coping with LOD and UV maps in blender

[Fizz Savira]

I've been trying to come up with a good workflow, in blender, for making well behaved meshes (in this particular case, shoes).

I've tried a couple of different approaches:

1. Make a low poly model, use that for the lowest LOD, then improve it to make (eventually) the highest LOD. This one kinda sorta works, except that I've run into an issue with the materials - because the lowest LOD doesn't have faces that reference some of the materials on the higest resolution LOD, the importer complains.

2. Make a high poly model, then reduce it until it's low poly. The problem with this one is that I end up with holes in the UV map where I've removed edge-loops. The holes lead to crazy texturing behavior.

I'd love some thoughts on how to cope! Right now, every which way I look amounts to a *huge* amount of work.

[Medhue Simoni]

I don't really use blender, but I'm not totally sure why you are that worried about LODs. I'm pretty sure that rigged meshes don't switch LODs like they normally do. Personally, I'd just make the shoes and let the uploader generate the LODs. You have some control over how many verts they are, and you are definitely not going to have a material problem.

[Kwakkelde Kwak]

---

Medhue Simoni wrote:

I don't really use blender, but I'm not totally sure why you are that worried about LODs. I'm pretty sure that rigged meshes, or even worn meshes don't switch LODs like they normally do. Personally, I'd just make the shoes and let the uploader generate the LODs. You have some control over how many verts they are, and you are definitely not going to have a material problem.

---

Hmmm, a strange answer from someone who is so keen on the efficiency aspects of mesh. Rigged mesh most definately switches LOD. In fact it's the worn items where mesh can either excel or fail miserably. Custom LODs tend to give me better numbers by more or less a factor 2 over the autoLoD, with prettier results ofcourse.

I've passed on buying some fantastic looking hair because the draw weight was too high. That number seems to vary between viewers, but when rezzed that hair had a landimpact of 170 or so. Maybe that's a lot better than most sculpted hair, but after looking around I found some hair worth about 15 prims, looking and behaving just as well. Even this hair had plenty of room for improvement with a fully modelled inside, hidden by the skull.

AutoLoD on a shoe will probably work better than on hair, but that doesn't change the principle. The material issue is easily solved by the autoLoD, but it's also easily solved by adding a single triangle with the high LoD material to the lower LoD models.

[Kwakkelde Kwak]

My workflow is different for different sizes of meshes, because the LoD models vary.

I often build LoD med by hand, then autogenerate high. Then for the lower LoDs I make new models again, they are very simple in most cases. Often I build all four LoDs, or the highest 3 and skip lowest. Again it really depends on the subject and the size.

The texture issue you can tackle by adding a single triangle with the higher LoD material to the lower LoD model and hide it somewhere. Not working with Blender, but I bet someone can help out if you can't get a material match between the different models. In 3ds max there's an option to reduce the vertices and keep the UV maps, I don't think Blender has this function, but who knows.

[Medhue Simoni]

Yeah, just worn meshes have normal LOD switches, but I never see an LOD switch on my rigged avatar. Maybe that has to do with the size tho.

[Maeve Balfour]

Yay, a fellow shoe/bootmaker actually interested in responsible LODs! :matte-motes-smile:

Like you, Fizz, I am still exploring workflows for effective footwear creation (and hopefully reasonably time efficient). I tend to be distracted easily though, so I seem to be taking forever... But I guess since I am just creating things for my own usage, it's okay to meander along LOL.

Anyways... some generic modeling thoughts (I don't use Blender specifically, but these should work regardless)....

In my efforts thus far, I have tended to make my full LOD mesh first (with as much efficiency in its construction as possible) and then worked backwards on the lower LODs. I leave the full mesh for LOD1 and LOD2, then create lower detail meshes for LOD3 and LOD4 (3 separate meshes in all). I then remap the UVs for the LOD3 and LOD4 meshes to try and match the original mesh's texture - not an easy feat (feet! LOL) as you can well relate to.

