Jump to content

Mesh parts simply disappearing - small detailed items


Codex Alpha
 Share

You are about to reply to a thread that has been inactive for 3792 days.

Please take a moment to consider if this thread is worth bumping.

Recommended Posts

Detailed mesh parts under 1m seem to like to disappear only meters away.

This has happened to a record crate of mine, and now an extending light's potlight head, as seen in the image below.

LOD-problems.png

Of course if I make the potlight big, like around 2m or so, then it doesn't seem to blink out of existence.

Meanwhile, using a simple, low 'no' detail potlight (cylinder) obviously has no problems rendering to the viewer at ANY distance.

Why is it as soon as I make something a mesh, suddenly it doesn't want to display, even when tests I have made have been simply uploading a cube or cylinder mesh and noticing they do the same thing! Same faces, same vertices, what gives?

Anyways, this seems to happen to objects I have designed (with more detail) that may only end up being 0.5m to 1m, eg. scale record crate, light fixtures, etc.

Should I not add so much detail to small items in 2nd LIfe (not that they are really)? Should I reserve detail for larger items only?

Even on a recent 3m 'spider' model uploaded, fangs that were sized below 0.5m, simply disappear at 5m or more, yet the rest remains, because they parts are bigger..

So let's see if bumping up the LODs will help;

LOD-problems2.png

Nope.

But now we got a whopping 8.76 LI, and still with no results. I can build this light in fewer prims and dont have this issue, which defeats the point of making a mesh (for extra detail for less LI).

Secondly, why don't the arms and base disappear?
Is it something I'm doing in Blender? As far as I know it's all clean, though I have used Modifier->Boolean->Union to create the shapes of the pivot points.


Now let's see what happens on a recent "Record Crate" model I made in Blender. Previously I just used a texture on a cube, and added a 'record' by using a 2nd prim, put both to 'no physics' or 'convex hull' and linked set is 1 Li. But at any size at least, there is no loss of LOD.

As in this image, my crate is unaffected by range as much as a larger item, but at scale, it blinks out. Also, i have observed and may admit, that at a 'scale' size of the record crate, perhaps a model isn't really needed at all.

LOD-problems3.png

 

What do you think I should do? It seems that my smaller models not only don't really need the detail for their size (But what if I created jewelry, that did?)

Is mesh settings of high, medium and low, etc not as relevant on small or 'tiny' or 'scale' models? Its an important thing to know before wasting time modeling detailed meshes for small scales.

What do you expert mesh makers think?

 

Link to comment
Share on other sites

On small items, the lowest LOD will have the biggest impact in reducing the objects land impact. Medium and Low LOD will have very little effect in the download weight calculation. Due to their small size they will switch to lower LODs really quick. Even quicker than sculpts, or some of the curved legacy prims of the same size.

So the most important LOD model will be the one for the lowest LOD, on these tiny objects.
If you build the LODs all on your own, you have full control over how things will look, even at a distance.

Below is an image with 4 LOD models and the collision shape model of some binoculars I made.

Further down are some in-world images of it. The last image shows it at it's lowest LOD. (Red square)

aR_Coin_Binocs_LOD_02.jpg

aR_Coin_Binocs_02.jpg

aR_Coin_Binocs_03.jpg

Also, sometimes objects just disappear from view when moving the camera at some angles/distances. That's more of a viewer glitch though, rather than an issue with the models.

How the LOD models affect the download weights is described here:

http://community.secondlife.com/t5/Mesh/Download-weight-and-size-Some-graphs/m-p/1057163#M5561

There is also a strange issue with certain meshes, which will switch to lower LODs ealier than others. Details here:

http://community.secondlife.com/t5/Mesh/Strange-LOD-transitions/td-p/1305609

 

  • Like 2
Link to comment
Share on other sites

For your 0.625 cube, the expected LOD switches are about 2.25 (hi->med) , 9 (med->low) and 18 (low->lowest) metres. If you use auto LOD, the first switch will be to a triangle (per material), which looks like what you are seeing. There is no need for using LODs at all for a sinple cube which has LI = 1 (0.5 if linked)  even at 64m. So for that you should set "Use LOD above" for all LODs.

