uvs: efficiency vs accuracy?
mod
just a general open question, I've seen a fair few uv maps now and I notice some are layed out super square and grid like (dazs dom war general comes to mind), and others (like mine) are all wibbly. The grids you can get excellent straight lines I guess from the pixels, but can suffer from distortion if your edgeloops follow contours on the body. The wibbly relaxed ones come out with minimum distortion, but you might not be able to fit as much on the sheet efficiently. any comments on which you prefer in what situations?
Replies
It's much easier to hide seams on most objects this way, plus it's easier and less mind-bending to pack everything together tightly and consistently.
Most of the time people won't notice a small amount of distortion anyway, and it'll save you a lot of headaches painting it, especially lower-res stuff. All unwraps I've seen for handheld/super-low-res stuff have been really tightly packed and perfectly straight lines. Avoids dodgy aliasing that way.
Alex
No, I'm partly joking. A lot of the time though, if you're modelling in quads especially, it's fairly easy to keep things lined up for ease of UVing, I usually think about where my seams will go as I model.
Alex
Think of it this way, if you have nice rectangular chunks, it's easier to arrange them tightly, so you can scale them all up. Now you have more texels on the model than you would have with the less distorted but more difficult to arrange chunks. So while there might be some distortion, your texture is actually higher res. Plus if you model in a neat manner, your UVs should start out fairly gridlike anyway.
I arrange the pieces willy nilly paying attention to what needs a nice clean edge and making sure it gets it. It helps minimize the distortion but maximize the detail. Honestly the end user doesn't give a rats ass how easy your UV layout was but if they see some weird distortion they notice.
Question for the griders: If you have to paint a distorted pocket to make up for the grid did it really save you that much time and was it really that much easier? (Not trying to be a dick, just asking)
It's pretty common when you work willy nilly to have a weird long string of polys hanging off a chunk, if that string doesn't impact large details (most of the time they are in odd places people don't look) scale it back so it makes arranging much easier. No need to force the chest piece to be small and misaligned because the few polys under the crotch are sticking off the left leg.
Besides "Pack UV's" normally does a good job of arranging pieces semi grid like and that is what I use as a base to start arranging.
In the end it all comes down to what looks good and is the fastest way for you.
One undistorted and "sloppy" one for painting that will then later be baked into the packed uv's.
I find lying out good uv's for nomalmapping especially challenging. E.g. if you are working with an engine which doesn't deal so well with differently rotated uv-chunks. There it is a big plus to have them layed out gridlike and as much parallel seams as possible etc.
Easy!
Question for the griders: If you have to paint a distorted pocket to make up for the grid did it really save you that much time and was it really that much easier? (Not trying to be a dick, just asking)
[/ QUOTE ]
Like Pior and Mop have alluded to, if you model in a clean manner, gridding your UVs really won't make any distortion. I've never had to paint a distorted shape to correct for square uv layout.
You have to decide where it's appropriate and where it isn't. It also only works if you've modeled in a way that keeps square UVs in mind. It's hard to explain and is definitely something you learn and discover the more you do it.
The main advantage of making the border edges straight is ease of painting, and making the texture seem sharper than it really is.
You deal with distortion by figuring out how your texture is going to look like (a strong, finished concept helps here), and by adjusting the UVs in a manner that shifts the distortion to less seen areas.
Smooth, non-detailed areas will never show distortion. So think about that. Areas that you know will have a lot of detail, you'll want to have really good UVs.
*shrug*
It's a process you tweak over and over as you go along. You have to find a good comprimise between the different elements... the square-ness of the UVs, the amount of distortion you're going to have to achieve said UVs, texel density, etc, etc.
And remember, UVs aren't the only thing that affects the distortion of your texture. Sometimes I find that moving verts helps, too.
then i paint it and when finished, bake this mess into a second uv-set which is tightly packed, uses square-ish shapes for uv-chunks whereever possible and fit's onto a single texture page.
saves my life - along with hand-editing normals to make floating geometry blend seamlessly onto underlying surfaces. thanks to you-know-who-you-are for letting me pick up your tricks and scripts!
Alex
That way you get both! Can't wait to try it out.
hahaha, the industry to a tee
I deconnect edges and autorelax everything superfast. It means that I'm well into the texturing by the time the guy next to me is done making his advanced "super-efficient" UVs. Of course my results look better than his then, regardless of the extra rez he's got.
[/ QUOTE ]
You took the words from my mouth. And Ben took the poop from it
[ QUOTE ]
Question for the griders: If you have to paint a distorted pocket to make up for the grid did it really save you that much time and was it really that much easier? (Not trying to be a dick, just asking)
[/ QUOTE ]
Like Pior and Mop have alluded to, if you model in a clean manner, gridding your UVs really won't make any distortion. I've never had to paint a distorted shape to correct for square uv layout.
[/ QUOTE ]
thats bs, if you have any sort of natural taper in your mesh, like an arm, and you map it to a square uv... and say you need to paint a stripe down it, it will be distorted, c'mon thats just common sense, regardless of how clean you model.
