Well-Optimized Alternatives to Resource-Heavy Mesh Bodies for SL

Last week's post on the leading mesh bodies for Second Life causing major performance issues has provoked a pretty heated conversation in comments, with some blaming users (get a better computer!), others blaming Linden Lab (if high poly mesh is a problem, the Lindens shouldn't have allowed it!), to blaming big brands like Maitreya, Belleza and SLINK* (they have a monopoly on the market!),. Others helpfully pointed out some mesh body alternatives that are much better optimized than the market leaders. For instance, writes Fred Beckhusen:

We now have the free Bento Ruth [pictured above] and Roth avatars. They have far fewer polys, and are very good looking, with body, nail and skin HUDSs. Both are free and Open Source with extremely permissive AGPL licensing. Avatars use only two levels of details. For Ruth, these are:
Level of Detail Settings (LOD)
High - Triangles: 11444 Vertices: 8442
Medium - Triangles: 5563 Vertices: 4960

For more details, see this Github page.

SL Marketpage listings here.

Reader Sarali has a whole hosts of suggestions:

Psicorp's Avatar Core seems good and quite low poly as well. It is made by a furry designer but works just fine as a human: https://marketplace.secondlife.com/p/PsiCorp-Avatar-Core-21-Female-Type-A/13988561

As a bonus it is super duper creator friendly, with rigging models and textures easily available on their website and all, instead of hidden away and needing a form and vetting process like some of the big name bodies.

I hear there's also one in the works by Asami Imako of Artificial Incarnations and all of their clothing is *incredibly* well made and low rendering weight, while still looking fantastic. They really know what they're doing as far as I can tell so I'm waiting with bated breath on that body.

I would also throw an honorable mention out to the Avatar 2.0 by Utilizator. It's a bit older now and caters a bit more to an anime market than the other bodies mentioned, but simply as an example of a low poly body there are things to be learned from it. It's not fully updated to use the latest tech and it's a bit *too* low poly on its feet in my opinion, but it's still a good case study. I absolutely recommend checking out the "Normie" head by the same maker though. [Pictured] It is updated and super well made: https://marketplace.secondlife.com/p/UTILIZATOR-Normie-Head/13071631

The mesh heads by Lightstar are also quite well done in that regard from what I recall: https://marketplace.secondlife.com/stores/40878

But yeah, definitely check out the bodies mentioned if you care at all about the future of SL, and keep an eye out for Asami Imako's body when it comes out, as it should be really really great. The biggest problem really comes down to that there's not much clothing made for these bodies, but if more people use them and start asking for clothes for them, then more clothing will be made for them.

Please post other suggestions in Comments!

*And please don't infer this post or the original one as a criticism as Maitreya, Belleza and SLINK, which are visually impressive products that are probably fine, performance-wise, in solo/small group settings.

Comments

[Han Held]

I'm trying to figure out (and not having a lot of luck) where Tonic fits in here. Wearing a tonic body my avatar complexity blahblah is 107. Here's a screenshot of what editing the body shows W.R.T. the mesh info.

https://gyazo.com/c92614e3abfc677524e62e32d93d6f8a

I use the Ruth 2 body and try to support the project (mostly by buying clothes for it on kitely since I don't have any particularly interest in spending any time in altrightland A.K.A. Mewe), but I don't think that it comes off well (visually) in a side-by-side comparison against Maitreya.

[Han Held]

Tonic (+ Catwa) in wireframe;
https://gyazo.com/7fe59d52d94a2fb32647f5f0825165e2
Ow.

[Chic Aeon]

My Maitreya body and (old non-Bento) Lelutka head weighs in at 38840 on the complexity meter.

I think that is pretty darn good.

[Lagatha]

Unfortunately the barn door is open and the horses are out. We’ve gotten used to the high detail, ease of use and wider availability of clothing and shoes for our current bodies so the solutions presented above all have one or many significant disadvantages when compared to the current main mesh bodies. Gals I know will try every trick in the book to tweak their settings, some will descript if the club they go to has script counters or pay attention to their ARC due to vanity because they ‘work too hard on their avi to be jellied’ or will even organize their inventory but none I know would be likely to trade down in their basic look to what they consider a less-pretty head or body.

Despite the much more friendly and helpful tone here, I predict you still won’t get many people to downgrade their avatars in resolution (and in most cases other aesthetics) and to willingly use what they perceive to be a frankly less-beautiful product and in most cases pay for the privilege. I don’t think that’s going to happen voluntarily no matter how much information you share about it. Nor will LL force it or they would have already, I guess.

