Drongle McMahon

"the maximum number of vertices a single mesh object can have 65,536." If only it could be that simple ... There is code in the uploader that would enforce that limit per material, but it is never reached. Instead, if you have more than 21844 triangles in a material, the uploader will secretly start a new material. That means you can get up to 8x21844 = 174752 triangles. You can upload a model with more, but the extra triangles are simply discarded. If you use up material slots with less triangles, then the available total is reducd accordingly. It;s a very bad idea to have more than 21844 triangles in a material because the secret materials can lead to very bad UV map fragmentation and consequent rapid increases in LI. It would be a very bad idea anyway, for performance reason as you say.

"Somebody's published all this, with graphs..." Here.

Frawmusl, If you use the "Analyze" button with this shape, it should give the same result whether you have five separate boxes, as here, or one with the corners in it. That's because it has to be split into five pieces for each to be a convex hull. So it should be exactly the same weight. However, it is generally easier to control what the uploader does by giving it the right hulls to start with. In principle, as long as these boxes are not tapered - all right angles, they could be treated as primitives by the Havok engine. It does that when you use linked boxes, and they only weight 0.1 each. However, at least for now, the uploader and server are not programmed to recognise primitives in uploaded physics meshes. So it actually makes no difference whether the boxes have angled corners or not. As long as they have flat sides (all four vertices coplanar) and there are eight vertices, all boxes will cost 0.36 each.

"....one boring royal correspondent after another...." Yes. I do feel a bit sorry for the royal correspondents though. They must know in their hearts they have reached the very bottom of the deepest resesses of their profession, with no hope of return to the real journalism they once dreamed of. I'm surprised they manage to keep going at all.

First, can you confirm that both the visual mesh and the physics mesh you had problems with are each (supposed to be) just one Blender object? Assuming that ... The physics shape being bigger than the visual mesh would be very unexpected. Is the bounding box you see when you edit the mesh and check "Stretch" also bigger? If it is, it suggests that you may have accidentally included some extra geometry in the visible mesh. The physics would then get stretched to include that. What do you see in the uploader preview after you press "Analyze"? Does the physics shape there match the visible mesh? Also, what do you see when you look at the physics shape with Develop->Render Metadata->Physics shapes? If you want to try a triangle-based physics shape (that's what you get if specify a mesh but don't use "Analyze"), you should use a mesh avoiding without the narrow rectangles that are edges of walls and windows, as far as possible without changing the bounding box (if you use one plane per wall, you will need the end edges only. If you use your physics mesh, just delete the narrow edges first. Don't use a mesh with the windows. This is because the narrow triangles increase the burden on the physics engine and the physics weight. Some of the cases of unexpected phantom prims have been reported to happen when a mesh is rezzed on top

If you don't tell the uploader to make a physics shape, it will use the default convex hull of the low LOD visible mesh as the physics shape. By definition, you cannot go inside a convex hull. If you give it a physics shape, then set it's physics shape type to "Prim" on the features tab of the edit box, then you may be able to go inside it if you have specified a suitable physics shape. For the simple case you describe, the easiest thing would be to choose the high LOD mesh on the physics tab of the uploader, and then click the analyse button. It should then tell you it has made a shape consisting of three convex hulls with 24 vertices. You can see how they fit by using the slider to spread them apart in the preview. If your shape becomes more complex, it becomes difficult to pruduce a satisfactory physics shape from the LOD meshes without having much too high a physics weight (and therefore high LI). So you will then need to make a physics shape mesh especially for that purpose.

No Problem, not for me. ... I thought the op was using Maya. I could have given Blender instructions.

Help ... someone from Mayaland, please explain how to join two objects into one.

Can't see from your pictures, but this is a poissibility ... You have to make sure the nanamesh and the extra triangle are arranged in such a way that their combined bounding box is large enough in all three dimensions. So, for example, the bottom left arrangement here will not work because one dimension is still too narrow to squash enough. The arrangement above it has the triangle moved perpendicular to the shared plane, making the bounding box larger in that dimension. The one on the right achieves the samee effect by rotating the pair so that the shared plane isn't aligned with the axes. This will not be achieved by rotating the object, as that will rotate the bounding box too. In Blender, this can be assured by Apply Rotation and Scale. I don't know the equivalent in Maya. Of course the whole thing requires the nanomesh and the extra triangle to be part of the same mesh. If you accidentally leave them as two separate objects, they will each have their own bounding boxes, and so the whole vthing fails.

"What would it take to upload a 1 million polygon dress into SL I wonder..." Impossible. :matte-motes-shocked: The limit is supposed to be 65536 vertices, but a bug in the code allows up to 174752 triangles by starting new hidden materials when any reaches 21844 triangles (with multiple materials in the input, the limit will usually be lower). Any more triangles than that and they are simply omitted. In fact, you can run into strange material artefacts at more than 21844 triangles, and, for unattached items, the LI can increase dramatically because of interspersion of triangles with different secret materials (depends on triangle order on input). So there is a sensible practical limit of 21844 triangles per material that everyone should keep in mind. However, for most things, including clothing, getting anywhere near that limit is probably irresponsible.