More generally, the autoLOD generator is unsatisfactory for anything but organic and other highly subdivided models. It doesn' know what the object is supposed to be and will susually make inappropriate choices in geometry reduction. You are always better off making your own LOD meshes.

It looks as if you are uploading the lamp as a collection of objects. If that is the case, it ends up as a linkset in which each linked object behaves independently as far as LOD is concerned. As they are different sizes, they switch at different distances, and for the smaller ones that can be very closer than you expect from the overall size. Furthermore, the LOD generator works on the triangle count of the whole scene, so that some parts might get completely decimated while others are left alone. You can get much more consistent LOD bahavior by joining the objects into a single mesh before uploading. Since it will thejn behave as a single object with the overall combined size, it will all swutch at once. However, the increased size will mean an increase in LI for the same geometry at each LOD as you used with the separate objects. You can control this by selectively removing the small details in explicitly made LOD meshes.

As I said in the other thread, using autoLOD with zero triangle counts in the low and/or lowest LODs is a bad idea unless you are sure the object is never going to be seen from beyong the relevant LOD switch distances, or unless your objective is to mak it disappear.

  • Like 1
Link to comment
Share on other sites


Drongle McMahon wrote:

"There is also a strange issue with certain meshes, which will switch to lower LODs ealier than others."

Was that never fixed? I can't see the jira - it was removed from the public list (link->"forbidden"!). I'll ask at the cc/mesh group.

I doubt it's fixed. I have to admit though, I haven't tested it by myself.

What I do know for sure is, that I get still different download weight results, with some meshes, depending on the orientation of the mesh. Like, if I make the longest side pointing Y+, or Z+, the download weight is higher than with the longest side pointing X+. It's only in the 0.5, or so range. But that's enough to push a model from 2 LI to 3 LI sometimes. That's with identical LOD meshes. same hard edges, same materials etc..

Anyway, would be great if you share the answer from the CC/Mesh group.

  • Like 1
Link to comment
Share on other sites

Wow, thanks, just how I suspected, but didn't have the knowledge to confirm that. I will experiment with different LOD models and see what happens.

I understand now why the smaller mesh will switch quickly to lower LODs other than larger objects. So am I on the right track to not really include too much detail for smaller meshes, and save it for larger ones? Eg. stick with basic cube prims in-game, and/or keep detail lower for 0.5m type objects (I'm building to default avatar/ RL scale, so a record crate is about 0.5m-0.625m).

I guess I could continue with the high detail at any size big or small, but EXPECT that the lower level LOD meshes will be preferred quickly. What do you think?

Link to comment
Share on other sites

Thanks for the actual distances involved in LOD, hard to find that info about  for mesh beginners to account for. P


Yes, I uploaded the lamp as a colllection of objects, as I did with a spider model as well, so it could be poseable into different positions 'in-game'.

If I join the objects into a single mesh, will I lose that ability?

Ok, so if I want to make individual pieces of a 'mesh' positionable, etc, then consider them as seperate meshes anyway and create LOD models (Or account for each individually) instead? 

Thanks for info, a great help.

Link to comment
Share on other sites

Yes, if you want to move them relative to each other, then they have to be separate meshes, but you can still combine some, such as a hinge and one of the arms it's attached to - so that the tiny hinge won't switch too soon. When you are mixing things of very different sizes, then you can compensate to some extent by using more "use LOD avove" for the smaller ones. Still, the most important thing you can do is not to use the automatic LODs. Making your own lets you control exactly what happens. This is especially true if you are uploading the parts together.

