Firestorm 5.0.11 New Mesh Building Tools

[ChinRey]

3 hours ago, Callum Meriman said:

That's exactly what I want. Coming to look at your works tonight Chinrey!

Ummm... I'm not sure if I had time to list that one for sale yet. As CSVirtual correctly hinted at, I have been focusing mainly on low poly "sim filler" vegetation so far because that's what I'm mainly interested in myself - I don't want just a few trees scattered about, I want the whole of Second Life to be covered with lush vegetation!

That brings me back to something I should have posted a fuller answer to:

16 hours ago, CSVirtual said:

It is exactly what I said. You use the forum to promote yourself by writing criticism on others.

I do not regard those high detailed feature trees as competition to my builds at all, it's a different market. But I do think those trees and my larger plant fields need each other: large low poly groups for volume and high detailed feature plants for accentuation. I miss those accent plants and I wish there were more of them - more well made ones that is. I can make them myself - I do make them myself occasionally - but I'd rather focus on the big volume fields because that's something nobody else seems to be doing at all.

 

16 hours ago, CSVirtual said:

Where is the laziness, on nice design trying to handle how Li is managed by Second life, or make billboard trees using just free textures ?

I should actually be flattered that you are comparing my entry level 1 L$ 16 polys plants to detailed high poly trees by others.

There is no laziness if the creator tries to handle the LI but look at Beq's animated illustration. The makers has simply zeroed out most all LoD models. And then read Chic's:

14 hours ago, Chic Aeon said:

The download cost in the example shown is 1.1. physics .7 and server 2.3. The download parameters could weigh in at up to 2.5 and still keep the same land impact.

So the maker could have kept far more details in the LoD models without increasing the land impact simply by changing the numbers in the uploader window. And that's just the first little baby steps in LoD/LI optimization. There is no sign of the maker trying to handle the LI in an appropriate way. I can't say if it's because the maker is lazy or ignorant of course but if it's ignorance, the answer is very simple: study and learn.

Finally before I quit this thread - since CSVirtual seems to be so hung up on my entry level plants: Yes, I am one of several plant makers to include simple, basic "crossed sheet" plants in my product range. And I do use every texture I can get my hands on for those plants to give people the widest possible range of choices. The only differences between my basic paltns and the oens from other makers are that mine tend to be cheaper and are made from ultra efficient low poly mesh and alpha masked textures rather than odl laggy sculpts/prims and alpha blending textures. I am considering taking them off the Marketplace and only sell them inworld, partly because there have been one or two misunderstandings but mostly because my more elaborate works tend to drown in the masses. But I do think that such simple plants still have a place in Second Life. Used in the right places they look good and enrichen the scene, they don't cost much and when made the way I made them you can fill up your place with as many as you like with no noticeable increase in the lag.

Edited March 7, 2018 by ChinRey

[Chic Aeon]

38 minutes ago, MIstahMoose said:

Counter argument (Sorry I love playing devils advocate) for as far as creators ignoring the optimization side, a lot of those larger creators do this as their sole income and optimization is largely unnoticed by the mass population of SL users. Having to create so much in a month for an event is pretty hard, and if you miss and event or two people start to forget your name. Time or necessity for optimization just isnt there imo, for most things at least. (I will say those creators who just do recolors should have time to optimize there stuff smh *throws shade* )

I think this problem really is largely on Linden Labs shoulders, Take a look at engines like Unreal Engine 4. That shiet optimizes for you, and does it so dang well I can scale my 4k models to MOBILE instantly, with good quality! Yes creators can be blamed, but we are still not given the tools we need to create quality content. A lot of the work arounds are dirty and they really shouldn't have to be. Spending so much time just to get an upload to work is frustrating beyond belief, and having to optimize on top of that? bleh. 

BAD argument :D. It only takes a couple of clicks to upload a model USING THE UPLOADER PARAMETERS in a much better manner. No time at all is involved. 

SL is  MY major income also and I am in many events every month. Still -- I manage to click those arrows in a more "sane" manner so that more folks can actually see and use my products LOL. 

Not much effort at all. 

Edited March 7, 2018 by Chic Aeon

[Chic Aeon]

5 minutes ago, ChinRey said:

But I do think that such simple plants still have a place in Second Life.

