Texture Space, Theory and Practice.

Texture Space, Theory and Practice.

Share and discuss visual creations and creation practices like texturing, modelling and musing on the meaning of life.

Moderators: MR.D, Moderators

User avatar
Argh
Posts: 10920
Joined: 21 Feb 2005, 03:38

Texture Space, Theory and Practice.

Post by Argh » 09 Jan 2008, 21:22

Let's talk about what I'm doing, and why. It's not crazy. It's not stupid, either. But it takes a little bit to explain the process.

Here's my 512 map, for my LightMech skin:

Image

Wow, that looks terrible right? Huge, wasted space.

Know what I'd redo, if I went back again, now that I've seen it all in game? I'd move the two hexagons for the sides of the legs (upper left) together, to share one space, and the two "leg vents" (near the middle) together, totally mirroring it all. I'd make the now-single hexagon for the leg vent bigger. Not much, just a bit more, so that I could use my next-larger version of my repeating round-vent texture.

And that's it.

I'd end up with more empty space at the end, too. The other stuff does not need to be bigger, or have more detail- the guys are maybe 32 pixels on a side, at typical zoom. And even at screenshot zoom, they look just fine:

Image

Now, what is the lesson I want to impart here? Simple. Cramming your uvmap as much as possible is not necessarily a virtue.

Detail levels are more important than uv space, in my book. So is having room for mistakes, kitbashing, and other things.

What is a detail level? Simply put, it's the amount of perceived space vs. the actual space. Or, to get technical, how many pixels of texturemap lie within a given set of uv coordinates.

This isn't a straightforward, there-can-only-be-one-rule sort of thing. It's an art. But here are some handy rules of thumb.

The bottom of a tank, or any other large object that doesn't fly, has a detail level of practically zero. Players never see it. No point in giving it a lot of space. People do it all the time, and then act as if they've been virtuous, by using up their entire uvmap efficiently. That's like praising yourself for gluttony, folks. Don't do that. If players can't see it, it should use as little space as possible. Save it for stuff that matters.

Same goes for small modeled details. Does a small tube need to have the same detail level as the top of a tank's turret? No. That'd be silly- players can hardly see it anyhow. Yet I see people do it all the time. Idiotic. Why waste lots of texture space on some pipe, that could just be a gray gradient with soft airbrush, when players will never see any more than that?

Give it less space... and if you're worried about screenshot-worthiness, confuse the issue by using the airbrush to blur things out a bit. That's what the pros do. Smoke and mirrors, people. Smoke and mirrors. We aren't building models for ourselves. We're building them for an audience. And the audience cannot see certain angles, and is usually at a long distance. Take full advantage.

The converse is true. Do you want an area to pop out with detail, and need the space? Give it to that area, at the expense of smaller stuff, even if that makes your use of space inefficient. Base your need on what you think you're going to need to get enough painted detail in. For example, if you go look at the map, the size of the "face" is disproportionate to the rest of the uvmap, to give me extra room to work with. Same for the red eye. Same for the barrel of the twin cannons. Extra detail, for the areas I thought would need it. Just one of those things that you have to try a bit, and fail a lot, before you will figure it out, trust me.

This gets a lot easier to figure out, with practice, when you know how you're planning out your painting well in advance of actually doing it. You can see, at a glance at one of my uvmaps, which ones I struggled with the color scheme the most- the maps are tightest. The really loose ones were ones where I knew exactly what I wanted, and where I'd want detail, before I got serious about laying out the map. Sounds nuts, but it's just how it works, for me.



When I design a uvmap, these are the assumptions I use:

1. Mistakes will happen. You may need to rework part of a map, or a model. Plan for it, leave yourself room for "oops". Then you won't have to start all over again. Sometimes I get all the way through painting, then have to change some physical part of a model, because I screwed up. It's called "being human", and it's just how it goes. I plan for it. Why repaint the whole thing, when you don't need to? Leaving plenty of space means you can often survive this.