I don't follow what you are sating about iusing less detail on smaller meshes. For meshes less than 1m, the "lowest" LOD has by far the greatest effect on the LI. You can have plenty of detail at the higher LODs without incurring high download costs as long as the lowest is very simple. What kind of physics are you using? The physics costs of small objects can get very high if it is triangle-based (specified but not "Analyzed"). You can see if its' the physics cost that's making the high LI by using the "More Info" link.

Link to comment
Share on other sites

With really small meshes you can get still a very low land impact, with thousands of triangles in the High, Medium, and Low LOD. As long as the lowest LOD has only a few triangles left. So making that Lowest LOD very lowpoly, but still look good enough to be pretty from a distance, is the key. Also, not everything has to be 1 prim/land Impact. Like with the binoculars. I could easily make the lowest LODs even lower, and have it at 2, or even 1 LI instead of 3. But it wouldn't look as good in the LOD transitions as it is now. And I think 3 LI is still a nice number for that thing.

With bigger meshes, the lowest LOD will have little impact in the download weight calculation, so your Low, and Medium LOD should be reduced more drastically already, to keep the land impact low. And if it's really big, so that the Medium LOD, Low LOD, and Lowest LOD, won't have any impact in the download weight calculation, your High LOD mesh should be very very lowpoly, to keep the land impact reasonable. And details should be added as smaller separate meshes.

As a general rule, reduce the triangle count of every level of detail model as far as you can, without destroying the model. You can't see all these very little details from a distance anyway. And having the lowest LOD  just represent the silhouette of the model, is good enough to make it look good from all distances, in most cases.

To achieve all that, it requires some practice, like with everything in life.

  • Like 2
Link to comment
Share on other sites

As for the LOD switch distances of multi piece meshes. To make the LOD transitions occur at the same same distance on both pieces of the binoculars, I have added a invisible triangle to the top part, to make it's bounding box as big as the bounding box of the lower part. Certainly, this increases the download weight slightly on the top part, but hey, first of all it should look pretty. :matte-motes-sunglasses-3:

Link to comment
Share on other sites


entity0x wrote:

...

So let's see if bumping up the LODs will help;

LOD-problems2.png

Nope.

But now we got a whopping 8.76 LI, and still with no results.

...

  1. First of all, stop using the lowest  LoD for your Physics! Doing so means your effective LI will most likely be driven by your Physics Weight, a situation only desirable in architectural models such as walls, where a detailed Physics may actually be required. A Physics model should be designed separately with its own collada file and kept to the least detail needed. In this case, where your model is simply a prop that will not be sat or stood upon, a simple cube or even a plane would suffice.
  2. Second, your Server Weight of 1.50 says you're using three distinct meshes for this (each distinct mesh has a default of 0.50). As Drongle pointed out above, if the bounding boxes of each of these 3 meshes differ in size, the distances will differ on their LoD's as well. When uploading a set of meshes such as this, having bounding boxes the same size throughout is advantageous.
  3. Third, your picture shows a Download Weight of only 0.487. Download weight is, without exception, directly set by your LoD's and, in this case, this is ridiculously low since it could go as high as 2.49 before it would cause your effective LI to go to above a mere 2 if it was the deciding factor.

 

As it is, I can see the advantage of having three meshes in this specific object  in that this allows you to position the light and boom angle/position. However, this will always result in a Server Weight of 1.50 which means the lowest effective LI you could have is 2. To maintain that, as well as obtain a properly viewable model, you simply require upload models made of each element of the equation and stop relying on upload configurations to do this work for you!

 

At the bare minimum, you need- An upload model for the Highest LoD (which may also be used for the Medium LoD) as well as a Low, Lowest and Physics models prepared for upload (no extra charge btw when you do it this way, just a bit more time/effort/thought on your part).

 

Relying on the uploader configuration settings to attempt to do this work will always, without exception, give poor results since you are giving up control when you use that approach.

 

[Edited for emphasis.]

  • Like 2
Link to comment
Share on other sites

Arton: "There is also a strange issue with certain meshes, which will switch to lower LODs ealier than others."