I still have some VERY old rose bushes that are lovely and that I still use. They load a little slowly with PINK showing up (I think they must be sculpts) but they are simple and don't seem to cause any noticeable lag, so for me they work fine. 

 

[MIstahMoose]

Just now, Chic Aeon said:

BAD argument :D. It only takes a couple of clicks to upload a model USING THE UPLOADER PARAMETERS in a much better manner. Not time at all is involved. 

SL is also MY major income also and I am in many events every month. Still -- I manage to click those arrows in a more "sane" manner so that more folks can actually see and use my products LOL. 

Not much effor at all. 

Those parameters are terrible for most models? What you can optimize yourself gives you vastly better results than those will. It almost completely destroys most models past the highest LOD. If you are wanting this to be visible from any view, then I am guessing you'd want their textures to not  get completely and utterly destroyed by the auto generated LOD too? Which is what happens if you use it most of the time.
They may see the silhouette but is that all you're going for? Because like you said, somethings melt right away. But even if you give it a little more polycount at lower LODs, the textures still get absolutely destroyed by the uploader while theyre still in view, and you have no control over that really?

I don't feel thats a bad argument, just a difference in opinion of what we want out of the uploader

[Chic Aeon]

1 minute ago, MIstahMoose said:

Those parameters are terrible for most models? What you can optimize yourself gives you vastly better results than those will. It almost completely destroys most models past the highest LOD.

I completely disagree. I use the uploader setting daily with VERY good results (certainly MUCH better than the example that falls apart). One only needs to MANIPULATE and TEST and for most things (not all) the uploader will do an adequate to good job.

It depends a LOT on how your model is made. Mine are ALREADY low poly.  

It would have been VERY easy to upload the tree with higher settings using the uploader. It could have been done with NO increase in land impact. Not doing that is beyond my comprehension.  Since the folks that are making this 'fall apart' mesh are NOT even using the uploader defaults, I have to assume that they are MAKING CHOICES -- just not ones that make much sense to me :D. 

 

I am not arguing that the uploader does a better job  than hand-made LODs (although I haven't been able to do any better in many tries). But saying it it isn't worth uploading with the uploader choices when it won't be ANY HIGHER LAND IMPACT  (in the case of the tree) is really ridiculous.  Not going to argue this any more. Heading to bed. 

 

 

[MIstahMoose]

Just now, Chic Aeon said:

I completely disagree. I use the uploader setting daily with VERY good results (certainly MUCH better than the example that falls apart). One only needs to MANIPULATE and TEST and for most things (not all) the uploader will do an adequate to good job.

It depends a LOT on how your model is made. Mine are ALREADY low poly.  

It would have been VERY easy to upload the tree with higher settings using the uploader. It could have been done with NO increase in land impact. Not doing that is beyond my comprehension.  Since the folks that are making this 'fall apart' mesh are NOT even using the uploader defaults, I have to assume that they are MAKING CHOICES -- just not ones that make much sense to me :D. 

 

I am not arguing that the uploader does a better job  than hand-made LODs (although I haven't been able to do any better in many tries). But saying it it isn't worth uploading with the uploader choices when it won't be ANY HIGHER LAND IMPACT  (in the case of the tree) is really ridiculous.  Not going to argue this any more. Heading to bed. 

 

 

Well there comes the difference, I was thinking of high poly models  If you try modeling something complex and using the uploader, it will be destroyed. The uploader is more than sub par in that category. With low poly and simple objects , yeah it works fine. 

Have a good night, Hope this didn't stress you out ^.^

[ChinRey]

2 hours ago, MIstahMoose said:

but what other context is it important for LOD to be correct. Props etc?

LoD is one of the main reasons why Second Life exists at all.

There is no gpu int he world that would be able to handle the full impact of a typical SL scene in realtime so it has to be simplified as much as possible before it can be rendered. The three main simplification mechanisms are:

• Object-object occlusion: Obhects completely hidden behind otehr objects are ignored
• Draw distance: Object further away than the set draw distance are ignored
• Level of Detail (LoD): Objects so far away you can't actually see every little detail (and that's the majority of objects in the scene) are simplified.

Now, Second Life included an automatic process for simplifying prims right from the start. They defined four different LoD levels and for each one a quarter of the vertices were eliminated. This workes reasonably well because prim shapes are relatively few and fairly easy to define. It didn't work so well when they tried to adapt the system to sculpts though, and ...