Another idea I am considering, but haven't gotten around to trying yet... is to reserve TWO of the materials exclusively for the LOD3 and LOD4 meshes (one per mesh). My concept behind it is to use two totally SEPARATE UVs/textures for those meshes - to hopefully avoid the UV-matching headaches you mention. The two exclusive materials can hold their own specific texture, which would be designed to fit the UVs of the respective LOD3 and LOD4 meshes. The main mesh could have the other six materials out of the eight available, or however many I decided it needs.

Now - a little bit of trickery is needed to make this work. EACH MESH needs to have ALL the materials, regardless of whether they are intended to be showing or not. So... for the main mesh, I would create two tiny triangles and hide them from view, and assign the materials intended for LOD3 and LOD4 to them, one material per hidden triangle. FOR THE LOD3 and LOD4 MESHES, I would do the opposite - hide SEVEN triangles, each with the unintended materials assigned (one per triangle), and then for those LOD meshes (the visible sections) I would assign the intended material (and hence the intended lower LOD texture, UV-mapped to suit, would be shown). If done right, as each LOD mesh kicks in, the intended texture will be shown. I would assume ALL the textures would be loaded upon rezzing, even if hidden away, so in theory there wouldn't be any lag involved for the LOD changes.

Obviously, there will be a slight increase in impact due to the extra triangles loaded into the lower LOD meshes (to accommodate the hidden materials), but I think it would be quite acceptable, especially since you are already intent on creating responsible meshes in regards to rendering cost. Also, for the lower LOD textures, I would recommend using lower resolution textures (256x256 most likely), since at the distances the LODs will be switching, you wouldn't be able to see much detail in the texture anyway. For the FULL LOD mesh, use 512x512s.

So yah, this is the concept I intend on testing eventually - technically it is probably as much work as working the other way (re-doing UVs to match existing textures). However, it will probably avoid the major headache you mentioned Fizz - getting the other LOD mesh UV's to effectively fit the higher LOD textures.

I hope this helps - feel free to ask me for clarifications!

Have fun! :matte-motes-smile:

[Chosen Few]

Excellent question, Fizz.  You're definitely on the right track with your two approaches.  (Which one to use will vary, depending on the particular model.)  You just seem to be overlooking the fact that you may need to manually adjust your UV layout for each LOD. 

If having to (partially or fully) remap the UV's each time is a little more work than you'd anticipated, all I can say is this is just the nature of polygonal modeling.  General rule of thumb: Whatever amount of extra work you did to create each new model for the LOD's, you can expect to do an analogous amount of work for each model's UV layout.  You can minimize this to a certain degree, with experience, as you gain more of a sense of best practices.  But there will always be SOME work required each time.

A lot of people tend to think of UV'ing as this annoying thing you have to do after the fun part, the modeling, is over.  I'd encourage you not to fall into that trap, yourself.  UV'ing is an inherent part of the modeling process, and if you do it right, it's just as enjoyable as any other part.  If you like to draw, you should like to UV.  (And if you don't like to draw, you probably shouldn't be 3D modeling in the first place, since it's likely that very little, if any, of it will be enjoyable for you.)

Certain modeling programs are equipped better than others for intelligent UV management while you work.  I'm not fluent enough with Blender to know what options it might have in this regard.

 

Let me respond to each of your specific questions.

---

Fizz Savira wrote:

because the lowest LOD doesn't have faces that reference some of the materials on the higest resolution LOD, the importer complains.

---

All you need is at least one triangle for each material.  If your low LOD model doesn't need one of the materials that is on your high LOD version, just bury an extra triangle somewhere inside the model where it won't be seen, and assign the material to it.

