LODs

[Charlotte Bartlett]

Hi I am hoping some of the experts here can give me some ideas!

I am making new furniture and working really hard on my LODs and efficiency.   

I started to look at other furniture in SL to see what a comparable LI should be to give myself a comparison guide. .  

I was struggling my own item which is a detailed sectional sofa that came out on the high to import once I retopo'd at 3800 Quads.   I can get the Sofa to about 9 LI but still I am having to struggle with the Lowest setting / Imposters as even if I get to blocky cubes it pushes my sofa up to 14 LI.  I am lost on how to get that lowest model to "work" within the parameters of SL without resorting to putting in "0"....  How would you approach the lowest setting for sofa ??  

I have attached a screenshot of a third party item (no names) who's items I love.   What I am struggling with is the LI on their item is low but the complexity is so so high.  Like this is more than my some of my houses!  Also that is just the top of the basket and handles, not the bottom part of the basket or the cushions inside.....


Am I missing something on my LOD quest??

 

[OptimoMaximo]

Take a look at the numbers. The lowestlod is 6 triangles, basically 3 faces. How can it conceivably be enough to reproduce that shape? Use the firestorm Lod switch and see how it was done. To me, that is the result of Lods made via the uploader, and will look like a mess of tris. It is very unlikely, to me, that the basket with that shape and size could ever have an impostor that works well from different angles.

[Charlotte Bartlett]

Assuming they zero'd out the two lowest....
I still wonder how is still so low poly with the high being over 128K?
it's small -  which plays a part. I reckon with the other parts it could be over 250K,,,

If I am on mid setting the basket was ok if in the same room (technically all you need for interior furniture)
If I zoomed out; it became a Tris collapse -  but only once I zoomed out far enough (and if in a house in SL you wouldn't then see as it would be in a different room)...




 

[Tonk Tomcat]

I'm really sorry to say that, but who ever made this has plain and simple no clue how to do mesh for games. Crap like that does not belong into SL at all. 128k tris for such a simple piece of deco is a bad joke. And the Li doesn't say anything, because what he/she did is just to auto generate crappy LODs that will break really ugly as soon as you zoom out.

[ChinRey]

54 minutes ago, Charlotte Bartlett said:

Am I missing something on my LOD quest??

I wish I could give you a long answer but it's way past my bedtime, I have a busy day tomorrow and I suppose you'd want some clues ASAP....

First. Don't look at the lowest model only. Once a model is below somewhere between 12 and 30 (depending on the overall shape) tris it doesn't add anything worth speaking of to the download weight.

Second: See if you can split it into several meshes. An item like this, if it's well made, will have a lot of very small tris and also a few big ones and that makes it hard to control the LoD swap distances. You want a fairly low LoD swap distance for the small details to get rid of all those little tris but you want the big tris to stay solid for much longer. Also, there's a quirk in the LI/weight calculation that means that even udner otherwise identical conditions, several small meshes will have a lower download weight (but higher server weight) than a single big one. Server weight is always 0.5 per mesh and the ideal is is to get those two weight as similar as possible.

 

23 minutes ago, Tonk Tomcat said:

I'm really sorry to say that, but who ever made this has plain and simple no clue how to do mesh for games.

Oh yes, you're right. It does look gorgeous though. Unfortunately most of the best designers in SL are horribly bad at the technical aspect of mesh making. We can't really blame the for it either, partly because they are building commercially for a market with very narrow margins and don't really have time to spend on factors that don't increase the sales, partly because LL who should have set the standard doesn't seem to care.

[Chic Aeon]

36 minutes ago, Charlotte Bartlett said:

Assuming they zero'd out the two lowest....
I still wonder how is still so low poly with the high being over 128K?
it's small -  which plays a part. I reckon with the other parts it could be over 250K,,,

If I am on mid setting the basket was ok if in the same room (technically all you need for interior furniture)
If I zoomed out; it became a Tris collapse -  but only once I zoomed out far enough (and if in a house in SL you wouldn't then see as it would be in a different room)...




 

Had to smile :D.

