best practices for necklace chains?

[Quarrel Kukulcan]

Sculpting individual links seems to run anywhere from annoyingly to prohibitively expensive. Even a "low-poly" trial mesh I made where each link is a hexagon with triangular cross-section (so 6x3x2 = 36 triangles), untextured, generates an upload with a LI over 19 and an upload cost over $75L.

I've found cylindrical cords to be lower (and the LODs are more straightforward) and great for a lot of things, but you can't do a linked look.

I tried a track of flat panels as well, with an alpha-blended chain texture painted on. That was the lowest complexity-wise, and you can use a tiny texture that's identical on every quad, but looks bad from viewing angles that catch it edge-on, plus it's extra-conspicuous when it doesn't track exactly against the body.

[Chic Aeon]

Possibly your upload costs are contingent on your lower LODs?   You can attach and invisible prim to the necklace in the form of a big cube that will "change" the size of the object and help with the LODs.  You can also use four teeny tiny cubes (or even stray vertices) to preform the same task. 

 

 If you have a high value for the lowest LOD that might be your problem. I haven't made that many necklaces but mine came in at $11 and had good long range viewing, so this is a "fixable" issue.  Land impact was one or two. 

 

 

[Wulfie Reanimator]

Gettings things like chains to be "low LI" is extremely difficult, due to the topology chains normally* require. Primarily because you just can't make the lowest LOD look good with mesh alone -- and not skyrocket the download cost. For small objects (go to Upload Options and make sure the dimensions are about the same as the expected result), you really really REALLY need to focus on the lowest LOD, since that's technically the most prominent LOD according to SL's algorithm.

Eg: 4968 tris

Lowest LOD: Link segments 2-3 times wider with only one "surface" per link, 672 tris

Uploading this at 0.5 x 0.5 x 0.2 scale costs 51 L$ with 39.792 download weight. Obviously no positives there.

Removing half the links and the inward-half of the rest: 23 L$ with 12.234 download weight, 156 tris

To show how much the lowest LOD matters, look at my higher LODs, which are just the first LOD minus one "surface" of each link:

But really, I would recommend making just a simple, non-segmented ring "surface" and use a veeeery low-res texture on it, or even leave it solid-color. It's the lowest LOD after all, it's not going to be very visible while in use. Even for the higher LODs, you probably shouldn't be using full mesh, my example is an exaggeration. And if we're being realistic about the current state of SL, the lowest LOD is basically never going to show up for avatar attachments so the quality-importance drops even further.

Edited July 29, 2020 by Wulfie Reanimator

[ChinRey]

1 hour ago, Wulfie Reanimator said:

Gettings things like chains to be "low LI" is extremely difficult, due to the topology chains normally* require. Primarily because you just can't make the lowest LOD look good with mesh alone -- and not skyrocket the download cost.

I'm not usually in favor of zeroing out LoD models but for a chain as thin as the one in Wulffie's it's worth a try since it probably won't be visible at "lowest LoD" distances anyway. But don't just set the triangle limit to 0 in the uploader. Instead, make a simple model with three or four small triangles defining the out limits of the necklace's bounding box.

Wulffie's 156 tri example looks like a good low LoD model to me. It can be reduced down to less than 100 but it wouldn't make much difference and would require a bti of retopo work so it may not be worth the effort.

I probably wouldn't bother making a mid LoD model but simply use the same as the high. There isn't much to gain by simplifying here and it would be quite a bit of work to get a good result.

For the high LoD model:

9 hours ago, Quarrel Kukulcan said:

Even a "low-poly" trial mesh I made where each link is a hexagon with triangular cross-section (so 6x3x2 = 36 triangles),

You have to use two different curve resolutions. For the cross section, that is around the circumference of the thread, you don't need more than six vertices, maybe only five (I assume you're using smooth vertices btw). But you need a bit more along the bends of the thread, at least four vertices along each half circle, possibly five or six.

You can also save quite a bit by manually deleting all the triangles that are hidden by other links in the chain. That's quite a bit of work though and it's up to you whether you'd bother with it or not.

[Quarrel Kukulcan]

On 7/29/2020 at 4:16 AM, ChinRey said:

You have to use two different curve resolutions. For the cross section, that is around the circumference of the thread, you don't need more than six vertices, maybe only five (I assume you're using smooth vertices btw). But you need a bit more along the bends of the thread, at least four vertices along each half circle, possibly five or six.

It sounds like you're talking about making a columnar tube here. I'm talking about individual links. Here's one of them.

[ChinRey]

1 hour ago, Quarrel Kukulcan said:

I'm talking about individual links.

That's what I was talking about too, individual links like this:

This means 168 triangles for each link in the chain but that is actually not a problem. For something as small as a necklace the high and mid LoD models add very little to both to all those different numbers (upload fee, weights and LI) and to the actual load so you can afford to fill them up with lovely details.