Drongle: Was that never fixed? I can't see the jira - it was removed from the public list (link->"forbidden"!). I'll ask at the cc/mesh group.

Arton: I doubt it's fixed. I have to admit though, I haven't tested it by myself.


I asked - it isn't fixed. Or, at least, the internal jira is still open with no fix mentioned.

Link to comment
Share on other sites

Good words, thanks for the help. I'll get back to the topic on my next test model to show what happens.

Now for each level of detail model, one still UV wraps and texture them individually right? So I can use the same blend file, but work on different object details?

Link to comment
Share on other sites


entity0x wrote:

Good words, thanks for the help. I'll get back to the topic on my next test model to show what happens.

Now for each level of detail model, one still UV wraps and texture them individually right? So I can use the same blend file, but work on different object details?

Right. Each LoD model has to have the same number of texture faces but on a lower LoD a face could be just a solitary triangle. Each model is separately UV-mapped but this mapping need be only done once if you're duplicating your model as I describe in a bit.

 

And, yes, only one .blend file is used, all your work stays together that way w/o any confusion.

 

There are different work flows to do this and I'll only describe what I've found works best for me. Experiment to find your own if you feel like it, I don't think there's a "wrong" way to go about it as long as the end result is what you want.

 

On Layer 1, I'll create my highest LoD model and I'll spend the greatest time on it. I'll do its geometry as detailed as I feel is necessary, assign the materials (which will become the texture faces inworld), UV-map it and (possibly) bake textures.

 

Once I feel that's good to go I'll duplicate it in Object Mode 5 times, placing an exact copy of it on different Layers. Layer 1 stays as is and I'll use that model for my highest LoD, Layer 2 for medium LoD (which I rarely need, having found that choosing the Use LoD above option in the uploader usually suits my needs. But, every once in awhile I'll need all five models after all and, if that's the case, it's there for me to work on.)

 

Layers 3 and 4 is where I'll start simplifying the geometry. This is just a matter of removing vertices and faces. I want to preserve my overall outline as much as possible but define it less strongly. For instance, if I have a cylinder using 32 sides on Layer 1, it may have 12 on Layer 3 and only 5 on Layer 4, which will be my lowest LoD. If I have (say) rivets defined on the model, they'll be flattened smooth on the lower LoD's and I'll let the texture itself define them if needed.

 

Generally I try to reduce Layer 3's vertices to no more than one third of Layer 1 and Layer 4 one third again. So, if my original model had 9000 vertices, Layer 3 would have 3000 max and Layer 4 1000. But that's no hard and fast rule and each model is unique. The main idea is to examine it and decide what can you get rid of or simplify at each level, keeping in mind it will be seen from a greater distance than the model above it.

 

The main point of all this is that doing it this way puts you in control of what is removed. Having the uploader decide for you is, at best, a crap shoot. It really has no idea what your model is supposed to be or how you need it to appear at 20 meters. It'll just decimate your work willy nilly and your great looking window frame may end up looking like a bad hairdo at a distance.

 

OK, now let's look at Layer 5, where you'll have your Physics model. Personally, I never use the model I've duplicated from Layer 1 except for sizing and placement of plane meshes. Some others I know use cubes, I find I get the results I need with planes. Just sayin'.

 

Some things to keep in mind about your physics model:

  • Material faces aren't considered, you can and should ignore them entirely.
  • Curved physical faces will eat you up! If your walls of your spaceship have gently curved corners, do the physics with, at most, one angled flat plane across that point.
  • Keep things flat, broad and only where necessary. Don't model the physics for each chair leg, stick a cube there to sit on.
  • Physics Weight can build very quickly if you're not making every surface you add to define it count. Skip surfaces if you can. Like ceilings, let the floor above or the rooftop define that.

Like I said, I use planes to define my surfaces, Adding and positioning them as needed. Once I'm satisfied with the arrangement, I'll delete the duplicated model I used for their placement from that layer. A single triangle may be all that's left, but it'll be a triangle I've placed exactly as I feel is needed.