I generally prefer to upload single objects separately. That's for the same reason I never upload textures with the mesh - if you combine things, you end up paying and waiting to upload the whole thing every time you modify just one part. However, I have done some multiple object uploads, especially to study the problem there was before Gaia % Co, got the "sort objects by name" option in the Blender exporter. Before that I had to edit the collada files by hane to get things in the right order. Now it's very simple. To export multiple object you just select all the objects that belong together at each LOD/physics when in Blender's object mode. Then use "Export selected" and "Sort objects by name"/ Then the appear in the collada file in alphabetical order. To get the right high LOD objects associated with the right low LOD and physics objects, you just have to name them in the corresponsing order. That's easy if you use the same names but with a suffix that says which LOD the belong to; such as "window_hi, door_high. chmney_high....." for the high LOD and "window_med, door_med, chimney_med, ...." for the next LOD, and so on. When you import multiple objects, the importer remembers the relative transformations, rotation and position, of the high LOD objects. It prefers to make them a single linkset. If it can't, it makes them into a coalesced object (with the pile of cubes icon). It doesn't take any notice of the transformations of the other LOD/physics meshes. Instead, they all have the remembered high-LOD transformations applied. That's why they get swapped around if they are not in the right order. Just as with a single object, each lower LOD/physics mesh will be squeezed of stretched to fit the bounding box of the corresponding object in the high LOD file. That also produces strange effects if they are in the wrong order, because they get the wrong bounding box.

Unless an object emits light, all you see is light from other sources reflected from its surface. Diffuse reflection is light that gets scattered equally in all directions, after having its colour changed by the diffuse map . The diffuse map is just the texture we have been used to used before advanced lighting. The amount of diffuse reflection is affected by the angle of the surface to the incident light, but not very strongly. The diffuse map is the old texture. Old textures didn't have any other kinds of reflection except the old shiny. Specular reflection is additional reflection of light from light sources falling directly on smooth surfaces. It only appears when a light source and a camera have the approximately same angle to the normal of the surface. The accuracy of the angle required for reflection depends on how polished the surface is. It is modelled by the glossiness parameter and the map in the alpha channel of the normal map. Its colour and intensity are not affected by the diffuse texture, but by the rgb channels of the specular map instead (as well as the colour selector under the spinners). While specular reflection comes light falling directly on the surface, environmental reflection is an attempt to represent the reflection of indirect light from the sky. It depends much less on the camera angle because it comes from all round instead of from a light source. So it makes the whole surface look wet or polished. It is more or less the same as the old shiny, but with much greater control. At midnight, when the sky is almost completely dark, you don't see the environmental reflection. As far as I can make out, indirect light from light sources you place near an object is not included in the environmental reflection. So you still see only the diffuse and specular reflections. At least for now, indirect light from light sources falling first on the ground or on other objects is not included.

More - It there is a specular exponent (glossiness = sharpness of highlights) map in the nomal map alpha channel, the value in the glossiness spinner acts as a multiplier for the map's value glossiness at each point. Smilarly, the environment spinner acts as a multiplier for the evironmental reflection mao in the alpha channel of the specular map, if it's there. If those alpha channels are absent, theyare effectively all white (= highest possible value). More more - the specular reflection is added to the diffuse reflection. It can sparkle if it gets too high. In contrast, increasing environmental reflection reduces the diffuse reflection. At the maximum environmental reflection, the diffuse texture becomes invisible. (This is like a chrome surface.)

In fact the requirement is even stricter than that. The material(s) must have the same names too, even if there's only one. In the Collada, the name that matters is the "xxx" in the material="xxx" attribure inside the <triangles...> or <polylist...> tag(s). That must be the same for equivalent parts of the geometry at each LOD. I can't tell you for sure where that name comes from in the Maya exporter, but if you make sure your two LODs are using exactly the same material, that should work.

In the oofficial viewer, at least, you can turn off shadows while ALM is on by simply setting the shadows option selector to "None". Someone else reported that this removed most of the lag otherwise caused by ALM on their system.

I don't understand this. You do have to switch the shape type to "Prim". Leaving it as "Convex hull" tells the server to use the default convex hull of the whole physics shap, which would fill in the depression of the road. But the physics shape pictures you showed did have the depression - meaning they must have been type "Prim". No? Maybe the shape display isn't working properly. Here it is for my simple thing - using physics shape for visible too... Top is the visual mesh, then the default convex hull you get when it's set to shape type "Convex hull". (Note that this doesn't quite fill the bounding box. I reported that as a bug, but I guess it's intended for some unknown reason.) The convex hull fills up the road space, as expected. So you don't sink into the roadway depression. Both the models below have the same default convex hull when the shape type is "Prim". Next is the physics shape uploaded using "Analyze", and set to shape type "Prim". It's a set of six convex hulls corresponding to the six separated parts. The roadway depression is now clear, and you can walk on it. At the bottom is the same model, but not using "Analyze". That results in is a triangle-based shape, as shown by the appearance of the triangles. This also lets you sink into the roadway depression. (The triangle-based shape isn't recommended for this model because the small triangles along the kerbs and the edges of the roadway cause a high physics weight.) Notes - pictures taken with water, surface patch and sky turned off (Advanced->Render types menu) fot clarity.