[ QUOTE ]
[ QUOTE ]
Question for the griders: If you have to paint a distorted pocket to make up for the grid did it really save you that much time and was it really that much easier? (Not trying to be a dick, just asking)
[/ QUOTE ]
Like Pior and Mop have alluded to, if you model in a clean manner, gridding your UVs really won't make any distortion. I've never had to paint a distorted shape to correct for square uv layout.
[/ QUOTE ]
thats bs, if you have any sort of natural taper in your mesh, like an arm, and you map it to a square uv... and say you need to paint a stripe down it, it will be distorted, c'mon thats just common sense, regardless of how clean you model.
[/ QUOTE ]
Looks like you're ri... oops, no you're wrong.
http://www.poopinmymouth.com/3d/naruto.jpg
http://www.poopinmymouth.com/3d/mr_mint_tex.jpg (the wiggle in this one isn't to adjust for distortion, it's to keep it from looking razor straight. If it was painted straight, it would appear straight on the model.
Maybe you just have poor visualization skills, or have never tried it. I looked for that example of Soul's with that chick with the red pants and leather vest, as he gridded that one and it had stripes on it that are perfectly straight. I'm sure Soul and Daz can chime in to it working too. I guess technically it scales a miniscule amount because of the taper, but that's nowhere near "bs" that's perceptually insignificant.
It's somewhat of a balance, but if our outsourcers sent work in to us looking pelt mapped, I'd send it back.
It's been touched on somewhere in here I think, but very VERY generally speaking I see the formula as the lower poly the mesh and target hardware specs, the more important 'squaring off' becomes.
For a cartoony model designed for handheld, like naruto, the square uv's thing can work just fine, since the character is mostly box-like and symmetrical. That's a special case and is not relevant to most of us who work on realistic 5k+ characters where the curve of the back, jutting of the stomach, breasts, and any other feature spells disaster for such a UV-layout. I think we can all agree on that.
When you're working on 1024 and 2048 textures, there's a lot more advantage to having a clean and smooth uv map over trying to take advantage of every single pixel possible. Plus, like I touched on earlier, the more time you spend UV-editing, the less is left to spend texturing.
[/ QUOTE ]
I've worked on next gen characters. I only provided current gen examples, but I use the same gridded UV process for all my assets, and it's never been a problem, no matter the taper. Speed is all well and good, but the asset will be done forever, and I'd rather spend a bit more at the start to get efficient UVs that then result in more texels on the model and hence, more details (which is what consumers pay for in next gen, right? more details) is worth some extra time. Plus at 2048, slight stretching is much less noticeable than on a smaller 256.
I do wish Soul could provide an updated link for his chick. She was at least 5k, I think closer to 15k, and had tons of curves, and was super gridded in the UVs, and it was unnoticeable. There are cases where it doesn't work, but that's why we all have brains, to decide when it's applicable and when not. We're talking about generalities in this thread.
Also, i have to disagree with the stretching comment, anything that has normals is going to be much more apparent that its stretching, and anything with a large texture map that has fine details will show badly when there is stretching as well, moreso than what would show up with just a lowres diffuse map.
I tend to do mostly hard surface stuff over the organic shapes and it is much easier to stick to the grid method but for some of the organic stuff I have been working on lately has made me rethink things. Our client is really picky on uvs and doesn't want any space to be wasted which is understandable but also with the engine things can't just be cut and rotated and shit to fix the maps and be more efficient so some seams are created because of this.
I just think unwrapping sucks in general. It is most definitely my least favorite part of the process and to be honest that is probably why half of my personal models have never been textured. Oh well. Let the fight continue.
Plus, like I touched on earlier, the more time you spend UV-editing, the less is left to spend texturing.
[/ QUOTE ]
Agreed. The auto unwrap and packing tools while NOT a perfect replacement for a human who knows what they are doing, it is a good place to start and can do the lions share of the work for you. I personally think there isn't standard unwrap method that should be applied to all characters but a large toolbox of tricks that allows us to unwrap characters and objects in the best way possible. What works fastest and best for the model at hand is what should be used.
Because of this thread I have paid a tiny bit more attention to laying UV pieces (that benefit from it) on a grid and the major pay off for me is that the texture artists understand it easier, that alone saves some time, but also costs me a little bit more time UV editing. In the end griding has its benefits, but I don't think griding every last piece is the solution to everything.
I don't really like doing mind you, but its a necessary evil ^^;
This?
Alex
thnx for the good read
On the recent domwar2 entry I worked on, I pretty much used basic full relaxed segments with as few seams as possible - there's probably more "wasted space" than if I'd spent longer gridding them out and making perfectly straight lines, but it was much faster to lay out and paint, less time worrying about seams.
I would argue that Per's points only really apply to current and next-gen characters though, when you only have a 256 to work with for a full character it's probably better to spend a little longer lining things up, because things like the seams down the side of trousers or zips etc. do get a lot more blurry at that resolution if the uv's aren't straight.
meh, whatever, to each their own. There's no true right and wrong with this shit. If the asset looks good (above) and the artist got it done in a reasonable amount of time, who fucking cares?