You can also use this technique to assign entirely different textures to each LOD.  Put a hidden triangle inside the high LOD model, and assign the low LOD model's material to it.  Likewise, put a hidden triangle (or several) inside the low LOD version, and give that the high LOD model's material(s).  As long as the total number of materials is less than eight, this will work.  It can be very effective; just be careful not to overdo it, since it obviously can increase the texture load and material load considerably.

 

---

Fizz Savira wrote:

I end up with holes in the UV map where I've removed edge-loops.

---

That's OK.  Just adjust the map to fill in the holes.

One way to minimize this problem, is to be careful not to remove border edges when culling your edge loops.  If all the edges you remove are between borders, you shouldn't end up with any holes at all.  As I mentioned above, once you get more experience, you develop a sense of what to do, and what not to do, throughout the entire process, so you ned up with UV's that won't get borked when you step the LOD down in this manner.

Also, don't forget that you can utilize different parts of the texture canvas for separate UV'ing of each LOD model.  In this way you, can UV each LOD entirely differently, if need be.  There's nothing that says they all have to find inside the same framework.

 

---

Fizz Savira wrote:

I'd love some thoughts on how to cope! Right now, every which way I look amounts to a *huge* amount of work.

---

If you're looking to avoid huge amounts of work, then don't be a mesh modeler.  This stuff takes time.

The same is true of any art form.  Is there any way to create an oil painting without a huge amount of work?   Absolutely not.  Any painter will tell you it often takes months to do it right.   How about illustration?  The average person probably has no idea how much work goes into a comic book every month.  It's enormous.  Sculpture?  Don't even get me started.  Depending on the material, and the subject, a good sculpture can take anywhere from days to decades to complete.

Artwork, true to its name is both art and work.  The two are inseparable.  You can't get the former without the latter. 

A lot of people seem to be under the impression that because digital artwork is done on a computer, the human effort factor is somehow removed, as if there's some sort of magic "make it look good" button somewhere.  In reality, that's just not how it works.  The computer is a tool like any other.  Getting good results takes a human being who's willing to put in the requisite effort.  If that's you, then you'll do well.  If it's not, then I'd suggest finding a different hobby.

What I can tell you is that as you do this more and more, the things that now seem like a lot of work will just become routine.  You do them without even thinking about them.

[Gaia Clary]

Hi.

Some remarks from a blender adicted

 

1. About Blender: With Blender B-meshes we probably will see a lot of improvement with UV-mapping. I have been told that keeping the UV map intact while decimating and removing edge loops etc. would be much better supported then. B-meshes looks like it will become available with Blender 2.63. At least that's the rumour i heard about it. BTW: i think there is no need to panic because of need to learn yet another new thing. As far as i know B-meshes will just replace the current mesh-system and add n-gons to blender. In the background it will add a much cleaner model for handlung meshes (and UV maps).
   
   
2. About the SL-Importer: I thought that the SL-Importer has been improved such that different levels of detail only need a subset of the used materials from higher levels. I just tested it and indeed you still need the same number of materials on each level (at least in Second Life 3.2.6 (247329). So is this a bug, a feature, or not yet implemented ?
   
   
3. About texture optimizations: I think adding textures (even low resolution textures) for LOD1/LOD0 only makes sense when the models get completely different (e.g. when you use a billboard for LOD0). Otherwise the server can happily reduce the resolution of a high resolution texture with degrading LOD and only send a low resolution version to the viewer when you are initially far away from the object. So i think the impact of adding low level textures for lower LODS may be higher than just letting the server/viewer do the job. I may be wrong here though, just guessing

[Chosen Few]

---

Medhue Simoni wrote:

I'm pretty sure that rigged meshes, or even worn meshes don't switch LODs like they normally do.

---

LOD's absolutely DO switch on rigged items, just as they do on every other item in SL.  However, there's currently a deficiency in the system, whereby if you attach a very large rigged mesh, it gets treated as if it's still its full size, even though it shrinks when worn, to fit the avatar's skeleton.  This leads to higher LOD's being visible from a much greater distances than they're supposed to be, which in turn means lower FPS for everyone within viewing distance!  Not cool.