 

First off, never look at the folks making HEAVY  mesh is you want to make "game asset" mesh.  And of course don't BUY it  :D.   There is plenty of this kind of thing out there --- more and more rather than less and less unfortunately.  

 

There are some folks that do a terrific job. I suggest going over to {what next} and look at her stuff.    ALSO it is important to test at least LOD2 which is the FS default (linden viewer is even less).  

 

I may get yelled at here, but depending on how the mesh is made (as in already low poly) it really IS OK to have a very very low triangle count for the lowest setting. The question is --- "is anyone going to SEE it at that low setting?  If you can cam out on a table and see it half way across the sim at LOD2, that is certainly good enough -- for me anyway.  So think about how each piece ACTS and not so much about the numbers.   I have seen some of that "OMG that is SO LOVELY" mesh with ALL but the top LOD zeroed out.  You can imagine.  Painful.

 

There are plenty of things inworld that should never have been uploaded into SL. That basket is one of them. SIMPLIFY.  If you start off simply enough you don't really have all the issues with the LODs.   I have been making low LOD stuff for almost a decade ON PURPOSE and for those with less than hefty machines.  Mine isn't the most gorgeous -- but it works and THAT was the plan.  "I" can make super heavy render mesh with 2048 textures too, I just choose not to.    

 

All that being said there are some folks who make impressive things with REASONABLE (not low poly) triangle counts and good LODs.  I just bought a lovely and very arty jewelry tray  that is impressively made.  Parts of the necklaces disappear fairly soon but then again they would in real life (for me anyway LOL).   

 

AND for your LODs you can add invisible  placeholders (small cubes, edges and I did get a single vertices to work ONCE  with no physics issues but never got that again - alas).   So you are tricking the viewer into thinking that the object is bigger than it visually IS.  This doesn't work for everything but can be handy for some things.

 

Hope you find a method that will work for you and your style. Good luck. 

 

 

 

 

[Charlotte Bartlett]

11 minutes ago, Chic Aeon said:

Had to smile :D.

 

First off, never look at the folks making HEAVY  mesh is you want to make "game asset" mesh.  And of course don't BUY it  :D.   There is plenty of this kind of thing out there --- more and more rather than less and less unfortunately.  

 

There are some folks that do a terrific job. I suggest going over to {what next} and look at her stuff.    ALSO it is important to test at least LOD2 which is the FS default (linden viewer is even less).  

 

I may get yelled at here, but depending on how the mesh is made (as in already low poly) it really IS OK to have a very very low triangle count for the lowest setting. The question is --- "is anyone going to SEE it at that low setting?  If you can cam out on a table and see it half way across the sim at LOD2, that is certainly good enough -- for me anyway.  So think about how each piece ACTS and not so much about the numbers.   I have seen some of that "OMG that is SO LOVELY" mesh with ALL but the top LOD zeroed out.  You can imagine.  Painful.

 

There are plenty of things inworld that should never have been uploaded into SL. That basket is one of them. SIMPLIFY.  If you start off simply enough you don't really have all the issues with the LODs.   I have been making low LOD stuff for almost a decade ON PURPOSE and for those with less than hefty machines.  Mine isn't the most gorgeous -- but it works and THAT was the plan.  "I" can make super heavy render mesh with 2048 textures too, I just choose not to.    

 

All that being said there are some folks who make impressive things with REASONABLE (not low poly) triangle counts and good LODs.  I just bought a lovely and very arty jewelry tray  that is impressively made.  Parts of the necklaces disappear fairly soon but then again they would in real life (for me anyway LOL).   

 

AND for your LODs you can add invisible  placeholders (small cubes, edges and I did get a single vertices to work ONCE  with no physics issues but never got that again - alas).   So you are tricking the viewer into thinking that the object is bigger than it visually IS.  This doesn't work for everything but can be handy for some things.

 

Hope you find a method that will work for you and your style. Good luck. 

