manual LODs: how necessary?

[Quarrel Kukulcan]

How important is it to manually create your own LODs? How bad are the auto-generated ones? How much do the answers depend on the environment? On whether it's an avatar? On what viewers and settings other residents are using?

[ChinRey]

1 hour ago, Quarrel Kukulcan said:

How important is it to manually create your own LODs?

If you only build for yourself, it's nobody's business but your own. If you're distributing your meshes, I'll say making LoD models manually is mandatory.

 

1 hour ago, Quarrel Kukulcan said:

How bad are the auto-generated ones?

Very bad.

The uploader can actually make good looking LoD models but only if you set the triangle counts much higher than the default - and much higher than what you need for well made manual models. This will result in bloated land impact and render cost. The Moles do it this way and Mole mesh typically has four times or more as high LI and render cost as properly made mesh.

 

1 hour ago, Quarrel Kukulcan said:

How much do the answers depend on the environment?

Not at all.

 

1 hour ago, Quarrel Kukulcan said:

On whether it's an avatar?

Well, at the moment the entire LoD system is broken for fitted mesh. And of course, there is no land impact restrictions for avatars. That means you can get away with all kinds of LoD cheating for avatar attachments. But LL is working on ways to fix those msitakes so it may not last .

 

1 hour ago, Quarrel Kukulcan said:

On what viewers and settings other residents are using?

If you are making thinks for others to see, don't assume they are willing and able to crank up their viewers' LoD factor to cover up your mistakes.

Edited May 12, 2020 by ChinRey
Typos

[Wulfie Reanimator]

Creating your own LODs is important for two reasons:

1. The auto-generated ones are completely unusable. They don't serve a purpose. They just destroy the object and make it unrecognizable from a distance.
2. Creating your own LODs is easy. Essentially all you need to do after your main model is done is dissolve some edges/vertices.

Can you tell me what any of these even are? You might guess the last one!

Edited May 12, 2020 by Wulfie Reanimator

[Chic Aeon]

It also depends on how complex your initial model is as well as how it is shaped.  For some things the uploader does an adequate job. So TESTING inworld is a very important step.  For ME, most of my items can be seen by most people  from a much further distance than is necessary -- so a table can be seen  well for instance at 64 meters when it will actually be used inside a house and would rarely need that level of detail .  I can't remember a time when I had an object look poorly from a distance and most of my work is lower poly game asset mesh.  

Be SURE and test at LOD 2 (or lower). Making mesh that only works at LOD4 is a VERY BAD IDEA.    I have to say that for the most part the creations I see inworld for sale have opted for higher land impact over bad LODs (to lower land impact). This isn't good practice either but "I" prefer it over lovely things that fall apart a few meters away. There are still those items of course. Just make sure YOURS don't belong in that category. 

[ChinRey]

1 hour ago, Chic Aeon said:

For some things the uploader does an adequate job.

Can you give an example of this? I've made thousands of meshes myself and studied thousands of meshes by others and I thnk I can count on one hand the occasions when the uploader managed to generate half decent LoD models. And even on those occasions it was fairly easy to make something better manually.

 

1 hour ago, Chic Aeon said:

So TESTING inworld is a very important step.

Yes!

Ummm... testing on beta grid of course unless you have a ton of L$ to burn.

 

1 hour ago, Chic Aeon said:

Be SURE and test at LOD 2 (or lower).

I'd call aiming for LoD factor 2 lack of ambition.

 

1 hour ago, Chic Aeon said:

I have to say that for the most part the creations I see inworld for sale have opted for higher land impact over bad LODs (to lower land impact). This isn't good practice either but "I" prefer it over lovely things that fall apart a few meters away.

Definitely yes. But there are some that take it too the extreme, over-strengthening the LoD models so much they effectively override the entire LoD system.

I have a downright grotesque example of this. Here's a bar counter I came across a while ago:

I think the numbers speak for themselves. But for any new mesh makers, this is a great example how not to do it.

Edited May 12, 2020 by ChinRey

[Chic Aeon]

I have lots of examples in my store. I have rarely made my own LODs. I know that makes you cringe but I have a fair amount of die-hard fans who seem to think I do a fairly good job.  I know that we have always disagreed on this. 

That I build mainly for LOD2 is of course my choice. I noted however that what most folks would call good for LOD 1.25  is where most of my things fall. I mean do you really need to see that table a quarter of the way across the sim when it will most likely be inside?  

There has to be a meeting place between the criteria the Moles use (and most folks choose to toss their items because of the high land impact) and render quality mesh that falls apart for those not using LOD4.  I have chosen mine. You have chosen yours and the OP and others need to decide for THEMSELVES where they want to be on the graph.   That's always been my position.

 

I don't BUY things that break apart at a few meters at LOD2. That is my choice as a consumer. 

So in my mind there is no perfect answer. 

[GManB]

#define NewbieAlert ON

Just started building models in Maya and the Substance Suite around the beginning of the year. Imported a few in SL and am gradually learning how to create something that looks good and has low LI. However, I know nothing about LoD other than it means Level of Detail (I'm almost certain). It seems from this discussion that LoD is pretty important. I have a couple models, a small wooden jewelry box (LI 3) and a teak stool (LI 13) that fall apart even when cam is just a few meters away unless LoD is set to 4. I have no idea why this is the case. I have other models with very high LI that do not fall apart.

So, could someone point me to any documentation or blogs or other where I can learn about LoD? I am especially intrigued about the statements above that indicate LoD can be done manually. 

