Ungoliant's mapping questions

[Nosslak]

thanks for this. I will check out highpoly baking, i was unaware it was a field of study all its own. The analogy is a good one, now I just need to learn how to bring the whole thing together into a final product.

 

This part I learned from some tuts online. the uv/image editor has a nice little "stretch" property that color codes based on stretching, and you can apply a uv test grid of checkered boxes to the mesh to see how it maps out in real time. When the boxes are uniform shape and size across the mesh, and the stretching overlay is mostly dark blue, your job is done. Can't wait to get that done on 86000 polys. Looks like i'll be getting some time in learning to use the 'pin' tool.

You don't need to unwrap the highpoly. When you're baking you're basically telling the software to shoot out a ray along the normal of every pixel for every triangle and then the last thing that your ray hits is what gets projected down. So there is no need to unwrap the highpoly. I usually unwrap the highpoly (very quickly without any regard for stretching or anything) for models that have lots of material that could be tedious to paint by hand. I just use 3 or 4 solid colors that I then bake down and uses as masks for all the different materials.

 

Now that I've learned of the sheer awesomeness of the Decimate modifier, it looks like i'll be going through 3 mesh stages. rough initial low poly, high poly, final low poly. What I'm still not sure of now is how I am going to layout the UV's of the high and final low polys so that they both fit the AO map in all the right places without excessive stretching in either mesh. I suspect the katsbits tutorial might have some detail on that subject.

The Decimator really isn't all that full of awesomeness. As Arcturus said earlier it will destroy your loops (this shouldn't be much of a problem here, but on a mechanical model it can be disastrous) and it looks like it completely destroys your unwrap (I've just done 2 quick tests so I might be wrong). You'll probably be better off just optimizing it manually while trying to avoid the seams. The only thing I ever use the Decimator for is to just count my triangles (the "fa:" value up top doesn't suffice as it just counts faces without any regard for if they are triangles, quads or whatever), but it's not even all that good for that as it just updates when you go out editmode and it hardens all your normals (so if you want to preview the model you need to render or turn the modifier off).

[ungoliant]

Thanks to katsbits, I am now able to see the light at the end of the tunnel. After i spend some time learning some UV unwrapping techniques, I think I'm actually going to be able to pull this off

[ungoliant]

question to somebody knowledgeable on the team: is it acceptable to cover seams by vertex painting, or is this just a blatant and unnecessary waste of end user system resources? (dunno how light or heavy vertex painting is on performance).

 

edit - more general question: do arrows hit the collision mesh? I feel like I should probably know the answer to that already, lol.

Edited April 10, 2012 by ungoliant

[Nosslak]

question to somebody knowledgeable on the team: is it acceptable to cover seams by vertex painting, or is this just a blatant and unnecessary waste of end user system resources? (dunno how light or heavy vertex painting is on performance).

You should try to hide the seams as much as possible when you unwrap the model and if possible you should try to place seams on otherwise hard edges or in seldom seen areas (bottom of objects, armpits of characters etc). If this still isn't enough you are supposed to be able to somehow essentially take a printscreen of the model, paint away the seams on the printscreen in GIMP (might work in Photoshop, I've never done this at all) and then project that fixed texture back on to the model. I've never done this as most seams aren't too visible on my models to start with and you'd have to work one layer at a time AFAIK with this method (it was a couple years ago since I read about it so it might be good now). Why would you want to use vertex paint?

 

edit - more general question: do arrows hit the collision mesh? I feel like I should probably know the answer to that already, lol.

Yeah, arrows hit the collisionmesh (that's what I've been told at least). So if you make a wooden object (or anything else that arrows might get stuck onto for that matter) take care to not have the collision model stick out too much (the arrow might float in the air in that case).

[Baddcog]

And the collision can have any number of TDM materials: ie:collision_wood (I think, there is a list on the wiki) collision_stone for sound, arrow breaking. A non-moveable can have multiple meshes (ie:wood and metal)

I've admittedly been a bit lazy with this lately, slipped my mind I guess... maybe before 1.08 I should clean up my newer models.

 

Main thing is getting it close, with stone it wouldn't be as critical, if an arrow hits the air beside it and breaks/ricochets it won't be that noticeable. If it's wood and the arrow hangs in mid air it will be.

 

Vert painting really doesn't work in tdm anyway. It works for blended materials (ie stone where black is, grass where white is), but i haven't had any luck using it for shading. I tried to use it for AO. Didn't show in game. (of course it could be used to blend a dirtier version of a material, but not just darken it)

 

If it did I'm sure it would be similar to just blending an ao pass, or a specular stage, etc.. just one more draw call to add up. So if you have normal, diffuse, spec and an ao you have 4 passes per model. If you have a cave with 80 of those models that's 320 draw calls. 3 passes is 270, so yeah it would have an effect on performance.

