct. Usually based on the 2D patterns made in Marvelous Designer. That way, I can UV cut the parts exactly, which is what I've done with these current clothes. :)