Consistent Texel density vs. Importance/Space efficiency?
greentooth
Based on this article: http://www.ericchadwick.com/examples/provost/byf2.html
1. Is it acceptable practice to have inconsistent texel density in your maps in order to prioritize detail in important places like say the face of a character?
2. Is it acceptable practice to simply scale up your UVs as much as possible to maximize UV space provided there isn't too much visual inconsistency. Like say as much as 50-100% more texel density in certain areas?
3. Is the "keep consistent texel density" advice still important for performance or is it just important for visual consistency now? I ask because the article was written ten years ago and in its explanation suggests that most PC-based systems and current (2003) high end consoles will automatically ensure a good texel cache coherency.
1. Is it acceptable practice to have inconsistent texel density in your maps in order to prioritize detail in important places like say the face of a character?
2. Is it acceptable practice to simply scale up your UVs as much as possible to maximize UV space provided there isn't too much visual inconsistency. Like say as much as 50-100% more texel density in certain areas?
3. Is the "keep consistent texel density" advice still important for performance or is it just important for visual consistency now? I ask because the article was written ten years ago and in its explanation suggests that most PC-based systems and current (2003) high end consoles will automatically ensure a good texel cache coherency.
Replies
2. I'd try to keep it as consistent as possible unless you are loosing too much detail in certain areas, like where there is text or a lot of visual noise. inconsistency is easily noticeable by non-artists and can look worse than low resolution textures.
3. It's mostly a visual consistency thing.
As noted, in the case of ME, or any other game that has characters faces in the front all the time, either due to dialogue or story reasons will usually get a bigger space, or even it's own separate unwrap material and unwrap.
Same can be applied to customizable characters.
Usually you will try your best to get them at the same ratio, the only problem is, if you try and squeeze every other pixel in, where different parts have different ratios, it can very easily degenerate to playing Tetris with your unwrap, and as you noted, you're getting back at most, what? 10% more fidelity on a certain part of the texture? Hardly worth it.
It's usually better to leave that extra wiggle room for your padding as those will serve you better in the long run, especially if your game needs to run on older hardware and has mipping issues.
Lastly, as noted again, it's a mostly a visual reason, if the asset is passing from artist to artist, and have iterative stages for different parts of it (armor, morph tagets, etc), having a unified and somewhat optimized system in place will do wonders.
In some situations such as first person guns can have more UV space dedicated to the back. Though this is done mainly to save texture memory and the fact they are generally viewed from a certain angle.
Faces are often higher to allow for close ups.
You also need to understand how mip mapping works, you cannot pack more detail into a smaller space if the texel density is higher than the pixel density on screen it will pick a lower mipmap. It cannot be stressed enough that you can actually make your art look worse if you use extremely inconsistent UV space. I've spent a lot of time trying to explain this to people but its amazing how many people just don't get it.
The most important thing though that a checkerboard texture applied to your model has square checks.
Faces, labels and areas of models that face the camera a lot are all suitable candidates for this. Take a look at ID Software's "RAGE" for example. The TD fluctuations are very, very big there - but it actually works and the game looks great imo.
2) I would say so yes. I doubt that any art director or lead artist is going to be so anal about TD that they are going to ask you to land on exactly X pixels per Y in game units. Usually you have a target TD but are allowed to go +/- a few % in difference.
But that also depends a lot on what kind of graphics you are making. Take toony graphics/animation for example: When working with that kind of graphics it's easier - and smarter - to take shortcuts: stretching and pinching shells as you see fit because in the end it is not as noticable as it is on high-end graphic assets.
3) Both. Visual consistency is very important ofc, but it's also a performance issue - especially when you work towards a platform such as mobile devices, Xbox or PS3 that have totally different limits compared to a PC. I work with mobile games and our biggest restriction is exactly that: texture map resolution. We can't have too large or too many texture maps because we hit the limit fairly easy - so we have to be pickier about TD and smarter about optimization I believe.
For something like a console for example, giving the same texel density to the back of the console as the front would be inefficient, so you might scale the back uvs to be 50%, or you might use a tiling metal texture for the back instead of including it in the unique unwrap. But on the other hand, if you have a big statue, you probably wouldn't want to have the top use less texel density even if it makes sense for the shot, because in the next room the artist might throw in a broken version where the top is lying on the floor. Or the intro cinematic might zoom in on the face, etc.
In my own workflow, I'll generally scale uv's off by up to about 20% to get the most pixel and time efficient layout. I save that 50-100% scale offset for unimportant stuff that will be hard to see or notice.
Besides the visual fidelity issue, if your texel density is too inconsistent, it makes it a nightmare to texture. In photoshop you might paint in all these scratches with the same pixel brush, and suddenly in game your scratches are all different sizes and look weird.
For something like a FPV weapon, I will scale up the areas that your closest to by about 115% or so, and the iron-sight/zoom in type areas another 125% to make sure those are super crisp, and them maybe scale down the areas that will rarely/never be seen from first person perspective to 75-85% range.
In particular check his shirt, the pocket flaps are at least double the texel density, maybe more, compared with the rest of the shirt. He also has a texture strip at the bottom for the shirt buttons that is much higher resolution than the rest of the shirt. I am guessing this is because he wanted more space because the buttoned areas had more details, smaller wrinkles, stitching, etc.
Works fine for a portfolio piece where you can zoom in on all the details in the belt and stuff but Would you guys consider that a poor unwrap for a real in-game model? Or does it work in this case because the details are where they need to be and not too distracting (like the mass effect screenshot)
Important in what way? For mip selection/performance? Or just for visuals?
In World of Warcraft they reuse textures aggressively without regard for stretching or maintaining square checkers (doesn't look so great sometimes), especially in the older models which would have had to run on older hardware.
Guess I'll just try to use my best judgment and try to avoid things getting too stretched out then