[Springheel]

The vast majority of our non-movable models don't even have collision meshes. It's a nice idea if the mesh is pretty high poly, but it's not crucial.

[ungoliant]

i dunno, somehow 460-480 collision tris just seems..... unnecessary.

right now i'm sitting on 480 after decimate, and it looks good, but theres some completely unnecessary verts sitting on flat planes so i think i can reduce a few more. i'd rather do extra work on the seams than extra work trying to eye the lowpoly to fit the highpoly. turns out the AO baking method i'll be using actually depends on making the low and highpoly version fit as closely as possible into the same xyz space, so i think procedurally generated with decimate is the way to go, as long as it has enough tris to do the job (which it does)

 

edit: not all my expanded models i plan to make will be in the 500 tri range. This one will be more of a centerpiece. I expect smaller stalactites and simpler stalagmites to be in the 20 - 150 range, with more convex shapes that would not even require an AO map. This first one is more of a proof-of-concept showcasing for my very first model.

Edited April 11, 2012 by ungoliant

[Baddcog]

Thing with collision is that it only is taken into account when something hits it. So sure, if something bounces off of it collsion calculations will be higher. Mainly when it's important is when a lot of things are bouncing around, which really doesn't happen often.

 

Still, having a 20-30 tri collision mesh is probably fine for something like that. Mainly you don't want the player to clip into it, or have an arrow hang mid-air, or the player be able to set a pot on it and have it float badly.

 

The shadow mesh though is very important for performance. That's where the engine can choke, and that's a constant drain.

 

But there are some models I've made that if I went to clean them up I'd probably get rid of the collision mesh/s. Just because they add to overall polys/draws and don't really do that much for performance.

-------

I don't care for decimate tools myself. Most of what I've seen comes out pretty ugly (but it's fast). You can do it yourself and take a little time and the results come out really nice.

 

Personally on LOD I've been shooting more for a mid range model to look nice, the high is nicer, but will only bee seen really up close and/or high end systems, the low doesn't have to look as good, but try not to go so low it looks like complete crap. Doing it manually lets you control that much better.

[ungoliant]

well I've got the current edition of the model working in TDM with the AO stage applied with a blend filter stage. It works. The results are much more... subtle than I was anticipating though. I suppose i could derp around with bake settings, or manually increase the contrast of the image. It looks pretty good as is, but its not the spectacular phenomenon I was hoping for.

[Baddcog]

Ao should be pretty subtle, that's why it works good with lighting.

[ungoliant]

sculpting with many more polys than before. trying to maximize performance at same time. Is there a way to disable viewing inside normals during sculpt? i can flip and recalculate and wateverthafuk, but i can't find a way to disable inside normals from being displayed in solid shading mode. i suspect if i can do this, it will be literally double performance.

any other tips for sculpting performance appreciated.

[rich_is_bored]

In the properties panel under object data there is a double sided check box.

 

 

I haven't noticed a difference in terms of performance there.

 

The greatest gains I've seen come from solid view port shading and additionally fast navigate which is under the sculpting options.

[Nosslak]

sculpting with many more polys than before. trying to maximize performance at same time. Is there a way to disable viewing inside normals during sculpt? i can flip and recalculate and wateverthafuk, but i can't find a way to disable inside normals from being displayed in solid shading mode. i suspect if i can do this, it will be literally double performance.

any other tips for sculpting performance appreciated.

Supposedly it should help if you first subdivide the mesh a fair amount with the subdivision surface modifier first and then use the multi-res modifier beyond that. I haven't sculpted a lot in Blender but that's what I've heard is good for performance. The only thing I ever sculpt in Blender is heightmap terrains (where Blender is better than ZBrush because of better axis-lock) and very simple sculpts (like a pumpkin). Otherwise if you want/need performance with a lot of polys (like 1 million polys or more) you should really get ZBrush. ZBrush isn't particularly easy to learn as it's got a pretty weird interface that in places don't make any sense to me, but once you learn the interface it is a very powerful program.

[ungoliant]

Ended up using about 450K polys instead of the 1.3 millionish i was originally intending. It seems to work fine, as long as you get the lowpoly as close to the final shape as possible before multires-ing. I suppose if I need to add more details to individual stretched out areas I can always subdivide the areas in question a little more without stepping up multires again. I managed to get some nice results with the displacement modifier as well, but I'm wondering if thats really going to do anything useful unless I decide to bake a normal map as well as the AO.

[ungoliant]

Ok blender is driving me batshit insane, and I cannot figure out a way to fix it. Using sculpt mode my model is now reacting to brushes in an almost random fashion. i drag a Draw brush across a surface, half of it moves outward, half of it moves inward. blob works like crease. polish works like crease. pinch/magnify and smooth seem to be the only tools working correctly. I think it has something to do with my model and not the tools though. Like the normals or something. There are certain areas of the model that seem to behave a certain way, like adding the brush instead of subtracting. I tried recalculating normals, and viewing them, they all seem to stick outward, but the tools are still fucked. does anyone know wtf is going on here?!?