 

 

 

 

here is where I am struggling - to compete sales with Heavy Mesh.... you have to make things look like Heavy Mesh as a large % of consumer SL wants that look(sigh).
Good pointers I will pop to her store to take a look.   The issue is not only am (I/we) competing with heavy mesh we are competing with the "it's only 3 LI" element too.

[Chic Aeon]

2 hours ago, Charlotte Bartlett said:

here is where I am struggling - to compete sales with Heavy Mesh.... you have to make things look like Heavy Mesh as a large % of consumer SL wants that look(sigh).
Good pointers I will pop to her store to take a look.   The issue is not only am (I/we) competing with heavy mesh we are competing with the "it's only 3 LI" element too.

Agreed, but BUT there IS another market besides heavy mesh.  And you can always find a more moderate middle ground. Honestly it is much more about the textures than the underlying mesh. The Looking Glass had builds that still hold up today that were made from prims a dozen years ago. The "trick" was that he made the textures himself in another program that simulated what we have today.   

 

While there are still plenty of people that are buying computer crashing mesh, others have stopped buying.  When I see a couch OVER A MILLION TRIANGLES there is no way I am going to support that creator. I sometimes buy the sale items for photo ops, but even though "I" can use that type of mesh now with my new computer, I can't support it's use.  

 

I hope you find a good middle ground.    You should paste in a picture of your problematic couch and maybe we could give you some ideas.   

 

OH, look at the newest relase of chairs from Trompe Loeil at Uber. THAT is impressive. Not "game asset" but so much better than so many creators. Might be some hints there too.    

 

Good luck. 

[Wulfie Reanimator]

Do you have examples of your own work and the LODs that are giving you trouble?

Are you sure your LI cost is caused by the LODs, or maybe the physics?

Also, what size is the object you're having trouble with? Different scale objects require different LOD budgets if you're going for LI.

Edited August 30, 2021 by Wulfie Reanimator

[ChinRey]

6 hours ago, Charlotte Bartlett said:

here is where I am struggling - to compete sales with Heavy Mesh.... you have to make things look like Heavy Mesh as a large % of consumer SL wants that look(sigh).
Good pointers I will pop to her store to take a look.   The issue is not only am (I/we) competing with heavy mesh we are competing with the "it's only 3 LI" element too.

I've told this story here several times before but it's one I think could do with repetition.

I was working on the beta grid one day when a newbie came up to me. Without even saying hello first, he asked "Why are meshes in Second Life so high poly?" It turned out he was a professional 3D modeller with years of experience in the film industry. He had joined SL because he wanted a place where he could build his own things, not only what his customers comissioned, and was absolutely shocked to see how wasteful SL content is. This is how we are regarded in the 3D modelling industry as a whole.

There are a lot of tricks and techniques to save tris (and vertices, pixels and LI) without reducing the visual quality. One of these days I'm going to write a set of practical instructions for the ones I know and use. It's a big job though, and RL has been rather demanding on me recently and will still be for a few weeks so it won't be anytime soon.

I did, however, write and post the introduction. It's all about the attitude, the general principles and why we need to optimise so there aren't many practical tips there at the moment. But that's where we all should start. It's easy enough to tell people what to do but they also need to understand why. Charlotte is one of the few great designers who also take modelling technique seriously so there's probably not much there she doesn't know already but others will read this thread too so here's a link to How to make efficient mesh - general principles.

While I'm at it, here's my Deadly Sins of Texturing blog post. It's a bit off topic for this thread but textures are at least as important for performance as geometry is and the two depend very much on each other.

Edited August 30, 2021 by ChinRey

[Charlotte Bartlett]

Thank you all and I am thankful the experts came out!!

I am fine with retopo - my sofa literally couldn't be less polygons and I used all the tricks (I bake the high poly onto the low in Substance etc so my creases are material led versus polys etc).    It's that final LOD low though - if I cheat and put in 0 - I am now at 5LI with the retopo versions and LODS above.  If I try to make a few blocky cubes purely to fake a bit of a shape it goes back up.  I want to balance commercial with being a good citizen but I have to say in this scenario I am VERY tempted to leave that as a zero purely as if you are outside the house (and indeed some distance away) you won't see the "collapse" at that lowest level. 