Needless to say, there are those in SL who love loopholes wherever they can be found, and those types probably think this particular one is great, silly as that is.  Whether it's because they're too lazy to create their own LOD's properly, or because they're just so selfish that they don't care what becomes of anyone else's frame rate as long as their precious avatar looks as high-poly as possible from every conceivable distance, or simply because they just don't know any better, it's a big problem.

To those people, I simply can't stress enough that LOD's exist for good reason.  As I've said countless times on this forum, the system doesn't run on magic.  Computers have limits to what they can do at speed.   The more polygons in a scene, the slower the scene renders, period. 

When viewing any avatar from a distance, there simply aren't enough pixels on anyone's screen for the high LOD to look like itself anyway, so there's zero benefit in forcing all those polygons into the renderer in the first place. Therefore, reduction over distance is the only strategy that makes sense.

 

---

Medhue Simoni wrote:

Personally, I'd just make the shoes and let the uploader generate the LODs.

---

I'd never let the system do it.  Make your own LOD's, always.

The system is only capable of dumb reduction.  It's going to remove polygons strictly by mathematics, without any visual sensibility whatsoever.  There's no way for it to know what looks good and what doesn't. 

In nearly all cases, getting proper results requires smart reduction, done by a human being.

[Medhue Simoni]

Well, my rigged avatar doesn't change at all. His jeans don't change, nor do any of the other rigged attachments. The mesh jacket, that I purchased, doesn't change LODs at all, either.

Oh, and I tried 2 other mesh avatars that were given to me by other avatar creators and neither of those change LODs either.

[Fizz Savira]

Great responses, everyone

The hidden triangle trick sounds super useful, I'll give that a whirl.

And it's very encouraging to hear that blender has some tech coming that will make UV management simpler. Right now, if I remove an edge-loop (not a border one), I have to manually select pairs of UV verticies and weld them together again to get the right texturing behavior. My hope was that modifying the UV map would be easier than doing a seperate unwrap and trying to get the two or more of them coherent so that the textures don't fly around all over the place, but right now it's very tedious.

Regarding the comment by Chosen about "it's just alot of work" - well, of course it is The number of hours I pour into my sculpty designs is large, but for that I already have a workflow sorted out, which does involve LOD management, but not UV management. The whole UV thing is new, and so trying to add that in, without going down paths that are just a waste of time, is what this thread is about

 

[Drongle McMahon]

Good Old Blender 2.49:

Going from high to lower LOD

(a) do it manually and suffer the burden of joining up all the islands left by eliminating faces. The main thing is to remember to check that superimposed vertices are welded (move it and see if they move as one, then escape or undo). This often happens automatically. If it doesn't, you can weld explicitly (W). If you don't, the uploader will see two different vertices and that will increase your download weight. I make an image of the high LOD UV map (or use it's baked ao) to make sure it's aligned properly. At least some of the uv vertices stay in the right place. You can get horrible problems if you haven't borne the reduction in mind while placing seams!

(b) use decimate modifier (Arrggggh). This not only makes a horrible mess of the mesh, with unresolvable triangulations of quads, but it loses the UV map entirely!

Going from low to high LOD:

Much easier if you can learn to get the low LOD right for the hogh LOD you want (I never can). Then you can use either multires or subsurf modifier to make (initial) higher LODs. Both subdivide the UV map along with the mesh. With subsurf, you can tell it to subsurf the UV map, but I don't think the effect is generally useful. Otherwise, and always for multires, it just leaves the original UV vertices in place. This sounds very nice, but as Chosen said, you will still have to move things around to get the optimal mapping. Furthermore, the subdivisions are only suitable/optimal for certain rather smooth shapes. For most things I make, I still have to do a lot of editing and still have quite a lot of filling in gaps to do. Still it's easier than starting at the high LOD.