(about 20,000 words of techy explanation omitted)

... that is the reason why we have all thise fuzz about LoD today.

It is of course hard to say exactly how many active triangles is acceptable in a scene. It depends on how powerful your gpu is, how sensitive you are to lag (older people can be happy with 10 fps, while the young and impatient may not be satisfied until they get 40) and of course, several other factors. But as rule-of-thumbs benchmarks: 100,000 shouldn't cause any problems, with 200,000 laptops wil start to struggle, 300,000 may be ok for regualr home desktop computers, 400,000 is game computer territory, 500,000 is high end game computer territory. Once you get close to a million triangles, you need a render farm, not a single computer - and SL doesn't even have support for multiple gpus.

Now, if we use the tree of conflict here as an example. The high LoD model has about 9,000 triangles. With the RenderVolumeLODFactor set to 4, the high model will still be the active one at 40 m distance. There won't actually be enough pixels on the screen for the tree to render all those triangles but the gpu will still have to deal with them and try to squeeze them in somehow (if I understood Beq right, that "squeezing in somehow" thing causes some additional problems on its own). Now, imagine 100 such trees or other objects of similar complexity within view distance.

 

2 hours ago, MIstahMoose said:

I personally still believe the texture sizes are bigger issues than LOD.

Yes, probably, but we have to look at both.

Fortunately it seems Linden Lab is finally doing that. From what I've heard, they are working on a new system for calculating land impact now, one that also takes client side load into account and doesn't penalize good LoD as much as the existing one does.

 

37 minutes ago, MIstahMoose said:

Counter argument (Sorry I love playing devils advocate) for as far as creators ignoring the optimization side, a lot of those larger creators do this as their sole income and optimization is largely unnoticed by the mass population of SL users.

Oh yes, I've seen and heard that one too. "Why should I care, people in SL buy it anyway". You can't argue against it but you won't find many merchants who are willign to stand up in public and admit that's the way they do it.

In any case, making LoD models is not actually very difficult. That is, it is difficult if you want them perfect - you can spend hours even on the simplest model for that. But making something considerably better than the ones out infamous tree has - that's only a few minutes work for a reasonably experienced mesh maker.

 

47 minutes ago, MIstahMoose said:

I think this problem really is largely on Linden Labs shoulders

I can't say I disagree with you there. But we have had mesh for more than six years now and I do think that when makers see that what they do doesn't work, they do have a responsibility to try to figure out why and how to fix it.

 

[MIstahMoose]

31 minutes ago, ChinRey said:

 

I can't say I disagree with you there. But we have had mesh for more than six years now and I do think that when makers see that what they do doesn't work, they do have a responsibility to try to figure out why and how to fix it.

 

Cant really argue with your points, they are fair ones to make However It -has- been 6 years.. And.. Little to no improvement from linden lab on mesh has been quite disheartening. (Don't say bento is an improvement please Its nothing like what they promised us YEARS before bento was even implemented...its also not entirely mesh related -even if animation is tied directly to mesh-) 
A lot of these methods are going to be set in stone until they adjust how things are calculated. Heck I remember arguing that I wanted LL to approved every model uploaded to SL when it was about to be released. Because what they were doing in the test server was setting up for disaster.. which is what we have now :\ 

Unfortunately no matter how much you argue with creators, the market is consumer driven. As long as there are people buying and demanding 1 LI items at high detail and high texture.. There will be a supply. It really depends on Linden Lab to make SL into a user friendly game. The Creators will always work to the limitations of what we're given, but right now we're not given much to work off of. Heck you can tell how much neglect the freaking uploader has just by something as simple as the "Name your mesh" Doesn't actually name your mesh!  

Anyway I just think im repeating myself at this point, I like seeing what everyone's opinions and thoughts are. I really wish creating game ready models was the trend in SL, but even when it does become the trend they need to make Advanced lighting usable, because the only way game ready beats high detail is when we can use normal and specular maps at -all- times. Which like I said in a previous post, Is unusable even if you make something with 100% efficiency. |

ETA: 
Big step in teh right direction would be if linden lab just deleted everything from 2014 and earlier LOL
That old stuff.. Lag central.