I hope Ruth and Roth 2.0 continue to develop but unless the faces, expressions, hands and alphas are refined and they are rigged to wear clothes that fit other mesh bodies, they will not be adopted and creators will not make for them. Classic 5-sizes rigging is becoming less and less offered, even in fullperm mesh templates. And what about their shoes?

With Altamura and eBody giving away a variety of bento mesh bodies complete with heads, some of whom fit much of Maitreya’s clothing and shoes, it’s a tough sell to market anything less detailed than that for sale at any cost. Who doesn’t have one or more of their alts in one of these by now? Sorry Ruth2, and Roth, these Altamura gals and guys are prettier, easier to use freebies. Most SL fashionistas don’t really mod too much aside from adjusting some parts to fit better so the open source feature is great but of limited interest to the majority.

A new human body maker would maybe be wise to rig a mid-density, mid-priced, mod-OK body model to wear Maitreya-rigged clothes and shoes, make their dev kits widely available and do heavy sales promotions in cooperation with the best clothing and shoe makers that opt in early to build for it, emphasizing to consumers that it’s the best of everything, beautiful, customizable, economical and more efficient. With bakes on mesh coming, the timing might be just right for it.

[Bruno]

@Lagatha A lack of optimization doesn't mean "more detail". It means the creator of said content didn't know how, or didn't care to know how, to manage their resources effectively to leverage the most detail for the least frame rate hurt. The Freya body doesn't look any more detailed than Maitreya despite using something like 10 times the number of polygons, and a competent CG artist can produce a body that looks as good as Maitreya at a fraction of the polygons. Those extra polygons are just wasted. Any body with a lower polygon count that looks just as good as bodies like Maitreya and Freya is a definite upgrade in quality.

You do make a good point that if someone wants to make a mesh body to compete with the likes of Maitreya they really should look into producing a body that is compatible with the clothing of one of the current popular bodies and price it competitively.

[Lagatha]

@Bruno But I have not seen any human bodies with a lower poly count or lower complexity that look as good as those main bodies with higher ones.
And if Chic’s head and body are that low in complexity then I am maybe not understanding why the fuss. I’m looking at more things in wireframe recently and I’m seeing some pretty dense mesh things that still weigh only 1 or 2 LI for decor stuff too. I don’t get it. If they are dense and unoptimized, why do they count so lightly? And why are there no best practices for optimized things defined for SL? With lots of things, if there’s no demo or you buy it in a box (like a gacha item at a yard sale) there’s no way to even tell.

I do wish that the various people that get fed up and go on a bender scolding everyone and bitching out LL would instead make some sort of consortium of good building or come teach at Builders Brewery or something...

[Pulsar]

The complexity number shown by the SL viewers isn't the best tool if you look for that and it's even worked around. Belleza has a monstrous polycount, but the complexity number is displayed low. Else, if the number worked better in that sense or the formula can be improved, jelly-dolls would encourage a better optimization, the same as jelly-dolls encouraged creators to make almost all the new products with a lower displayed number (and sometimes to update).

Some data by rezzing the bodies/parts (in no-script zones, some has anti-rez scripts) and checking with Firestorm object inspector tool.

Belleza Freya (body+hands +feet)
Vertices: 271.384
Triangles: 458.323

------------

Maitreya Lara 4.1
body only:
Vertices: 68.016
Triangles: 100.142

one hand (so multiply x 2)
Vertices: 20.844
Triangles: 38.036

Feet:
Vertices: 28.392
Triangles: 52.138

------------

Ebody curvy, free version 8.6

Vertices: 19.652
Triangles: 22.514

2 hands (single item)
Vertices: 21.640
Triangles: 34.552

feet
Vertices: 22.196
Triangles: 37.476

This is a pretty detailed body and yet it has fewer tris than the above ones

------------

Altamura Estephania (body+head+feet+hands)
Vertices: 77.176
Triangles: 129.924

------------

Psycorp Fem A
Fem A Body
Vertices: 30.048
Triangles: 38.916

2 hands (single item)
Vertices: 1.536
Triangles: 2.330

feet
Vertices: 3.152
Triangles: 5.254Vertices: 3.152
Triangles: 5.254

--------------

And I think the same as Bruno.
Lagatha has some good points, and some teaching would be good. I'm sure people would be happy to teach, and there are people who do and even blogged about optimization for years, some reactions are discouraging though and some people believe they are going to lose their pretty bodies (nope, just the wasted extra triangles).
You can get the idea already from the above counts already that you won't lose anything.