2. You'll never have enough space, even with mirroring, for the areas you want the most detail in, without sacrificing heavily on ease of painting. And then you'll slow down. A lot. Because if one part needs a bevel of 43 degrees, and other needs 67, it's just a lot of extra, fussy detail. Why bother, when sacrificing some space can leave you with clean 90s? Think about it.

Moreover, every model is different. Things like tanks or cars invite mirroring, but you may or may not need to use full details on the sides, front and rear. It's better to assume yes, do a quick fill-test, then shrink some areas down, freeing up space for no. 1, above, than the converse, where you might not have room later, and be screwed.

3. As a consequence of assumptions 1 and 2... assume you'll need to cram as much visual information into as small a space as possible, and plan accordingly while building the map. This often means giant space wasted. Oh frickin' well. If we were in the Quake days, I'd be painting a pixel at a time, to get good results in a 128 or 64, but I don't think I'd change my method much, it works pretty well. And we're not in the Quake days- nobody uses that low of a spec any more, unless they're goofing off, or doing a contest. It's not imperative that we grab every possible pixel. Just that we grab enough to get the job done. Go grab a professional map, sometime- they aren't perfect, either.

Heck, why take my word for it? Here's the uvmap for the Land Raider, from Dawn of war. It features no mirroring, even though it would've been smart, among other issues:

Image

That's one of three large textures, folks. From a pro. I make do with 1 512. Am I starting to sound a little less insane now?

Sure, this skin looks good, but it could have been a lot more efficient. This guy didn't even try.

Now, that said, the Marine textures are marvels of efficiency- all of the Marine stuff, for a whole guy, upgrades and all, is in a 512/256. Just wanted to put that on record- when the DoW people really, really wanted to, they were efficient as all get-out. But I'll bet it took them quite awhile, just to get the model and all of the maps sorted out, and some test runs to make sure they had the detail levels right..

They also lack mirroring in places I would have- stuff like the undersides of the rims of the shoulder pads, and other areas. But, meh, they were under deadline. Oh, and it's obvious that they painted at twice rez, then shrank it, then sharpened, then cleaned pixel-by-pixel, by hand. Very hard lines in places. Very soft in others. It's why the skins look pretty grainy in spots in Army Painter, and blurry in others. But my hat's off to those guys, for making it look as good as it does, that close up.

Now, I get a model edited, uvmapped, and painted, in one night, when I'm motivated and know what I want. Making it easier to keep good lines and good bevel lighting, for example, saves a lot of time in the long run. And I group uvmap sections by what object they're in, because that makes it easy to paint by eye very rapidly, keeping levels of color and shade about right. These things all "waste space".

So, we done now? Or do you want me to run a contest, and we can find out how good I am against people all starting with the same model, no uvmap made, and some rules about color choices? I'm game, if you people are... bet you can't meet a 6-hour deadline, though ;)
0 x

User avatar
KDR_11k
Game Developer
Posts: 8293
Joined: 25 Jun 2006, 08:44

Re: Texture Space, Theory and Practice.

Post by KDR_11k » 09 Jan 2008, 21:49

Argh, teaching people bad practice again? Leaving room on your UV map means that it'll eat system resources that it doesn't really need. Considering how much you complained about if statements eating your performance I'd expect you to make damn sure that your texture is no larger than it needs to be. If you organized that better you could cut the texture size in half without any rescaling.

Sure, undersides should use less UV space but the point of that is that you need the UV space for something else, not that you just leave it blank! If you aren't going to use that UV space anyway you can just as well make parts larger than they need to be, you're wasting texture memory anyway.
0 x

User avatar
Argh
Posts: 10920
Joined: 21 Feb 2005, 03:38

Re: Texture Space, Theory and Practice.