 

edit: even after restarting several times, nothing happened. Then, randomly while playing with tools, the whole thing decided to work again. Looks like i must have accidently toggled the "fuck everything up" option on and off.

Edited April 23, 2012 by ungoliant

[ungoliant]

having trouble making a normalmap out of my highpoly. The quads are very very unevenly spaced, which causes displace and texture painting in sculpt mode to give bad results.

Is there an easy modifier or function to re-distribute polys in the same shape as the original mesh? I tried shrinkwrap with a big 1million poly cylinder, but for a shape like that stalagmite, its a disaster.

The other ways i guess could be using weighted vertex groups, or further subdividing areas manually, but I'd like to go for a one-size-fits-all solution first, if it exists.

[Arcturus]

You could try 'relax' addon with shrinkwrap. But the tiny bumps don't need to be real geometry. You can sculpt only the biggest bumps, and then add noise texture set to influence Geometry -> Normal. This little noise will not be visible in real time but when you bake normalmap, it will be treated as a heightmap and combined with real geometry.

 

Or even better, use Doom engine. In your material you can have two stages:

- normalmap (for big bumps)

- heightmap (for tiny details)

They will be combined during mission loading. This way you can have different bumps for different diffusemaps while still maintaining shape of the highpoly model. You probably don't even need a separate heightmap and can use a diffusmap for that. This way bumps will match diffusemap.

[ungoliant]

would applying the heightmap texture as a displacement on a flat plane work effectively as a preview so I don't have to keep rebaking over and over?

Edited April 27, 2012 by ungoliant

[Arcturus]

Displacement is dependent on the polycount so it will not look exactly the same. You can hit render to see how a heightmap looks.

[ungoliant]

Is there some kind of trick to rendering a good image in Blender?? I've been working on settings for half an hour, and no matter what I do, it looks like shit compared to d3. I set the ambient low, the point light to a color similar to our torches, specular to pretty low intensity, with a lot of hardness, set the blend stage on the AO to multiply, and turned off self shadows via the material, and made sure to set all images to use uv coordinates, and the end result still makes me want to barf. Should it really look this much different from D3 screenshots?

[Nosslak]

would applying the heightmap texture as a displacement on a flat plane work effectively as a preview so I don't have to keep rebaking over and over?

I'm not sure what you mean, but why can't you just bring the baked map and the detailed texture (converted to b/w and then to a normalmap), change the blendingmode to overlay, turn off the blue channel and mask away the parts you don't want? If you do it like that you should just need to save that out, update the texture in Blender and then see it in the viewport in real-time.

 

Is there some kind of trick to rendering a good image in Blender?? I've been working on settings for half an hour, and no matter what I do, it looks like shit compared to d3. I set the ambient low, the point light to a color similar to our torches, specular to pretty low intensity, with a lot of hardness, set the blend stage on the AO to multiply, and turned off self shadows via the material, and made sure to set all images to use uv coordinates, and the end result still makes me want to barf. Should it really look this much different from D3 screenshots?

Show us what you've got and we might be able to help you.

[ungoliant]

why can't you just bring the baked map and the detailed texture (converted to b/w and then to a normalmap), change the blendingmode to overlay, turn off the blue channel and mask away the parts you don't want? If you do it like that you should just need to save that out, update the texture in Blender and then see it in the viewport in real-time.

 

If that works, that is pretty friggin cool. I don't follow the part about masking though, I don't know what it does in this context or how to do it. Also, whats the blue channel in the normal map do? In the meantime, I've baked a normalmap without all the fancyness of the bumps, just the smooth highpoly, and I'm just looking at the rendered image. You guys were totally right about that, the difference is night and freakin day. Just need to create a good generated heightmap slap onto it.

Having trouble figuring out how to make anything other than random patterns, considering that its going to be projected from UV's. Some concentric rings would be nice, but I can't fathom how to get it done without applying the stupid image onto the highpoly mesh as a displacement.

[ungoliant]

went for a closer look at the normal map results in d3. something is wrong. most of it looks great, but the normal map appears to be affected by the seams just as badly as the diffuse. The AO bake did not have this problem with seams, so why is the normal bake getting all crazy?

 

edit: might this be because the edges of the UV islands do not recognize their counterparts during the bake, and are not colored correctly? how can i fix this?

Edited April 27, 2012 by ungoliant

[Arcturus]

You have to invert green channel of the normalmap (in Blender 2.5 and later) because of the differences between Blender and Doom engines. In uv/image editor: image -> invert

[ungoliant]

that explains why it looked fine in blender. thanks.