Also keep in mind that the skin plays a big role with the look. Ruth from the marketplace surely need improvements, but it comes with a very simple skin.
The fun thing is sometimes I have been asked what mesh head brand was mine, while in that moment I had a "system" one LOL.

[Blaise Glendevon]

Are designers making clothes/skins for them? Are the top tier designers? Until Blueberry, ISON, and their equals are, no sale.

[kenz]

misleading topic, meshbody never heavy when compare to clothes and accessories, sometimes hair. 20k to 30k rendering costs for anyone popular mesh body. As today standard 100k costs is not really heavy. But some hair or rings, or little stuff can be cost over 100k. No one will expect the old standard of 5k cost. We select body according to the shape, skin, and commonly accepted by clothes designer. Every businessman already invested that time and money for mesh body market. And most user already bought mesh body, not even one. Believe me, mesh body are already out of topic, except another wave of technology change.

[Fauve Aeon]

IMO, LL is probably not going to impose limits except to make the changes they have made to correctly reflect the render weights and maybe that will continue, and causing the jellydoll. Ditto for the continued adjustments to correctly reflect some other geometries that have been under-valued for their true weights in the past. And scripts. Outfits of mine that reflected green last year in a script counte nor are bright red so...there have been changes.
And changes to the viewer that have largely allowed us to control what we see or do not, often with a simple preset.
That's a LOT of progress in the individual's experience, at least it seems so to me.

So...the rest so far seems to be up to us as individuals too and HIGHLY SUBJECTIVE.
How optimized is optimized, how wasteful is wasteful, how detailed is detailed enough, how laggy is too laggy, how many clothes are 'enough' and how efficient ones the whole need to be once we get the shoes and earrings on?
All subjective and I don't see us reaching anything but the loosest coincidence in our answers.

Venue owners want their visitors to be under certain limits? They can impose the limits
...and we can choose to attend or not.
Lower weight items can be sold or even given freely
...and unless they are supported by makers, we won't wear them much if they fashion choices.
Anyone non-furry is not often going to embrace a furry-appearing avatar on themselves...or their partner(s).

Most end-user SL consumers are either not motivated or not skilled enough makers or modders to do much beyond basic resizing and mayyyybe linking and unlinking some things and possibly some basic retexturung, tinting, etc, less than these skills probably if they never had to adjust any prims by hand before resizers...and that's a majority, I'm guessing.
It's what I see when I have to coach people to fit their costumes. It's not required anymore to do much of that beyond slightly tweaking a hair or raising/lowering a necklace or moving an earring once in a while.

[NiranV]

@everyone

Pulsar is right. Complexity is not what you should look at. It is essentially meaningless.

I'm here to give you some Viewer-insight as to why.

The Viewer calculates complexity based on what feature is enabled, how much textures and how big these are as well as how much polygons something has. The last thing, the polycount however is badly weighted. It favors mesh content immensely over anything else, disregarding any impact bad mesh has. From the original calculation i learned that it roughly counts 15+ (up to 80) polygons as a single complexity point. In comparison, adding just a single light (create a prim and set it to cast a light) will add 512 complexity. That is an equivalent of 512*15 = 7680 polygons at minimum and in worst cases (taking 80) 512*80 = 40960 polygons. That is unrealistically disproportionate. We are essentially saying that a single light, which doesn't even do anything at all to your framerate is as heavy as a mesh with up to 40.000 polygons, one that when seen will absolutely have a visible impact on your framerate (a small one but still).

On the other hand, simple items such as prims have a minimum complexity (i think it was something around 1024 or so), even if that minimum was as low as just a single complexity, it would still be disproportionate. Taking the 15 polygon example and comparing it to a cube which has 6 sides, each side having 2 polygons meaning 12 polygons we would still have 3 polygons left to spare but still get 1 complexity. This disproportionate calculation drags through the entire way complexity is calculated not to mention it basically completely ignores rigged content and treads it complexity wise as static mesh which is completely wrong, rigged content has a lot of additional, very resource heavy stuff going on which is completely ignored here. All of this together caused a single avatar with 3.5 million polygons to have roughly half as much complexity as my own avatar which i spend a major deal optimizing all over the place (including textures)

Example Picture

So lets break down what is happening in this picture.