Edited March 7, 2018 by MIstahMoose

[Rya Nitely]

I wonder how many people use or know about this trick. I stumbled on it by accident, and said 'ooh, that's useful'.

I sometimes get the uploader to generate my LOD models and then after I upload the item I save the 3 highest LOD models back to inventory, and then improve them. It gives a good starting point. You could do this even before the latest firestorm feature by zooming out and save what you see.

Maybe most people here know this, but for those like me who didn't until a few months ago when I stumbled on it by accident when saving a collada and I was zoomed out too far, this is very useful, or to get back an LOD model you lost or deleted by accident.

The lowest LOD model is the most critical and has the biggest impact on LI, and so that still takes the most work as the uploader hardly ever will give you a decent one.

 

Edited March 7, 2018 by Rya Nitely

[Rya Nitely]

1 hour ago, ChinRey said:

In any case, making LoD models is not actually very difficult. That is, it is difficult if you want them perfect - you can spend hours even on the simplest model for that. But making something considerably better than the ones out infamous tree has - that's only a few minutes work for a reasonably experienced mesh maker.

It really does depend on the model and the LI you are aiming for. For me, trees are not so easy to get a decent representation that doesn't change shape too much going from high to med to low. I don't do lowest on trees, and sometimes I skip low. Would love to see the LODs on your tree

[ChinRey]

4 hours ago, MIstahMoose said:

Unfortunately no matter how much you argue with creators, the market is consumer driven.

No, it isn't. It's marketer driven. Second Life is such a fractured, huge and flooded market nobody really knows what is available here. Obviously people won't buy from you if they don't know about you at all and unless you want to spend ten years slowly building a fan base, the only way to get noticed is to get a foot inside on the big prestigious events and blogs. They are the ones who decide what people want and buy.

[ChinRey]

4 hours ago, Rya Nitely said:

The lowest LOD model is the most critical and has the biggest impact on LI,

That depends a lot on the object's size but for most meshes, yes.

 

4 hours ago, Rya Nitely said:

Would love to see the LODs on your tree

Aww, that would be telling.

But here are some numbers to look at at least. This is only a tree trunk without foliage:

Maybe I shuld mention that in this particular case, the lowest model doesn't affect the land impact at all, it's all about the mid and low models.

There are no secrets here, I think all the technical details have been posted several times ont his forum over the years. But yes, it can be hard to find it all. There won't be any good, complete SL mesh manual until Linden Lab is willing to pay me somebody from IMMUFPG (the Informal Mesh Mystery Unveiling Forum Project Group) to write it and I can't see that happening this century.

There is an artistic side to good LoD models too, it's not all techy stuff. You have to have an eye for details to be able to determine what is necesseray and what isn't and you have to be willing and able to follow the one universal rule for all arts:

1. Kill your darlings!

Edited March 7, 2018 by ChinRey

[Love Zhaoying]

28 minutes ago, ChinRey said:

There won't be any good, complete SL mesh manual and until Linden Lab is willing to pay me somebody from IMMUFPG (the Informal Mesh Mystery Unveiling Forum Project Group) to write it and I can't see that happening this century.

Was thinking the other day, there’s so many books (even if they are like “Dummie” books) - there should be books on creating mesh for SL. It’s a conspiracy, to keep the knowledge in the hands of the few!

[ChinRey]

5 minutes ago, Love Zhaoying said:

Was thinking the other day, there’s so many books (even if they are like “Dummie” books) - there should be books on creating mesh for SL. It’s a conspiracy, to keep the knowledge in the hands of the few!

It's not a conspiracy but it would be a big job, far more than you can expect anybody to do in their spare time. Linden Lab should have done it and I understand they even promsied to but it never happened. It may be just as well they didn't because I don't think they have the qualifications in-house. So much of it is about ways to work around the mistakes they made.

Edited March 7, 2018 by ChinRey

[Love Zhaoying]

1 minute ago, ChinRey said:

It's not a conspiracy but it would be a big job, far more than you can expect anybody to do in their spare time. Linden Lab should have done it and I udnerstand they even promsied to do it but it never happened. It may be jsut as well they didn't because I don't think they have the qualifications in-house. So much of it is about ways to work around the msitakes they made.