 

At this point, I can export to my collada files. I go to Layer 1, select the object(s) there and export for SL with the selected only option. I'll name that file MyModelHigh.dae. Third Layer the same but named MyModelLow.dae, fourth is MyModelLowest.dae and you can guess what my fifth layer is named. Whole thing is saved as MyModel.blend into a project directory that I'll keep my textures and such.

 

Then it's off to the beta grid to start testing it out. I'll choose the Load from file option everywhere but on the Medium LoD, which I'll (attempt to!) use Use LoD above. I'll have the uploader Analyze but not Simplify my physics model (some prefer to skip the analyze bit). I'll make note of the numbers across the bottom before I upload, checking to see which of Physics, Download or Server Weights is highest, because that will decide my final Land Impact. And, unless it's the Server Weight (which will always be 0.50 for each mesh element), I might see an area of improvement- possibly simplifying my lower LoDs even more or moving some geometry from the model onto the texture, such as one might do in the case of a fluted column.

 

After I've uploaded it, I apply its textures and then check it out. Anything I'm not satisfied with, it's back to my blend file to correct or tweak. Wash, rinse, repeat.

 

And that's basically it. One doesn't need to spend all that much time on these extra models either. In fact preparing them can go fairly quickly after what's needed is figured out. But doing it that way will give you complete control over the process and, I'm certain, a better result every time.

  • Like 1
Link to comment
Share on other sites

Just a couple of notes on LeprKhaun's very complete post...

"I'll have the uploader Analyze but not Simplify my physics model (some prefer to skip the analyze bit)."

It may be useful to understand the distinction here. If you don't "Analyze", you will have a triangle-based physics shape. That is to say, the physics engine will see the object a the collection of triangles that make up the physics mesh. If you do Analyze, the uploader will generate a set of convex hulls (solids with no indetations) to represent the shape. It is usually more efficient for the physics engine that way. There are two big differences in the bahavior of these type of physics shape.

First, the Analyzed shape has a constant physics weight independent of the size of the object, but the triangle-based shape's weight is very dependent on size. The physics engine hates small, narrow triangles. So the weight increases markedly as the object gets smaller. The result is that triangle based shapes are never a good idea for small and medium sized objects. However, for large objects, such as large buildings and walkable lanscapes, the triangle based shape may have a lower physics weight than the functionally equivalent Analyzed shape.

However, and secondly, the physics engine knows that the convex hulls of the Analyzed shape are solid. So if you get inside one, it will gently nudge you back out. This is not true for the insides of parts of a triangle based shape, So if you get stuck inside one of these, for example, inside a wall, then you are stuck. (Actually it's usually easy to get out by sitting on something out there, but not always).

""I'll make note of the numbers across the bottom before I upload, checking to see ...

The physics shape you have specified is only used when you set the physics shape type to "Prim" in the features tab of the edit dialog inworld. Until you do that, a default physics shape is used. This is a single convex hull made for the whole mesh (as opposed to the multiple hulls made by "Analyze"). This is what gets used with the default physics shape type of "Convex hull".  If you specified a physics mesh, it will be the convex hull of that mesh. If you didn't, it will be the convex hull of the low LOD mesh*, and there will be no "Prim" type option.

Unfortunately, the physics weight shown at the bottom of the uploader is the weight of the default convex hull, and can be very different from the weight you will see when you switch to type "Prim". So sometimes, although you checked that weight, it may jump to a vale higher than the download weight when you switch to type "Prim". That is just one of the reasons to always test you mesh on the beta grid, as LepreKhaun suggested.

One other thing to remember when making your extra meshes, LOD and physics, is that they will get stretched so that their bounding box fits that of the high LOD mesh. So it is important to make sure they have the same bouding box if you want them to fit properly.

Finally, if you want to upload multiple objects together, with physics mesh, you need to have a physics mesh for each of the high LOD objects. If you only supply one mesh, it will be squeezed to fit just the first object. If you stick to a naming scheme like LepreKhaun's, you should have no trouble with getting the right physics associated with the right LOD mesh (using Blender and the "Sort by object name" export option). All that applies to the lower LOD meshes too.