The 5 LI will compete with the heavy mesh makers whilst keeping a least a level of integrity.
I went to the stores mentioned and can see LODs are in good shape but still seeing a few of them are using the zero out method on the last LOD on small furniture.

On that basket I cry hah because I LOVE IT.  I know how well that creator sells (top 5-10 of sellers I suspect in SL H&G) and sometimes I just think like giving it up and chucking in a Z Brush model with that level of detail and a bare flick of retopo work.   Because out of the top sellers bar a few of the olderbies and teams like Cory/her partner it is what "sells" and I mean really sells...

I am going to put myself in the naughty corner for using a zero now.

Edited August 30, 2021 by Charlotte Bartlett

[Charlotte Bartlett]

11 hours ago, Wulfie Reanimator said:

Do you have examples of your own work and the LODs that are giving you trouble?

Are you sure your LI cost is caused by the LODs, or maybe the physics?

Also, what size is the object you're having trouble with? Different scale objects require different LOD budgets if you're going for LI.

I can put up a screenshot of the lowest LOD effort this week - it's is purely on the lowest LOD you can't get away with imposters for a sofa shape (that I can figure)

 

[ChinRey]

54 minutes ago, Charlotte Bartlett said:

It's that final LOD low though - if I cheat and put in 0 - I am now at 5LI with the retopo versions and LODS above.  If I try to make a few blocky cubes purely to fake a bit of a shape it goes back up.  I want to balance commercial with being a good citizen but I have to say in this scenario I am VERY tempted to leave that as a zero purely as if you are outside the house (and indeed some distance away) you won't see the "collapse" at that lowest level.

How big is the sofa? I do agree that when it comes to indoors objects you shouldn't worry too much about what people see from the outside through a window.  Assuming it's 1x2x1 m the swap distance to lowest is 32.67 m even with LoD factor 1 so it shhould be perfectly safe to zero out that model.

That being said, it is possible to make something that has a generic sofa shape with 12 tris and a lowest LoD model like that will not increase the download weight compared to a zero model. It may actually reduce the weight a little bit.

 

54 minutes ago, Charlotte Bartlett said:

The 5 LI will compete with the heavy mesh makers whilst keeping a least a level of integrity.

If you can do it at 5 LI as a single mesh, you should be able to do it at 3 LI as a linkset of five or six meshes and maybe even 2 LI as a linkset with four parts. As and added and quite significant bous, splitting will also allow for multiple avatars to sit on it at the same time.

 

Edit (forgot to comment on this):

54 minutes ago, Charlotte Bartlett said:

I am fine with retopo - my sofa literally couldn't be less polygons and I used all the tricks (I bake the high poly onto the low in Substance etc so my creases are material led versus polys etc).

That's how Hattie Panacek does it and when it comes high performance feature items, I can't think of a better example to follow than her. She's also an absolutely brilliant texture artist so unlike others who do it in a similar way (no names mentioned), she manages to do it without overloading her works with uber-laggy texturing.

Edited August 30, 2021 by ChinRey

[Wulfie Reanimator]

4 hours ago, Charlotte Bartlett said:

I can put up a screenshot of the lowest LOD effort this week - it's is purely on the lowest LOD you can't get away with imposters for a sofa shape (that I can figure)

For a 'regular sofa,' you need to pay attention to more than just the lowest LOD. For example, I made this intentionally unoptimized (30K tris) mesh with autogenerated LODs...

...with the lowest LOD like this (14 tris), the Download Cost is still 20.9 for a 3x1x1.3 object. Even if it was a single tri, it wouldn't get respectably lower (if at all).

The scale of the object affects how much influence each of the LOD levels have on the download cost.

This is why you should build your models to-scale for SL, so you can get more accurate estimates in the uploader for faster iterations. The basket you're comparing to is much smaller than a sofa, so it benefits a lot more from zeroing out the low/lowest LODs because they have much greater influence than the higher ones. (Because the object is small, the low/lowest LODs are expected to be most visible...)