So recently I have been comprimising: Make the high LOD and UV map it; export a picture or ao image; Do all the loop removal etc to get the low lod (lowest is usually too simple); UV map that and adjust to the saved image; then do a multires to get the medium ... getting the subdivided UV map with minimum adjustment needed. This doesn't work for everything, but saves time when it does.

The best way to save time with UV maps is to recognise all the symmetry and repetition in your model before you start. Then just make and UV map the repeated components for all LODs. . Then duplicate as appropriate (array, spin dup, shft D, mirror, etc.) and eliminate double vertices where necessary. Then you can make adjustments for small differences between repeats or others required for effective LOD. There is little more frustrating than doing the same tedious UV map edits on a dozen essentially identical pillars. Sometimes you can detach them and stack them up and edit them all at once (good for reducing texture area consumption too if you leave it that way), but it's still much more work than doing it once at the right time.

@Gaia...

"1. With Blender B-meshes we probably will see a lot of improvement with UV-mapping. "
Hmm. That might get even me to move on from good old 2.49.

"2. So is this a bug, a feature, or not yet implemented ? "
See VWR-27873  https://jira.secondlife.com/browse/VWR-27873

[Fizz Savira]

I just tried a few experiments:

First, I went from high LOD to lowest LOD by shredding my mesh. I added back in some hidden triangles to capture the missing materials, and that satisfied the uploader *YAY*

Next, I stumbled upon exporting the UV map as a PNG and did that (an amusing coincidence since I then read the reply by Drongle ...), then used that as a guide to unwrapping and moving the lowest LOD UV's. That works OK. Unfortunately, there are some parts of the lowest LOD mesh that don't map nicely on the high LOD map *pout*

It seems that I need to account for this during the design of the high LOD UV map...

However, at least I'm getting somewhere finally!

Thanks again for the sage advice everyone

 

[Kwakkelde Kwak]

---

Medhue Simoni wrote:

Well, my rigged avatar doesn't change at all. His jeans don't change, nor do any of the other rigged attachments. The mesh jacket, that I purchased, doesn't change LODs at all, either.

Oh, and I tried 2 other mesh avatars that were given to me by other avatar creators and neither of those change LODs either.

---

Did they not change LoD or did you not see them change? Those two are not the same... What's your object detail? What's your RenderVolumeLODFactor?... Dial them down and you will probably see the change. If you don't, I bet the render/draw weight of those items is far too high because it can only mean all LoD models are the same.

Anyway, the hair I'm wearing changes, my shirt does and my pants do... all hardly visible unless you really try. All have a rezzed landimpact of 10 prims or so. I think their display weight is in the hundreds, hair might be a bit more.

[Maeve Balfour]

Chosen Few wrote:

Needless to say, there are those in SL who love loopholes wherever they can be found, and those types probably think this particular one is great, silly as that is.  Whether it's because they're too lazy to create their own LOD's properly, or because they're just so selfish that they don't care what becomes of anyone else's frame rate as long as their precious avatar looks as high-poly as possible from every conceivable distance, or simply because they just don't know any better, it's a big problem.

To those people, I simply can't stress enough that LOD's exist for good reason.  As I've said countless times on this forum, the system doesn't run on magic.  Computers have limits to what they can do at speed.   The more polygons in a scene, the slower the scene renders, period.

___________________________________________________________________________________________

Sadly, I have seen FAR too many examples of this with mesh already inworld and on the Marketplace. As you say, it appears people are too lazy to bother with LODs, or are simply just too selfish about their AV appearance to care. Ignorance to rendering cost in general is another factor - creators and users just don't seem to want to know, or learn.