You have to be ready to spend soem time and effort on the low and especially the lowest LoD models though because they make all the difference. Hopefully me and Wulffie were able to give you some good tips how to do it.

Oh, and you also have to consider the physics model. That's very easy for anybody who knows how to make SL mesh but not obvious for those who don't: One single triangle is all you need here and for all other meshes that don't actually need a physics model. (Good tip for all SL mesh makers: make two very simple meshes, one just a 4 vertice/12 tris cube and one even simpler with only a single triangle. Save them as dae files and store them for the future. Use the single triangle for objects that don't need physics and the cube for objects where you only need physics to keep avatars from walking straight through them. That covers the vast majority of meshes in SL.)

Edited July 30, 2020 by ChinRey

[Quarrel Kukulcan]

12 minutes ago, ChinRey said:

You have to be ready to spend soem time and effort on the low and especially the lowest LoD models though because they make all the difference.

Yeah, I see jewelry quickly shrinks to near-invisible. I'm down to LI 4.04 and $15 upload cost if I make the links in the Med. LOD simple, solid boxes and set the Low and Lowest to a flat strip. I already had a trivial shape for the unnecessary physics mesh.

Thanks for the pointers. I hadn't realized the lower LODs affected the cost so much.

[ChinRey]

1 minute ago, Quarrel Kukulcan said:

Yeah, I see jewelry quickly shrinks to near-invisible. I'm down to LI 4.04 and $15 upload cost if I make the links in the Med. LOD simple, solid boxes and set the Low and Lowest to a flat strip.

That's relly well done!

You should be able to save even more if you reduce the lowest LoD model down to four tiny triangles as I suggested in my first post. Assuming it's a really thin necklace like the one in Wulffie's illustrations, it won't really be visible at "lowest LoD distances" anyway.

 

4 minutes ago, Quarrel Kukulcan said:

 I hadn't realized the lower LODs affected the cost so much.

It depends on the size of the object. For small objects it's the lower LoD models that count, for big ones it's the higher ones.

Since I'm posting anotehr reply here anyway, I've never made a necklace and I'm not going to make one just for this thread but look at this sign:

Here's a closeup of the chainlinks:

This sign, bracket and chains included, is only 1 LI and the upload fee was 11 L$ (or possibly 12, this is a four years old build so I don't remember exactly) yet even with LoD factor set to 1.0 it looks good at any view distance with no noticeable distortions or abrupt switches between LoD models. This isn't for beginners of course. Something like this takes a lot of skills and experence I wouldn't be surprised if it turned out to be less than a dozen people who would be able to make it. But no matter what skill level we are at, there is always room for improvement and the big sevret is to understand where it's posible to save.

[Kyrah Abattoir]

To be fair, for thin jewelry-type chains, the BEST practice would be to make your mesh a flat polygon strip and to make an alpha masked, normal mapped chain texture repeat along it.

I don't have an example on hand, but a friend of mine once showed me one that they made, and It was so convincing that I couldn't figure out that they did without them telling me.

(I was actually gonna start lecturing them that it looked a bit too high poly when they smirked at me and told me it was just clever normal mapping)

Edited August 1, 2020 by Kyrah Abattoir

[ChinRey]

1 hour ago, Kyrah Abattoir said:

To be fair, for thin jewelry-type chains, the BEST practice would be to make your mesh a flat polygon strip and to make an alpha masked, normal mapped chain texture repeat along it.

That's an interesting idea but although this definitely is the best solution for fitted mesh, it probably isn't for rigid mesh.

I'm not sure what the overall size of this mesh is but if we assume 0.25x0.2x0.1 m, which seems reasonable, the first theoretical LoD swap distance will be 0.56 m, the second 2.23 m. That means the polycount for the high and mid LoD models will only have a marginal impact on the calculated wiehgts and LI and and under normal circumstances none at all on actual performance. Normal maps on the other hand, are not culled by the LoD system and will always add a little bit of extra load. Not much but in cases like this almost certainly more than some extra polys in the high and probably also mid LoD models.

There is a common misunderstanding that subsituting normal maps for "real" geometry always improves performance and I don't believe the impact this misunderstanding has on the overall performance of SL is insignificant.

---

I'm glad you mentioned texture repeats btw. When it comes to actual performance, the texture is far more significant than geometry here and the only really good solution in this case is to use a repeating texture, no more than 256x256 resolution (and even that may be too much).

Edited August 1, 2020 by ChinRey

[Kyrah Abattoir]

It all depends how big the map is, to be fair, while the diffuse might need to be 128x128 (or a bit more) to get clean alpha mask borders, the normal map can be lower than that. And the specular map can probably just be the default "blank" texture.