Post by Argh » 09 Jan 2008, 21:52

:roll: I did not say "tight is bad". I said that "tight is not ultimately important". If you're working with a 256 skin, that's your limitation- the size of the uvs within that square are the real issue, and how you size them matters a lot. KDR, once you've made that decision, it's made- you're going to be using that much texture memory. After that, it's about aesthetics. You want ultra-efficiency, you need to start with a small skin size, period. You can't go the other way around.

Again, you guys want to have a contest? I'll whip something up, we can go right now, deadline of 2AM CST... it'd allow everybody but Warlord to participate. We can do some crazy-small skin size, too, like 128s- I'll still very happily kick your butts :mrgreen:
0 x

User avatar
KDR_11k
Game Developer
Posts: 8293
Joined: 25 Jun 2006, 08:44

Re: Texture Space, Theory and Practice.

Post by KDR_11k » 09 Jan 2008, 22:13

Argh wrote:If you're working with a 256 skin, that's your limitation- the size of the uvs within that square are the real issue, and how you size them matters a lot. KDR, once you've made that decision, it's made- you're going to be using that much texture memory.
Nope. If you find that you can get the resolution down without much loss you go for it. Of course if that happens it means YOU'RE DOING IT WRONG anyway.
Again, you guys want to have a contest? I'll whip something up, we can go right now, deadline of 2AM CST... it'd allow everybody but Warlord to participate. We can do some crazy-small skin size, too, like 128s- I'll still very happily kick your butts :mrgreen:
Go play on Polycount, you'll get the trimming to your ego that you desperately need.
0 x

User avatar
rattle
Damned Developer
Posts: 8278
Joined: 01 Jun 2006, 13:15

Re: Texture Space, Theory and Practice.

Post by rattle » 09 Jan 2008, 22:29

ImageImage
Image
I claim teh prize lulz
0 x

User avatar
Forboding Angel
Evolution RTS Developer
Posts: 14589
Joined: 17 Nov 2005, 02:43

Re: Texture Space, Theory and Practice.

Post by Forboding Angel » 09 Jan 2008, 22:38

Something that I think deserves to be said...

In Evolution I use almost exclusively 512 textures. They aren't always very well mapped, bbut one thing is constant no matter how your UVMAP is done...

The resolution on the uv DOES make a difference. Even if you have wasted space, If you cram all that crap into a 256x256, your texture will be much less detailed.

It's the nature of the beast. Generally do this. Try not to waste space, but if it happens, don't jump off of a mountain over it.
0 x

Saktoth
Zero-K Developer
Posts: 2665
Joined: 28 Nov 2006, 13:22

Re: Texture Space, Theory and Practice.

Post by Saktoth » 09 Jan 2008, 23:04

Mirroring, sure, if you want consistent lighting from multiple angles you cant always use it. Not what i was talking about.

UVmap size. Sure, okay so a 512 might look better. Great, use a 512.

But use up the space. Scale it up so it uses the space. You gave the rhino as an example, and its a great one- of my point. It is tight. Every single object is crammed in, it looks like well-stacked tetris blocks. The only bad parts are the large black pieces, which i assume are parts of the model that are rendered- and he still puts things in them.

The spaces around the texture are cram-packed right up against one another though.

Size up the objects so that they fill the grey space though. Give yourself a bit of room to move, sure, but you dont need that much room.

You can keep doing it as you like Argh it obviously works for you- but you shouldnt be teaching other people bad practice.
0 x

User avatar
Forboding Angel
Evolution RTS Developer
Posts: 14589
Joined: 17 Nov 2005, 02:43

Re: Texture Space, Theory and Practice.

Post by Forboding Angel » 10 Jan 2008, 01:11

Regardless of which way you do it, it will still be taking up the same abount of texture space, so at the end of the day, it's not a damning offense.
0 x

SpikedHelmet
MC: Legacy & Spring 1944 Developer
Posts: 1945
Joined: 21 Sep 2004, 08:25

Re: Texture Space, Theory and Practice.