I mean, even 1-2 experts could get together and publish an e-book, update it occasionally - with they same info they put in their blogs.

[ChinRey]

4 minutes ago, Love Zhaoying said:

I mean, even 1-2 experts could get together and publish an e-book, update it occasionally - with they same info they put in their blogs.

It would still be a big job and... ummm...

I don't want to sound arrogant and I do admire some of the good technical blogs out there, particularly Chic's helpful newcomer tutorials and Beq's indepth technical explanations, but if you want the full picture, explained by a content creator for content creators in a way easily udnerstood by cotnent creators, you need to involve either Drongle, Aquila, Arton, Optimo or me and none of us have a blog.

[Love Zhaoying]

2 minutes ago, ChinRey said:

It would still be a big job and... ummm...

I don't want to sound arrogant and I do admire some of the good technical blogs out there, particularly Chic's helpful newcomer tutorials and Beq's indepth technical explanations, but if you want the full picture, explained by a content creator for content creators in a way easily udnerstood by cotnent creators, you need to involve either Drongle, Aquila, Arton, Optimo or me and none of us have a blog.

Slacker!!!

 

[Chic Aeon]

6 hours ago, MIstahMoose said:

Big step in teh right direction would be if linden lab just deleted everything from 2014 and earlier LOL
That old stuff.. Lag central.

I would have to see a lot of "proof" to go along with that. Some of the current mesh being made is worse than the old stuff. Now that it is "semi-legal" ? to upload models from the 3D mesh catalogs, more folks are doing it and most of those models are WAY too heavy for SL no matter what you do with the LODs.  This includes some popular creators which is pretty disturbing to me on a creative level.   

 

In 98% of cases NO ONE should be uploading this high poly mesh. I can't actually think of any instance but I don't like absolutes. 

I will say that PHYSICS has improved overall during those years. We don't see quite as many phantom houses with prim walls as we used to and it is more likely that folks can rez on the floor these days. 

I have some very good low poly mesh from 2013.  I even have some that I made back then that is still very good and usable and I haven't "remade" simply because the new product wouldn't be that much better.  

[Whirly Fizzle]

7 hours ago, ChinRey said:

It is of course hard to say exactly how many active triangles is acceptable in a scene. It depends on how powerful your gpu is, how sensitive you are to lag (older people can be happy with 10 fps, while the young and impatient may not be satisfied until they get 40) and of course, several other factors. But as rule-of-thumbs benchmarks: 100,000 shouldn't cause any problems, with 200,000 laptops wil start to struggle, 300,000 may be ok for regualr home desktop computers, 400,000 is game computer territory, 500,000 is high end game computer territory. Once you get close to a million triangles, you need a render farm, not a single computer - and SL doesn't even have support for multiple gpus.

Too many triangles in a scene will cause a "black hole" where any object rezzed in the area (prim or mesh)  will be invisible.
If you have RenderVolumeLODFactor set high in the viewer, you are more likely to see these black hole spots inworld.

To quote Runitai Linden:

Quote

You've run up against a performance/stability fence in the viewer

Grouping multiple millions of triangles close together is making the viewer run out of memory for that spatial partition node. The threshold is controlled by a debug setting "RenderMaxNodeSize" and defaults to 64MB per node.

The triangle count is just unreasonably high

Details are on this JIRA issue: BUG-9439 - When certain objects are rezzed, they cause other nearby objects (including default prims) to render invisible

This is yet another reason why product notecards telling the user to set their LOD at some stupid high number is a really bad idea.
Even raising LOD from 2 to 4 will cause a drastic increase in patches of invisible objects for the user.

[Whirly Fizzle]

7 hours ago, ChinRey said:

There is no gpu int he world that would be able to handle the full impact of a typical SL scene in realtime so it has to be simplified as much as possible before it can be rendered. The three main simplification mechanisms are:

• Object-object occlusion: Obhects completely hidden behind otehr objects are ignored
• Draw distance: Object further away than the set draw distance are ignored
• Level of Detail (LoD): Objects so far away you can't actually see every little detail (and that's the majority of objects in the scene) are simplified.