Since you're working on a larger object, zeroing out Lowest, or even Low LOD won't be enough to solve your LI problems. For the kinds of scales we're probably talking about, you need to spread your focus across each LOD. Only very large or very small objects are "simple" to solve, since you generally only need to focus on High or Lowest.

One good thing about auto-generated LODs is that you can use them to figure out what your triangle budget should be to get the LI you want.

• In my case, zeroing out Low/Lowest would get me 20-4 LI, depending on what I do with the Medium LOD.
• Optimizing my High LOD to 7.5k tris (intentionally high again) gets me to 2 LI with auto-generated LODs. This already gives me a very good idea of what my budgets should be. I can now create custom LODs such that my Medium and Low levels match the given triangle counts. I could go even lower to hit that 1 LI without ruining the model.
  

Edited August 30, 2021 by Wulfie Reanimator

[ChinRey]

1 hour ago, Wulfie Reanimator said:

You can cut it down to 12 tris by changing the thickness of the back to 0. Here's a very rough mockup:

It probably doesn't matter much though. Wulffie's 14 tri model is symmetric enough there's probably nothing to gain from reducing it further.

 

1 hour ago, Wulfie Reanimator said:

For a 'regular sofa,' you need to pay attention to more than just the lowest LOD.

Charlotte told us she had already reduced the other LoD models as much as possible so it's probably not relevant here. But as a general point, how much each LoD model adds to the download weight depends much on the object's size or to be more precise, how big an area each LoD model is assumed to be visible across. For a very small object it's pretty much only the lowest LoD model that counts. For a very big object the high LoD model is the only one that matters. A sofa is somewhere in between and all LoD models will have a significant impact on the download weight so every tri in every model matters.

---

There are three more Li saving tricks I should have mentioned here and they are all based on compressability and the fact that even though we usually talk about triangle counts, it's actually the vertices, not the triangles that add the most to the download weight:

1. Split vertices. In 3D modelling programs like Blender and Maya each vertice can have different normals for the different tris assigned to it and it can be assigned to multiple materials and multiple locations on the UV map. This is not the case in SL and other game/virtual world engines so vertices often has to be split here. We can often save quite a lot by reducing the number of split vertices. There's usually not much to do about metrial splits but try to have as few vertices as possible along UV seams and try to use smooth normals whereever you can (but be aware that smooth normals sometimes backfire and increase the download weight).
2. Sorting order. The dae file includes a list of all the vertices in the mesh and the order they are lsited in can have a significant and hard to predict impact on the download weight. So try different options but keep a backup of the original in case it turns out to be the best one after all.
3. The exact location for each vertice both in 3D space and on the UV map affects the download weight and even minor shifts, too small to notice, can have a significant effect. There are lots of unpredictable factors here but the general rule is to get as much "symmetry" that is as many vertices with the same cordinate along at least one axis, as possible. One very quick and easy tool to improve the symmetry a little bit is the Snap to Pixels option on Blender's UV map. So make make it a habit to apply it unless your model really needs very precise UV mapping. Ideally you want to snap to the pixels on the default 256x256 UV map but it helps even if it's a larger 512x512 or 1024x1024.

---

Edit (Forgot something important again):

Some readiers may be confused because I keep talking about download weight, not LI, but it's the same thing really. Download weight is only one of the three weights that can determine the land impact, the other two are physics weight and server weight. Physics weight is determined by the physics model and hardly ever an issue for reasonably well made meshes. Server weight is always 0.5 for single unscripted objects and only ever relevant for linksets and some extremely rare cases of heavily scripted objects.

Edited August 30, 2021 by ChinRey

[Chic Aeon]

Not sure if this will help or not, but I made this recently. It was actually 3 li before I added all the animations and extra cubes so that many could sit (insert sad face).  

 

 

Edited August 31, 2021 by Chic Aeon
adding info

[Charlotte Bartlett]

You are all amazing - I am going through these.

Some of the theory I had, but having it clearly laid out is beyond helpful.  The amount of effort you all put in to help a hobbyist like me is so kind and thank you!