Post by SpikedHelmet » 10 Jan 2008, 01:26

Regardless of which way you do it, it will still be taking up the same abount of texture space, so at the end of the day, it's not a damning offense.
The point is is that if he were smart he could've fit that texture into 256x256 without any loss of quality of the in-game product, and thus sacrafice only half as much performance loss.

And YOU, Forb, there is absolutely no reason why 99% of the units in Evo can't use a 256x256, if not a 128x128 texture. I'm shocked to hear you use 512x512 for everything, and I seriously almost wet myself when I actually looked at your textures.

And that explains a lot of why the game I had with you and the others last night was a bit low-performance, despite the low-poly models and everything. I realize now it is because each of those tiny low-poly units has a fucking 512x512 skin!

PLEASE for the sake of your own mod and the improvement of gameplay lower them to 256 or 128... and try to arrange the uvmaps more efficiently... the reduction in quality will BARELY be apparent, and will be impossible to detect at the regular ranges the game's played at. I mean seriously, all tanks in S:44 use 512x512 or less.
0 x

User avatar
smoth
Posts: 22298
Joined: 13 Jan 2005, 00:46

Re: Texture Space, Theory and Practice.

Post by smoth » 10 Jan 2008, 01:35

argh is mad because I claimed that his textures have more wasted space then nebraska.

For you non us people, the state of nebraska has SOO MUCH NOTHING that you can drive for 3 hours and see nothing.


as in he wastes space.

as in he does not make the best use of texture space and thus wastes memory.


Image

LOLZ SMOTH IS WINAR!
Last edited by smoth on 10 Jan 2008, 01:38, edited 1 time in total.
0 x

SpikedHelmet
MC: Legacy & Spring 1944 Developer
Posts: 1945
Joined: 21 Sep 2004, 08:25

Re: Texture Space, Theory and Practice.

Post by SpikedHelmet » 10 Jan 2008, 01:44

Sure, if you want

Seriously though here's an example of that tiny little tank. Keep in mind this is probably a closer zoom than most will ever use in a normal game. Also keep in mind this is just a base reduction in size, you could easily improve the quality tenfold by maximizing space efficiency

Image

Can you tell right off the bat which has a 512x512 texx and which has a 128x128? Do you even care?

And I'm saying this to try to recover some of the performance loss I noticed. I'd really like for your mod to succeed (as much as any mod thats not a TA clone and actually introduces something unique) and the performance increase would help.
0 x

User avatar
Forboding Angel
Evolution RTS Developer
Posts: 14589
Joined: 17 Nov 2005, 02:43

Re: Texture Space, Theory and Practice.

Post by Forboding Angel » 10 Jan 2008, 01:53

Yes I can tell, and the 128x128 looks like crap. If you get anywhere within 100 feet of it ingame you can tell how crappy it looks.

I have tested the textures at different sizes... long ago. I like 512x512.
0 x

SpikedHelmet
MC: Legacy & Spring 1944 Developer
Posts: 1945
Joined: 21 Sep 2004, 08:25

Re: Texture Space, Theory and Practice.

Post by SpikedHelmet » 10 Jan 2008, 01:56

Then tell me which is which, Forb. :)

Edit: FYI, I don't really care about the quality of your models or textures; there's nothing wrong with low-poly models for a game meant to have plenty of them. I am just saying that performance is seriously being hijacked by your use of 512x512, and not only that, it's bloating the size of your mod file to 40MB, IMO it should be half that, if not 1/4.

But if you want to get all defensive because you think I'm insulting your mod, then fine. I'm not. If I'm insulting anything (which I don't mean to be) its your uvmapping skills.
0 x

User avatar
Argh
Posts: 10920
Joined: 21 Feb 2005, 03:38

Re: Texture Space, Theory and Practice.