G

[Quarrel Kukulcan]

1 hour ago, GManB said:

So, could someone point me to any documentation or blogs or other where I can learn about LoD?

There's a wiki!

it hasn't been updated in eight years.

Probably not that.

[Quarrel Kukulcan]

11 hours ago, ChinRey said:

Well, at the moment the entire LoD system is broken for fitted mesh.

So any mesh object attached to an avatar renders using the most-detailed mesh (if it renders at all)? At least currently?

[ChinRey]

27 minutes ago, Quarrel Kukulcan said:

So any mesh object attached to an avatar renders using the most-detailed mesh (if it renders at all)? At least currently?

If it is fitted mesh, yes. Without going into details, there are at least three known bugs that totally throw all the performance and swap distance calculations off for fitted mesh.

 

46 minutes ago, Quarrel Kukulcan said:

There's a wiki!

it hasn't been updated in eight years.

Probably not that.

Actually, this page isn't half bad: http://wiki.secondlife.com/wiki/Mesh_and_LOD

One thing that surprises me there is this:

Quote

The uploader automatically generates lower LODs suitable for preview or rough import.

So LL's official documentation admits that the auto generated LoD models only are suitable for "preview or rough import" yet LL's own content creators blatantly ignore this. Dare I say they have a little bit of internal communication problems?

[animats]

It's the lowest LOD that usually has to be manually created. The built-in LOD generator isn't that good, but if you ask it to reduce, say, 1000 triangles to 500, it does OK. Reducing 1000 to 10 will produce trash. A useful approach is to have MEDIUM LOD be half the triangles of HIGH, and LOW LOD be half of MEDIUM. Then make a LOWEST LOD by hand.

Here's an example. Some wheels:

HIGH level of detail.

LOWEST level of detail.

The two wheels on the left turn to total junk at the lowest level of detail. The wheel on the right has a lowest level of detail which is simply a flat surface with a picture on it. 25 meters away, at LOD factor 1, that looks fine. This is called "impostoring". When you can do this, it will reduce triangle count and LI more than anything else.

Managing multiple LOD models in Blender is somewhat difficult. Blender 2.82, with "collection", helps. Beq Janus is building a tool for Blender which should help.

[GManB]

Thanks for the replies. The wiki page got me to understand a lot more about LoD. I am still confused, though, about how LI is computed relative to total number of triangles, LoD (whether auto, manual, or with simpler models), absolute size, .... For example this model I pulled off a public domain website has about 68K triangles, it LI, as you can see is sky high at 116, but LoD are all 14. And the model does not change in appearance as the cam moves away. So, it seems, in this instance, LoD just doesn't work as advertised.

I would appreciate pointers or advice on how LI is computed and the contribution of LoD to LI.

Thanks,

G

[Wulfie Reanimator]

4 minutes ago, GManB said:

For example this model I pulled off a public domain website has about 68K triangles, it LI, as you can see is sky high at 116, but LoD are all 14.

It's a linkset. The edit window only shows the main object you have selected, which is the root, not the rest of the model.

Press Ctrl-Shift-E or tick the "Edit Linked" checkbox and select the actual statue.

I also wrote a little tutorial earlier that also explains LI calculations.

 

Edited May 13, 2020 by Wulfie Reanimator

[GManB]

Ok, Thanks for the info on 'Edit Linked'. Read the tutorial and I will experiment some.

 

G

[GManB]

@Wulfie Reanimator I created a cube,  selected it in the Physics section of upload but my physics cost did not change (using a different model from the statue above). This model has a LI of 19.5 but a physics cost of 14.5. Using the cube does not change that.

Any ideas?

Thanks,

G

[Wulfie Reanimator]

12 minutes ago, GManB said:

@Wulfie Reanimator I created a cube,  selected it in the Physics section of upload but my physics cost did not change (using a different model from the statue above). This model has a LI of 19.5 but a physics cost of 14.5. Using the cube does not change that.

Any ideas?

I assume the model you're trying to upload has multiple parts to it. The process of importing a set of physics models and LODs is a bit complicated so I would strongly suggest that you either upload the model in pieces or join it all together as a single object before exporting from Blender.

The LI is higher than the physics cost because the download weight is higher.

Edited May 13, 2020 by Wulfie Reanimator

[GManB]

Got it!

Thanks,

G

[Extrude Ragu]

One of the nice things about Blender 2.8 is the ability to rotate normals of a mesh.

By rotating normals on your lod models, you can make the shading more closely match the higher detail model, which can significantly reduce visual pop-in as the viewer switches between lod models. 

[GManB]

Thanks Extrude! Do you know if something similar can be done in Maya?

Thanks,
G

[OptimoMaximo]

4 hours ago, GManB said:

Thanks Extrude! Do you know if something similar can be done in Maya?

Yes it can be done in Maya, since ages 😁 Mesh Display menu, edit vertex normals

Eta: actually, you can copy vertex normal orientation from a mesh to another using the transfer attribute tool in Mesh menu. It will average vertex normals of the receiving object to resemble the original, regardless of geometry/topology differences

Edited May 14, 2020 by OptimoMaximo

[GManB]

Thanks!

 

Sooooo many new things to learn. I am having a great time!

G

[Kyrah Abattoir]

On 5/14/2020 at 8:59 PM, Extrude Ragu said:

One of the nice things about Blender 2.8 is the ability to rotate normals of a mesh.

By rotating normals on your lod models, you can make the shading more closely match the higher detail model, which can significantly reduce visual pop-in as the viewer switches between lod models. 

DataTransfer existed in earlier blender versions too.