Look at the top right, the purple avatar, that's me, i'm completely naked, no clothing, no extra attachments other than hair and lights and i get 68000 complexity with 160000 polygons. Sounds fine right? Now look at the bottom right, there are three avatars, one of them has 3500000 polygons but only 40000 complexity. This avatar has over twenty times more polygons than me, many times more attachments (clothing and all that) which use way more textures than me too and yet she only gets 40000 complexity a bit more than half of what i get for much less. How is this possible? Why am i being punished?

I've reduced texture sizes.
I've reduced texture amounts (combined textures where possible and re-used the same as often as possible)
I've completely removed all invisible extra states i don't need. (even replaced the hand states with bento hands)
I've reduced my invisible light prims to a single small pyramid (down from a sphere) and gave it an empty white texture, even made them visible to prevent them being calculated as slower alpha objects.

What has she done? Nothing. Spammed everything possible on her avatar as she saw fit, probably incapable of doing any such modifications and optimizations anyway (due to no-mod) and yet there she is, having less complexity. It drove me nuts.

The explanation is above. Here's the breakdown. My avatar itself is ~8000, my flexi alpha masked (once again an optimization so its handled as solid, not alpha object for faster rendering and no alpha flickering) hair weights 27000 alone, my lights on my back and head weight roughly 31000 thousand, the whole rest. Most of my complexity comes from a few tiny lights that most users can't even see properly because they are not running Deferred Rendering (aka Advanced Lighting Model) and are unable to see more than 6 lights at any time, even if they were, these few lights basically don't touch your framerate in any noticeable way.

I went absolutely apeshit when i saw this and set out to change this faulty complexity calculation to heavily punish these high poly cases. I even went to LL to make a huge drama and start a big discussion about it with the other TPV devs which finally got LL to rethink it, they will be looking into re-evaluating the complexity values soon but don't think it's going to be good, their answer "we're not going to punish bad content" basically explained this whole project as dead before it even began so don't hold your breath. They are most likely going to do minimal number shuffling and at the end your complexity will be the same, 3 million poly avatars will still run around with 40000 complexity and heavily optimized avatars will still get punished for things like flexi or lights.

TL:DR complexity is pointless, it makes so sense and favors heavy avatars over optimized ones and mesh over prim regardless of their actual impact.

For anyone interested, i've changed the complexity calculation in BD and had much better luck with my own version of it, doing a much better job at jellydolling heavy avatars and keeping my framerate decently high even in heavily crowded places.

[uriesk]

@NiranV
Why do you compare a mesh that has no light enabled with a prim that has light enabled?
Light adds lots of complexity because it can cast shadows and actually influences the environment. Shadows are resource intensive.

Also a prim cube has not just 6 sides * 2 triangles = 12 polygons. That's not how second life prims work. A cube in SL, which can be changed to be cut or hollow, has 48 triangles.
I tested it and if i attach a default prim cube to my right arm, it adds 404 complexity. Then i exported it as dae and uploaded it as mesh with all LOD levels to the highest possible (because a prim doesn't have LODs, just mesh have that) and attached this mesh cube to my right arm. It adds 404 complexity.

It is exactly the same number, a prim gets calculated by the amount of polygons it actually represents in SL. It gets treated exactly the same as mesh.

The reason why an equal or high quality mesh item gives you less complexity is because it has LOD levels. When you look at it from 40m away, just some crude approximated shape gets rendered. When you stand right in front of it, it gets rendered in full detail. This is a huge performance advantage! Of course this has to be part of the complexity calculation.

Another thing is the calculation of rigged mesh.
LOD levels of rigged mesh work differently, they get favoured.
If you wear a unrigged mesh attachment that has LOD Lowest and Low to 0 (means just as many polygons at it has faces -> it disappears on those levels), the attachment will have super low complexity and it actually disappear if you zoom away from the avatar (like it's supposed to).
If you do the same thing on rigged mesh, it will still show from far far away, the Lowest LOD levels change nothing. But it is still recognized in the Complexity Calculation and it will be super low. This is what Ruth 2.0 here does to.
This is a wrong calculation.
Giving a mesh object a super low complexity because it vanishes as soon as you are 5m away from the avi (and therefore doesn't need to get rendered) is logical and right. But the Low and Lowest LOD levels from rigged mesh gets basically ignored, so rigged mesh shouldn't get that bonus for those being zero.