*This has caused consternation to people who use a single triangle as low LOD to get a low LI and then wonder why their physics is just a triangle.

PS - L's advice not to use "Simplify" is also very right. Simplify is a bag or worms, especially if you are making something hollow with openings. If the physics weight is too high because the number of hulls is to large, or they have too many vertices, simplify the physics mesh yourself instead.

  • Like 1
Link to comment
Share on other sites

Ok, I'm following this right now. So each level is going to have it's own UV Map, it's own texture image as well?

For example, Level 1 will have its UV Map image, then perhaps Level 3 will be UV'd a bit different, but I would need to bake out another image for that, and etc for Level 2 or less if needed?

So I would expect to upload LOD models for most levels, and each image for each LOD would also needed to be uploaded for each? Not sure about how this aspect works.

I will keep the same 2 materials I'm using, but my model will have less vertices and faces, but still maintaining its shape and size, etc

Then UV Map again, and bake a new image for each?

Link to comment
Share on other sites

Most often you can use the same texture for the LOD meshes. You just have to use UV-map-conserving edits when you reduce geometry for lower LODs - Dissolve is your friend there, and some others.

It's best to avoid using different textures for two reasons. First, performance will suffer from using extra textures. Second, you have to have all the textures on at least one triangle in every LOD. You can use hidden triangles at the LODs where a texture isn't used, but you are still using four of your eight available slots for each SL face, limiting you to two. The extra triangles may also significantly raise the LI in certain cases. It is the case that some textures, especially baked normal maps, aren't going to be exactly right when they are applied to the simpler geometry at lower LODs, but that's usually at least as acceptable as the geometry reduction itself.

On the other hand, using an alpha texture on a simple plane or box at the lowest LOD is often the best way to get low LI, for things like windows, railings and slatted chair backs. In that case, you have to have use the hidden triangle trick, but you only have one extra texture. So it's less wasteful than using a different texture at every LOD.

 

Link to comment
Share on other sites

This is going to be a bit mind boggling. Blender can be a real mess when you click the wrong button at the wrong time and don't even know it. Trying to figure out why each layer object needs to have its own new material, and assigning the same image gives different results.

I'll have to get back to you after seeing what the problem is.

Other than that, getting the lower LODs is working out fine. Mapping them to the original baked texture is a pain in the ass though, and doesn't give very nice results.

Capture.JPG

Now I'm just lost.

UV Map this, UV Baked that. Material uses this texture, then on new model Material disappears have to make a new one, it says 2 items are using the image, but now there's a copy...

Might be better off just creating a new blend file for each edition, though I do like your way.. but the materials/textures/UVMaps get confusing in the same file for a newbie like me.

Link to comment
Share on other sites

I'm not sure I follow exactly what you are doing. The way I do it is thus: (1) make high LOD (2) apply and bake texture - this is the one texture to be applied to all LODs. (3) duplicate for next LOD and remove lots of geometry using dissolve where possible - texture should remain correct - adjust UV map manually if needed. DO NOT BAKE AGAIN. (4) repeat  step 3 for low and again for lowest LODs. There are variations, such as making the but this is the main process.

The point is that you only make one baked texture, using the high LOD mesh. Then the UV maps of the lower LODs have to fit that. You shouldn't have to re-uv map the lower LODs at all, but you might have to if you use some destructive editing to reduce geometry. If you do have to, I believe you can "Pin" the vertices that are in the right places first, so that they don't move. Make sure the correct texture image is selected before you remap. The (equivalent parts of the) meshes at all LODs MUST be using the same material because they will use the same texture in SL.

When done, select each LOD model in turn and export it with "Export selected" checked, as xxxx_hi, xxxx_med, etc.

I keep all the LODs on the same layer, just moving them apart in object mode (object mode translation is ignored by the uploader). Others prefer to keep them on separate layers.