I'd add texture discard to this list.
Enable TextureLoadFullRes ( If TRUE, always load textures at full resolution (discard = 0).  Will cause excessive memory pressure) & see how badly your performance suffers.
Even a top of the line gaming system will be brought to it's knees with zero texture discard.
A 32bit viewer will have an out of memory crash within minutes. A 64bit viewer will plummet to 1-2 FPS within a few minutes, even on a top of the line system.

Unfortunately TextureLoadFullRes is another setting that some creators tell users to enable in their product notecards.

[Love Zhaoying]

Hmm..reading this reminds me that I need to replace my “tropical foliage” with mesh. Since I will need to be able to put my script in them and Rez on demand, do you think I will have trouble finding good copy/mod mesh tropical?

Edit: Sorry to be off-topic. Guess I was asking @Chic Aeon.

Edited March 7, 2018 by Love Zhaoying

[OptimoMaximo]

32 minutes ago, Love Zhaoying said:

Hmm..reading this reminds me that I need to replace my “tropical foliage” with mesh. Since I will need to be able to put my script in then and Rez on demand, do you think I will have trouble finding good copy/mod mesh tropical?

I will make you some, with these features:

• sculpted leaves, 4 different shapes, averaging a nice 65k vertices each
• each leaf will be split in 8 materials
• each material will carry a 1024 texture
• each texture will have also its own normal and specular map WITH alphas for more data to read (glossiness and environment)
• and finally, the Medium, Low and Lowest LoDs will be one tiny triangle i will upload separately to make sure it's small enough to never be seen at any distance

Oh, and for good measure, aside from the LoDFactor you will have to push to this level described here ,

57 minutes ago, Whirly Fizzle said:

TextureLoadFullRes is another setting

you will have to enable to enjoy real high quality

How does it sound?

 

[Love Zhaoying]

3 minutes ago, OptimoMaximo said:

I will make you some, with these features:

• sculpted leaves, 4 different shapes, averaging a nice 65k vertices each
• each leaf will be split in 8 materials
• each material will carry a 1024 texture
• each texture will have also its own normal and specular map WITH alphas for more data to read (glossiness and environment)
• and finally, the Medium, Low and Lowest LoDs will be one tiny triangle i will upload separately to make sure it's small enough to never be seen at any distance

Oh, and for good measure, aside from the LoDFactor you will have to push to this level described here ,

you will have to enable to enjoy real high quality

How does it sound?

 

I must be smarter than I look!

You’re just checking if I’m paying attention!

[OptimoMaximo]

16 hours ago, Rya Nitely said:

Hidden triangles work very well, but I hesitate to use them. If customers want to change the texture on an item (which will happen with full perm items that I sell) then you need to explain to them about that hidden triangle(s)

It's what a developer does, Rya: document the work they intend to sell or give away to others to include in their projects. When you purchase a full perms script, the least you expect is that it contains fully commented code, so that you know what is done and where to modify it, don't you? It's how development material has always been. And you can certainly add a few lines to a notecard explaining why and how to do so. Perhaps this might be one of the reasons for some creators to keep their stuff no mod.

[OptimoMaximo]

10 hours ago, MIstahMoose said:

I think this problem really is largely on Linden Labs shoulders, Take a look at engines like Unreal Engine 4. That shiet optimizes for you, and does it so dang well I can scale my 4k models to MOBILE instantly, with good quality! Yes creators can be blamed, but we are still not given the tools we need to create quality content. A lot of the work arounds are dirty and they really shouldn't have to be. Spending so much time just to get an upload to work is frustrating beyond belief, and having to optimize on top of that? bleh.

Can i ask what is a 4K models? 4 K texture? However, aside from that, the UE4 LoD making editor does a pretty good job in almost all cases but it still suffers from a decimation type of approach. Smarter and better than SL's own LoD generator, still they recommend complex objects are better off having their custom LoDs.

I don't remember any one failed upload to SL i attempted that wasn't my fault, forgetting to do something prior the export; if you understand the concept of "clean mesh" in terms of inputs and geometry it's not hard to figure out what doesn't work. It's not a matter of optimizing on top the troubleshooting of the misbehaving mesh which doesn't want to upload, it's more a matter of working optimized from start to end. Because it's easier to add geometry at later time, rather than remove it. And if the process implies the use of a high definition object, it's for maps extraction to fake those details on a low(er) polygon version of the same item. So yes, you build the same object twice in order to make it good, both looking AND technically.