I hope other people get to see this too as it's really gotten me on the right track!   Thank you again!!!!!

[Moe McAlpine]

I am reading this with great interest right now. Thank you for all the helpful information and explanations. A question for understanding:

Quote

Sorting order. The dae file includes a list of all the vertices in the mesh and the order they are lsited in can have a significant and hard to predict impact on the download weight. So try different options but keep a backup of the original in case it turns out to be the best one after all.

How do I influence the sorting order? and were can i see it ? What should I pay attention to. This is interesting and completely new to me.

[arton Rotaru]

On 9/7/2021 at 3:10 PM, Moe McAlpine said:

How do I influence the sorting order? and were can i see it ? What should I pay attention to. This is interesting and completely new to me.

For Blender, check this out: https://docs.blender.org/manual/en/latest/modeling/meshes/editing/mesh/sort_elements.html

[Beq Janus]

On 9/7/2021 at 2:10 PM, Moe McAlpine said:

I am reading this with great interest right now. Thank you for all the helpful information and explanations. A question for understanding:

How do I influence the sorting order? and were can i see it ? What should I pay attention to. This is interesting and completely new to me.

It's really just a hack though. It does no harm and I don't particularly worry about it, there is a slight argument that it will load faster but that's unproven and it would certainly not be noticeable. 

The reason this "works" is simply that the download cost of mesh is based upon the size of each LOD in the asset file. The Asset file "zip" compresses the mesh data and by sorting it in different ways you can achieve better or worse compression. For  most meshes it makes little to no difference (it might be enough to shift you from 3.7 to 3.4 and thus save 1LI in the inworld sense. 

A few months back I actually got around to "making stuff" instead of fiddling with code. I made a bunch of hand tools for New Babbage where I live. 

I wanted each of these to work well enough that you could have them on a tool bench or workshop scene on the background or be carrying them in your hand, and ideally I wanted them to be 1LI (though I don't typically wed myself to these figures as people make far too many sacrifices for the sake of 1LI as it is)

The problem we have is that the current Land Impact system (deliberately) penalises triangle usage in the lower LIs harshly, especially so for small items such as these. As general (sweeping) approximation you have at most 20 triangles in the lowest LOD before you have no chance of getting your prized 1LI. 

Imposters are the way forward, not the lazy crumpled triangle nonsense of the item in the OPs image, which sadly we see far too much of (and is of course the very reason I wrote the LOD viewer capability for FS in the first place).

For the wrench, which is adjustable. making an imposter per segment allowed the impostered view to remain correctly adjusted (a detail nobody but me would likely ever notice!!)

https://i.gyazo.com/709b8bc22060110c18ffbdcf9b7fe993.mp4

My process for making LODs is typically to start form the high, or occasionally the medium, and deliberately cut away smaller sections of mesh. Get used to think about how small they will be on screen when they are at a given LOD in order to drive your design decisions. This can be tough, and the trade off used by @Chic Aeon of either padding the bounding box to inflate the radius (which will affect the LI) or just accepting that the item is not going to be seen at that distance is entirely valid (I swear my RL house keys have the lowest LOD zeroed, as I frequently fail to see them when they are right there on the table).  

Here is a small video of me quickly whizzing through the attempts I made at making viable LOD models. 

https://i.gyazo.com/f1d06b356e38500465cf44ff71072357.mp4

Note here, that most of the tools I am making are well suited to imposters, you see them side on for the most part, if they are not side on (on a tool rack or similar) then you probably don't care much (they'll be lost against the avatar holding them etc. The exception being something like the oil can which was a total pain because it needs a proper volumetric LOD model and it is asymmetrical meaning I cannot use the "plant pot" imposter trick of a star mesh.

In the end each one was 1LI, and each one is a single texture. I use a 512 for all except (apparently - I just looked)  the wrench and the matches to which I have granted a 1024 but I think that's because I haven't bothered to try the 512s yet!. I also consider those a guilty pleasure given that I don't sell stuff much so I'm not polluting anyone else's Second Life 😉