So Fizz, you have my genuine applause in your efforts in keeping your render impact down. You can be rightly proud of that. I can't name and shame examples in here due to TOS (plus I don't like being rude to individuals)... but... I have seen some hideous examples of footwear with MONSTROUS triangle counts and no efforts to create any kind of LOD that I can discern. Seriously, I know of examples where one SINGLE shoe (from a pair) I have seen on the MP would exceed the prim allowance on an empty 512m2 plot if rezzed on the ground - and would be hard pressed to fit in an empty 1024m2 plot (234 prim allowance). With heavily sculpted footwear, this has become the accepted norm (unfortunately), BUT with mesh, its utterly inexcusable. Just seriously sloppy meshwork, which would be excessive in general 3D rendering, let alone a realtime 3D environment.

Sorry for the mini rant, things like this just makes my blood boil, LOL.

So yah, Fizz, be proud - You are doing the right thing, definitely! :matte-motes-smile:

[Kwakkelde Kwak]

---

Fizz Savira wrote:

 Unfortunately, there are some parts of the lowest LOD mesh that don't map nicely on the high LOD map *pout*

 

---

Did you look at the lower LoD model in Blender or inworld? Chances are you won't see the "misfit" at all inworld, since the object is so small....

[Kwakkelde Kwak]

Maybe it's time to start a thinking about a jira with a request for a very hard cap on draw weight for a single attachment. Crappy build? impossible to attach..now that sounds fair to me.

40 000 is acceptable according to LL, I'd say 10k per attachment or something is reasonable.... or maybe just an overall cap, go over 100k and get a message when trying to wear something new...something like that....

[Fizz Savira]

---

Kwakkelde Kwak wrote:

---

Fizz Savira wrote:

 Unfortunately, there are some parts of the lowest LOD mesh that don't map nicely on the high LOD map *pout*

 

---

Did you look at the lower LoD model in Blender or inworld? Chances are you won't see the "misfit" at all inworld, since the object is so small....

---

I did look at the model in world, and for my first attempt at this, there was some visible jumping around. The second version, after I made some changes to how the high LOD unwrapped, looked alot better, so I'm now feeling that this problem is more solvable than I had first thought, with less work than I had thought.

 

[Medhue Simoni]

---

Kwakkelde Kwak wrote:

---

Did they not change LoD or did you not

see

them change? Those two are not the same... What's your object detail? What's your RenderVolumeLODFactor?... Dial them down and you will probably see the change. If you don't, I bet the render/draw weight of those items is far too high because it can only mean all LoD models are the same.

---

I don't screw around with the default LOD settings. All the draw weights are within reason. My avatar, with clothing, is about 15000 draw weight. I made the avatar so I know exactly what the LODs are. The only way I got any of the rigged meshes to change LODs, was when I turned Object all the way down in the Graphics Preferences. The slider had to be at 0 for the avatar to change, and it changed to the low LOD. Beyond that, even at a distance of 120 meter, no change at all. My LODs, for rigged items, are always the med LOD is the same as the high, the low LOD at around 25% and almost nothing for the Lowest. Even if you want to say it is my mesh and how I do it, this doesn't explain why I get the same thing from rigged mesh that I did not make.

[Fizz Savira]

---

Maeve Balfour wrote:

*snip*

Sorry for the mini rant, things like this just makes my blood boil, LOL.

So yah, Fizz, be proud - You are doing the right thing, definitely! :matte-motes-smile:

---

 

I agree, those things make me upset as well. I suspect we will see plenty of that given the difficulties involved

Well, I'm trying to do the right thing, that's for sure. Sadly, I have no metrics to work against except my sculpted products. I've pushed those to be efficient as well, so I think that if I can come in face-count wise, less than those, I'm good, right? right? LOL. Seriously, some metrics to design against would help. The new informational display showing the breakdown of an object, including prim costs and rendering costs is super helpful, but I'm still working in a bit of an information vacuum sadly.