It seems to me that the precise, cubic nature to the look of the above textures, is as much a *style* choice as a technical one.
Model for Uvs, Uv for design. MGS and FF model win!
Oh and, thanks to Per and Daz for the clear prose.
While i very much enjoy your wonderfully writen post, you still fail to disprove Ben Mathis' main point; "I've done this before and on NEXT GEN assets, so it must be correct." You sir, must have never actually done this, or perhaps you just have poor visualization skills.
I've noticed people mentioning the FF and MGS texture flats. Is there somewhere they're available for viewing? (I no longer have my MGS2&3, nor FFX DVDs available)
You kids and your luxurious high res textures. Most of you will never know what it's like to squeeze everything out of a 128x128 texture. Haha
"If you can't make a 128x128 look good, what are you supposed to do with a 1024x1024?"
Daz, first you express frustration regarding over-emphasis of how gridmapping is best applied to low-rez art, and now when I leave that point alone (I had already touched on it in post post#200260), you complain that I didn't bring it up. Too much blind hatin', man, time to bring out the chill.
You ask who cares, well I care. I'm working freelance, man. While suckers are slaving behind their desk all day, I'm out in the sun with my stomach full of pancake. How come? Because I'm mad fast at this shit, I'm a motherfucking speed-train. I'm on flaming monorail sonic booming all over the place leaving a trail of high quality models behind, working for some of the most prestigeous clients out there, I'm the black superman, the sleazy punk of 3d funk, the polygon prince. I almost make money faster than I can spend it, and I get to spend off-computer time more than aplenty. And I'm FAST because I'm completely anal about every single aspect of game art production. I analyze, I test, I benchmark, I time myself, I use keyloggers to build statistics over my most frequently used shortcuts and remap them ergonomically. I research every single 3D software package I can get my hands on. I write scripts. So of course you can figure it's no skin off my back if someone else lames out and decides to stick to arcane bits of workflow for whatever reason. I've said my bit, and a bunch of guys could benefit from listening. If they do, that's cool and I like to bring stuff back to the community that originally taught me not to be a noob, but if someone spazzes out, that's their loss.
So, yes, *I* care.
Hey poop, did you say that stuff eq sez you did? Get ready for the pain-train, brother - I'm merciless!
[/ QUOTE ]
erm, ok. What you said. Looks like you lost some o' dem dollars writing that out though man!
But then again, that's PS2.
The summation of my post is that I agree with Per128.
erm, ok. What you said. Looks like you lost some o' dem dollars writing that out though man!
[/ QUOTE ]
no,following his logic it would mean, his workflow actually gave him the opportunity for time to do waterfall-esque replies.
as for the looks, obviously the best way would be "screenspace" mapping, givin the best pixel sharpness in return. While this doesnt work, whatever comes closest wins. So if your zipper goes vertical and you see humans mostly standing upright, that would yield a good result. (rotating by 90° wouldnt hurt). About wasting texture mem, projecting a model to 2d (such as sprite art) of course means lots of "empty" space around, but is that really wasted detail? I'd say sprites are probably the "best detailed textures ever". If the area was stretched just to get more texels in the texture, it wouldnt change the amount of pixels on the screen. Consistency really wins over everything I'd say. When you stretch uvs or rescale, whatever, you change local detail frequencies. Take a "greyscale nose detailmap" as example, if you stretch it, it looses all its sharpness. You are essentially doing the same thing here. You must paint "blurry" so that the "unstretch" fixes it back. Which means you loose precision twice...
anyway I dont think you can beat the maths perfect 2D parametrization of a shape, which minimizes stretching error, when it comes to texel/pixel relationship. Getting those charts to lineup, and packing them tightly is were the human would beat the algorithms yet. You must always keep in mind that everything is purely maths here, no room for interpretation, as the maths implemented is mostly reduced to the simplest form.
so much about watefalls
2nd beef: No offense to the model because it looks good but its a higher poly version of a low poly box/cylinder model so of course it will grid easy. More loops doesn't mean more detail unless you make them count. If you have to compromise the silhouette of the model for a gridded UV layout, which was suggested in other posts as "if you model with a grid in mind" then whats the point?
I would be really impressed by the 100% grid method if I could see it fully worked out on a non cylinder/box model. If it really is the end all be all to UV editing, saves time and heartache I would like to see it fully tested not just paraded around with very little proof.
Until then I'll stick to the hybrid method of doing what works and looks best on the model at hand. Back to reading a few monster posts, sorry if this turned into one =P
EDIT: Oh hey look at that I managed to repeat Per and the die hard gridders aren't so die hard any more, looks like I just posted a useless post in a concluded thread =P
UVs are fun!
I use keyloggers to build statistics over my most frequently used shortcuts and remap them ergonomically.
[/ QUOTE ]
That, my friends, is HARD CORE.
Awesome thread.