Models by Jorzi (AR)

[Olrox]

About turning the wheels: It was as always a question of realism vs design.

Yeah, but such small details sometimes make big difference.

From my point of view, it could be easier to distinguish the half-tracks from the tracks if the front part was a bit slower, especially at farther zoom levels

[Jorzi]

The wheels should be slightly smaller and further apart now

[Emdek]

Yeah, a bit better now, although at first glance not big difference (in such cases it is good to repost previous one for easier comparing ).

[Olrox]

Yeah, a bit better now, although at first glance not big difference (in such cases it is good to repost previous one for easier comparing ).

oh, I just opened both in separate tabs for comparison


The difference is indeed very small.

[Safety0ff]

The difference is indeed very small.

Playing "spots the differences" the only things I see that are different (other than the wheel spacing) is:

• the front light is slightly smaller, and there appears to be a crease above it.
• A links have been added to the front/top track.

Great work as usual Jorzi.

[Jorzi]

Finally got the heavy halftracks unwrapped and baked today.
Oh, and @SafetyOff, the front light thing was an accident, thanks for pointing it out

[Jorzi]

I've been trying out the new tangent space features of WMIT.
At first I had a few problems. Instead of smoothing out the sharp edges, the normal map amplified them. So I tried flipping the cross product generating the bitangent (cross products always tend to go wrong for some reason) and it produced much better results. It still produces a bit of dstortion at the seams (most notable at the top) but I think it is within acceptable limits.

The change I made was at line 16 in pie3.vert, which I changed to:

Code: Select all

localSurface2View[1] = normalize(cross(localSurface2View[0], localSurface2View[2]) * tangent.w);

I also changed line 28 in pie3.frag to:

Code: Select all

		N = normalize(localSurface2View * (2 * texture2D(Texture2, gl_TexCoord[0].st).rgb - 1));

This does not affect the result in any way, but I think it makes it more readable.

I made a quick comparison to see how well the normalmap translates from blender to wmit.

[MaNGusT]

looks like edges(with their points) are divided in these parts.... or it's just the NM texture seams that can be fixed manually(usual tanget space NM bug).
btw, looks very smooth and nice, I like it.

The change I made was at line 16 in pie3.vert, which I changed to:

Code: Select all

localSurface2View[1] = normalize(cross(localSurface2View[0], localSurface2View[2]) * tangent.w);

this inverts the Y direction of tangent space normals
at least on my drivers
but it's not a problem, I can always invert the texture's green channel to fix this.

[Jorzi]

the lower mesh is actually made up of separate duplicated parts (8 parts for lower and 6 parts for upper mesh). The normal map is made to compensate for the edge splits.
Inverting the green channel should be equal to uncommenting line 29 in pie3.frag:

Code: Select all

		N.y = -N.y;

This didn't work for me, however, so if it works for you, good thing

[MaNGusT]

Inverting the green channel should be equal to uncommenting line 29 in pie3.frag:

Code: Select all

		N.y = -N.y;

This didn't work for me, however, so if it works for you, good thing

We have to find a way when shader's code will work on most adequate gpus without such hacks. :/ the original code locates here

can you show me your nm btw? I want to look at colors' directions.

[Jorzi]

The result is not gpu dependent (other than some intel gpu's not supporting it) so that's not the problem.
What I meant was that I tried inverting the green on my normalmap (by changing the code, but it should do the same thing) and it didn't produce the right result.
Thanks for the code link btw

[MaNGusT]

well, you could invert the green channel in gimp instead of did changes to the code.

I see that you use 1 smooth group for entire model, why don't you use flat shading on parts with hard edges? I'm surprised that the result looks OK and doesn't have such bad parts that I've got here. I fixed my bug so: baked problematic parts as flat shaded(turned off smooth groups which caused it), removed texture seams manually, exported model with needed smooth groups turned on.

[Jorzi]

Hmm you were right. Apparently inverting the green channel in gimp equals this:

Code: Select all

localSurface2View[1] = normalize(cross(localSurface2View[0], localSurface2View[2]) * tangent.w);

and not this:

Code: Select all

      N.y = -N.y;

So yeah inverting the green channel is a much better idea than changing the code

About smoothing groups:
The trick is that all my hard edges are at texture seams and all my seams are hard edges (I like to talk about hard edges instead of smoothing groups because blender has no such thing as smoothing groups)

[Jorzi]

I've been texturing both the medium and heavy halftracks in parallel. Still a few details left but they pretty much have the general look already.

[Emdek]

Very nice, although I really hate those edgy wheels, but such detail would took lots of vertices...