For example, a recent design of mine (a test really) uses 5 sculpties to construct. That's over 5000 quads? at highest LOD. Now that 5000 includes all the low LODs as well, so comparing against a mesh requires some care. I made a mesh version of that same design, ripping out all the hidden parts and eliminating lots of over kill in the design, and that comes in at 683 quads at the highest LOD. I made a low LOD version of that, which came in at approximately 50 quads (a few triangles snuck in to solve materials issues). So assuming I'm lazy, that total works out to 683*3 + 50 or 2099 quads which is well under half compared to the sculpted equivalent.

I think that's a win, if my math is right? But I still don't have a sense of whether or not that's good enough...

 

[Medhue Simoni]

This is just my opinion, but you shouldn't use anything as a reference. You should use the amount that the items requires. Using a sculpty as a reference would be a huge mistake, as most of the verts are being wasted. If the item requires 1000 verts, in your opinion, then you use 1000. If the items requires 50 verts, then you use 50. Generally, you want to use as few as you have to, but the last thing you want to do is use any sculpty version as your reference. Heck, sculpties might require 10000 to make the items, where as the same thing in mesh could only require 100 verts.

[Kwakkelde Kwak]

---

Medhue Simoni wrote:

I don't screw around with the default LOD settings.

---

Yes but your default LOD settings might not match the default setting of others:)

---

Even if you want to say it is my mesh and how I do it, this doesn't explain why I get the same thing from rigged mesh that I did not make.

---

You make it sound like I want to push you in a corner and prove you wrong or something... that's not the case, at all. I just noticed LOD changes on my av and dialing down the numbers seems to affect it aswell, in other words, it does happen. It does look like they change rather late compared to their size, possibly the mesh adopts the avatars size, maybe there's a bonus for rigged mesh where a LOD change is skipped, maybe it's a bug. You seem to be right about it not behaving like other objects...worth some testing I'd say.

 

 

[Drongle McMahon]

I believe the avatar and rigged attachments change LOD together. The LOD distances are controlled by renderAvatarLODFactor, not renderVolumeLODFactor. It ranges from 0 to 1 with the Avatar graphics quality slider.

[Kwakkelde Kwak]

It also reacts to the objects detail slider... and I had the feeling the RendervolumeLOD aswell, which would make sense then...I just had a quick peek, as I said, something worth testing..... not at 1am though:)

[Maeve Balfour]

Fizz: In regards to your mesh creations, I would just use a simple rule of thumb - use only as many triangles/polys as you need to, to get the shape you require. I wouldn't even try to compare triangle usage to sculpties - Just keep your models tight and trim in regards to poly usage, and you are already there.

In other words, do as you are already doing - working for efficient usage of triangles as you model. I know you are determined to do that, so you are already well on the way. If you constantly keep an eye out for ways to trim excess polys as you work, then you will quickly develop an eye for lightweight, efficient models.

Being someone who likes modeling footwear meshes myself... to me, your quad count sounds quite reasonable (683 in your post). I guess the poly count would double once converted to triangles... but it would still be quite reasonable. If you can see areas where quads could be trimmed next time, then yah, reduce if you need to... but since you are using LODs in addition to your main mesh, the impact overall should be quite acceptable (especially in comparison to some hideous examples I have seen inworld LOL). You're doing the right thing Fizz - be proud of your efforts!

:matte-motes-smile:

PS: To give an idea of the monstrosities I have seen inworld... although unable to get actual triangle counts, when I viewed said footware meshes in wireframe mode.... I kid you not when I say this.... the triangles in the mesh were SO DENSE / TINY / TIGHTLY PACKED that they had an appearance almost like a solid surface in their own right, when viewed from certain angles. I'm talking many tens of thousands, quite possibly hundreds of thousands of triangles... in a single damned shoe or boot. When I rezzed a demo on the ground... initially I thought it was server lag.. but no... it was SO HEAVY it took what felt like a minute to rez. Not to mention no effort at all to remove hidden faces.... so you can just imagine the wastage and huge render/server impact. Arrrrrgh!

(Edited to fix paragraph formatting (forum borked on me earlier LOL).