Post by Argh » 10 Jan 2008, 02:03

128s are really hard to work with, as I just re-discovered. 256s aren't bad, for most things, just a bit tricky on detail areas sometimes. 512 is where I'm happiest, because I can be fast and fairly sloppy, 1024 is like having too much food, 2048 is ridiculous- I don't understand how most FPS's use at least one, if not multiple 2048s these days, yet still look pretty crappy up close :P

For me, the performance change I see, with a 7800GT, between 512s and 256s... is very minimal, tbh. Textures aren't a big performance drain. I didn't expect that, but it's what I saw when I looked at raw FPS. 1024s are a bigger hit, but even then... there are several units in PURE with 1024s, and even people with crappy systems have been able to play- it's the explosion stuff that drags FPS down, because it's so bloody CPU intensive. Polycount matters more than the texture space seems to, though, in my tests.
0 x

SpikedHelmet
MC: Legacy & Spring 1944 Developer
Posts: 1945
Joined: 21 Sep 2004, 08:25

Re: Texture Space, Theory and Practice.

Post by SpikedHelmet » 10 Jan 2008, 02:09

It depends on the unit. Are you expected to have 100 of those 1024x1024 units ingame at any one time? Between 4-6 people?

It all adds up when you start multiplying players and units. Not only that, but it bloats download sizes for mod files.
0 x

User avatar
smoth
Posts: 22298
Joined: 13 Jan 2005, 00:46

Re: Texture Space, Theory and Practice.

Post by smoth » 10 Jan 2008, 02:13

Argh wrote:I'm not mad :roll: I just think you're all wrong. I'm allowed to do that, y'know ;)
Argh, you think we are wrong and rather then offer a strong counter argument you make things up. Fact of the matter is that I cram as much stuff into my uv as I can because I want the most detail that I can get. Your texture could have been done on a 256X256 or even a 256X512. You are wasting space.
0 x

User avatar
kiki
Posts: 859
Joined: 05 Nov 2007, 03:06

Re: Texture Space, Theory and Practice.

Post by kiki » 10 Jan 2008, 04:03

really, tho arg. Less wasted space = higher quality AND better fps. Just takes effort and practice. Im not one to talk about that btw.
0 x

Archangel of Death
Posts: 854
Joined: 28 Jan 2005, 18:15

Re: Texture Space, Theory and Practice.

Post by Archangel of Death » 10 Jan 2008, 04:47

(why am I budding my nose in here? Do I I even look like a texture artist to you?! I wish...)
Argh's texture is currently taking up at least 60% of the texture space... which means it wouldn't be possible to get it into a texture of 1/4 the area (simple maths?). Shrinking them down, of course, would make pixelation become apparent sooner as you zoom in closer (is that what some of you are saying as "loss in detail"?), though the graphical elements in the texture would still be there (is that what the rest of you are calling detail?).

I just want to say you have that space there, just fill it. Stretch elements for more visually significant parts of the model even further, even if you aren't going to try to squeeze more greeble into that space. Sure, it won't make a bit of difference once the video card gets to its resizing them for whatever zoom your at unless you get REALLY REALLY close. So why? Cause you can... its not that much more effort is it? And it makes people think you used the space more effectively because it doesn't have as many holes. Win win, aye?
0 x

User avatar
smoth
Posts: 22298
Joined: 13 Jan 2005, 00:46

Re: Texture Space, Theory and Practice.

Post by smoth » 10 Jan 2008, 04:49

it could have fit in a 512X256 or 256X256
0 x

User avatar
smoth
Posts: 22298
Joined: 13 Jan 2005, 00:46

Re: Texture Space, Theory and Practice.

Post by smoth » 10 Jan 2008, 05:03

15minutes in photoshop and I can promise you I could have done better then this.

Frankly it is crazy to me....

Image

I could have done more and there are a few more areas to point out that are redundant but I am in a hurry, Gotta run. Point is that the texture is wasting a lot of space. In fact, I dare say at least 1 more unit could share the space.. in fact, I would use it for a corpse.
0 x

Post Reply

Return to “Art & Modelling”