 

  • Like 1
Link to comment
Share on other sites


Drongle McMahon wrote:

I'm not sure I follow exactly what you are doing. The way I do it is thus: (1) make high LOD (2) apply and bake texture - this is the one texture to be applied to all LODs. (3) duplicate for next LOD and remove lots of geometry using dissolve where possible - texture should remain correct - adjust UV map manually if needed. DO NOT BAKE AGAIN. (4) repeat  step 3 for low and again for lowest LODs. There are variations, such as making the but this is the main process.

Understood.


Drongle McMahon wrote:

The point is that you only make one baked texture, using the high LOD mesh. Then the UV maps of the lower LODs have to fit that. You shouldn't have to re-uv map the lower LODs at all, but you might have to if you use some destructive editing to reduce geometry. If you do have to, I believe you can "Pin" the vertices that are in the right places first, so that they don't move. Make sure the correct texture image is selected before you remap. The (equivalent parts of the) meshes at all LODs MUST be using the same material because they will use the same texture in SL.

My mistake so far is using destructive methods. I wasn't using dissolve, I was deleting the faces, creating a new face, then wondering why it didnt' fit my UV map anymore. That has been corrected, and now use dissolve. However, this introduces another hurdle to overcome, or at least what to watch for when rearranging the UV Map before baking..

If the individual sections are not kept together when the UV Maps are re-arranged (to scale things up that deserve detail, etc), such as this front face of the couch arm, if dissolve is used, this can happen; which certainly won't help with keeping the texture intact.

Capture.JPG

(faces that were initially Uvmapped side to side, were moved apart by accident when manually repacking/scaling uv items...) Kind of hard to not do that though, since some shapes cannot be manually associated with their UV mapped counterparts.


Drongle McMahon wrote:

When done, select each LOD model in turn and export it with "Export selected" checked, as xxxx_hi, xxxx_med, etc.

I keep all the LODs on the same layer, just moving them apart in object mode (object mode translation is ignored by the uploader). Others prefer to keep them on separate layers.

 

Understood. My next challenge is scaling these models down with dissolve, without having any UV mapping issues, as shown in the image above.

Continued:

Ok, so I've learned that better initial UV layouts will help later when simplifying. The parts of my couch that were the easiest to dissolve were mainly the back and seat cushions, as I had unwrapped them nicely, whereas others not so nice.

I managed to deal though, and came up with these models;

couch levels.png

I ended up making a simple 4 plane box for the physics model.

I used the Hi for both High and Medium setting, since I hate how it switches to the Medium model almost immediately like 3+meters away. Linden Labs should just make this one setting IMO, at least at this scale it seems pointless because who wants to lose all the detail and work they put into something just because the avatar moves 3m away from it?

I skipped my 'low' model even though it is 1/3 of the medium model because the Li was 35 prims for that. So I used my 'lowest' model for low, and just let the uploader choose my low setting, which at least holds the basic shape from viewing too far away. This last setting with the lowest seems to be the defining factor for upload weights and such.

After a break, I could go back to my models and experiment some more, but this is my results so far anyway.

Link to comment
Share on other sites

My guess is thjat you dissovled an edge that was UV seam, leaving the new face spanning across two separate UV islnds.  I should have mentioned that. Ideally you need to plan the seams in advance so that you can leave them all there, undissolved in the lower LODs. It's not as difficult as it sound. Just keep the seams at the major edges that are going to be needed at all the LODs. When you can't, or you made a mistake with that, then you can pull the vertices around on the UV map after you add the nes faces, until they are right. That's tricky though. It's easier if you use a good test texture like this.

tgrid.png

  • Like 1
Link to comment
Share on other sites

You are about to reply to a thread that has been inactive for 3792 days.

Please take a moment to consider if this thread is worth bumping.

Please sign in to comment

You will be able to leave a comment after signing in



Sign In Now
 Share

×
×
  • Create New...