"Ruler/Protractor at the bottom of the Tool Panel" What Blener version? It's not in mine (2.66a).

Select the Scene button in the properties panel (usually on the right) next to the camera. In there, under "Units" is a set of buttons "None", "Metric" and "Imperial". If you activate "Metric" you will see the Blender measurements labelled "m" for metre, "cm" for centimeter, etc. I think the default, when you have "None" selected, is that 1 Blender unit = 1 meter as far as the collada exporter is concerned (Collada has to have the units). If you scale the scene (below the buttons) the Blender (None) units stay the same, but the Metric m,esaurements scale. I don't know if this scaling gets into the exported collada. You can also scale the model in the last tab of the upload dialog.

I cant see anything wrong with those. They look like they are "Analysed". One possible problem - the avatar doesn't actually touch the surface of a prim it walks on. Mine hovers about 0.075m above it. So if that is your kerb height, then maybe that's what's giving the impression that the shape is not concave. They way to get round that is to make the surface of your physics shape a bit lower than the visible road surface to compensate. Then when you stand hovering above the physics shape you appear to be touching the road. The bounding boxes still have to match up, so the visible mesh has to extend down to the bottom of the physics shape, (e.g. with an extra triangle). Here is a physics shape I tried to check there wasn't a new bug. It's six simple boxes, shown by colours. They are all in the same mesh and are just not overlapping. Uploaded with the LOD mesh in the physics tab after clicking "Analyze". I could walk on the depressd road surface as expected. Since it is not triangle-based, it works even if the mesh (visible+physics) is less than 0.5m thick. Since you describe connection problems, I wonder bif your connection could be flaky. The physics is all done on the server - so a poor connection could lead to all sorts of physics errors, which might include not setting the av height correctly. ETA - I here because the page-blanking problem went away, at least for a while.

"Noooooooooooooooo" See note I added ... may not be their fault. So I might relent. Still, a rest would be good.

So. Now I'm not allowed to see my own posts onless I agree to be subjected to garish flashing advertisements for loan sharks and gambling dens. And that from a company who I pay a lot of money directly. That is intolerable. I'm afraid I will not be contributing any more to these forums unless they correct this ... service. ETA - on reflection, it seems that this adblock-dependent page-blanking behavior is intermittent. Its timing is like that of the appearance of a particular advertisement. So it may be that it is coming from content of some of the ads rather than directly from LL or Google (although it does occur during the execution of a highly obfuscated google script). Therefore I will reserve judgement while awaiting clarifying information.

Just for fun. I made a similar door in Blender and baked normal and ao maps from it onto a plane, to see if that avoided the overlapping normal effects where the gaps between planks met the edges of the metal bars. You can see in the picture that the artefacts are still there, causing breaks in the highlight along the top edge of the bars. So this would need the same retouching as did the normal maps made in Gimp. In contrast, the baked ao does a much better job of supressing the specular highlights where the light can't go. Technically, I suppose this would be better if I did the retouching, but for some reason I prefer the handmade one. Not sure why.

You have to see whether it can flush away a whole copy of the Daily Telegraph.

"The door bugs the heck out of me btw ..." To expand on this for anyone who's interested, here's a simpler door (pics half size) ... ... and how I made it in GIMP (slightly simplified) ... Yellow is transparency. Everything except the woodgrain was drawn in GIMP. The diffuse layers were upward in the stack shown from left to right, superimposed with Normal mode. It was easier to make the masks and normal maps that way. Staring with a flat image, you would have to decompose it this way first. The normal maps were made by the normalmap addon from greyscale versions of these diffuse layers with appropriate contrast and occasionalblurring, as shown (actually, a bit more complicated, but...). They were combined, stacked in that order, with Overlay mode. To stop getting things like the gap between planks on the metal bars, it was necessary to fill intersecting parts of these maps with the <128,128,255> (ordinary normal). The selections for doing this were set up on the diffuse layers. Had to invert the green channel in this map (decompose, invert, recompose) to get the normals the right way round. (You can do that with normalmap from the start.) The main problem was that the sloping part of the normal maps generated from the diffuse maps extended further than the diffuse feature used to make the. So, for instance, the sloping metal bar edges overlapped the gaps between the planks, leading to incorrect highlights. So I had to adjust selections and/or masks to compensate for that. I began to think it might have been quicker to make the thing in Blender, using the diffuse images as template and bake the normal map. There was also the problem of trying to eliminate specular reflection from the bottom of the crease between planks and the junction between the metal bars and the bolts, where no light should be getting in. You can see my attempts to do this in the specular map, but I didn't get them right yet, as close inspection of the door would reveal. Once again, baking an ambient occlusion map from a model and superimposing it on the specular map might provide a more accurate and faster way of doing this. ETA - the picture is with glossiness = 80, environment = 0, under a single local spotlight at midnight. The specular map is also used as the alpha channel of the normal map (specular exponent map).