Seams...
http://img148.imageshack.us/my.php?image=swordlevel2ti7.jpg
As you can see, the seams appear at a distance. Does this always happen? I've written my own normal map shader and tried different samplers but that didn't fix the problem (when I'm in MAX).
Ideas?
As you can see, the seams appear at a distance. Does this always happen? I've written my own normal map shader and tried different samplers but that didn't fix the problem (when I'm in MAX).
Ideas?
Replies
Also you should fill the empty parts of your maps with average colors (not black or some other contrasting color). For normal maps use 128,128,255 as the empty color.
I use edge padding but I multiply an AO bitmap that is outputted from Xnormal's simple AO baker.. it makes your textures look 10x cleaner for display.. That's my issue then.. Cool, for the real textures I simply wont use it.. but when I put them on display I will.
I'll report back! But before I leave... does anyone have technical information as too why this happens?
http://en.wikipedia.org/wiki/Mipmap
Thanks so much EricChadwick!
Mips are only stored by a program if you're using a bitmap format that can store them, like DDS, and only if the program actually creates those mips.
Every game engine has a mip-creator that it runs either at load time or as a pre-processing step, for all bitmaps that don't already have their own mips.
What exactly do you mean? Are you saying that your shader multiplies the diffuse by the AO, or that you're adding it directly to to color channel?
Don't do this. Don't multiply an AO map only on the flat but not on the bitmaps used on the model, because an employer wants to see exactly what you use. Not some prettified revision.
A properly-filled even-colored bkg is actually more professional. But I can see the benefit of using black in the empty bits, makes it easier for someone to evaluate your UV layout.
[edit]lol no i mean 12 padding for nm/ao =D
I guess the question goes back to you as to how much bleeding are you getting that 6px isn't enough?
To each their own I guess. It's not like your a slouch when it comes to this game art stuff. I'm sure you're making the best decision for what you're working on, it just seems a tad bit like overkill to me.
I normally have at least 4 pixels space between all the uv chunks but that doesn't mean that there aren't any spaces that have a bit more, so why not go higher? I only have less then 4 pixels distance in unimportant areas where the bleeding wouldn't impact the look.
6 seems to be the max for 512 I want to use and not have it impact my layout. If I work on smaller textures then I use the padding ratio for that texture size. 1024 still seems big and a bit of a waste in most cases, but maybe not anymore.
Maybe I'm getting more seam bleeding then I realize... But then again most of the textures I've been making have been for Source and Unreal runs on a bigger scale and might run on higher res textures? last I checked 1024 was almost standard in Unreal.
This is pure theoretical and maybe 6 pixels are enough and no difference is noticeable but on power of 2 textures i always try to keep my pixel ratios also power of 2.
Thanks for the heads up!
My model doesn't suffer the black lines anymore since I followed the advice here, and my normal maps look better on edges too! Awesome!
For a small prop that's using a 256, padding of four or six should be plenty, as the object will be so small at that point that you won't notice mip bleed anyhow.
I had this issue with a bunch of buildings for Sega Rally Revo that had to use a single 1024. Some of the buildings were quite large, so had to be diced up into squares or strips to be UVd onto the single texture sheet. Since they were so large, if I used a tiling square somewhere on the sheet it needed a lot of padding around it to avoid mip bleed. If it were strips, then I'd have to make sure that there was a lot of padding vertically.
Another really easy thing to do in addition to the propper padding is to just set your background color in the texture to the neutral that most closely matches your problem area's color. In the case of the sword, the background should be gray. That way if you still get some bleed, at least it's less noticeable.
I was looking at thomasp's site the other day (nice work, if you're listening), and here's an example, where it's tougher to make out how tight the UV is because it's filled with average colors.
http://www.thomaspecht.de/img/ninja
Shows he's a pro for using the technique, but on the other hand it's not as easy to judge the UV. Which I think is probably more important for up-and-comers' portfolios.
Most of the props I've done lately are roughly player size and I haven't noticed any major mip map problems using 2-6px for padding. I've also been careful to keep a consistent pixel density with the game world so I don't have a sharply detailed prop sitting on a pixelated blurry wall.
But I DID notice it when I scaled the same objects up to 20x. Which breaks one of my personal rules, don't use 10 gal of paint to cover a 2000 gal wall. Either use the right amount, or get crafty. Which is easier said then done most of the time...
So I guess there is another really important factor to keep in mind =/
log2(max(width,height)) + 1 = nMipmap Levels + 1
for example... for a 2k x 2k texture is (12+1) = 13
in that way it's guaranteed that if two pixels use the maximum and minimum mipmap LOD then a 2x2 box filter result won't present visible seams. The default in xNormal is set to 16 so it should be ok even for a 8k x 8k map... So keep it at 16, it will take just a few miliseconds more to process the bitmap but you will avoid better the possible seams. However, as Earth mentions, you cannot increase it without limits because then a lot of texture space will be wasted...
And remember... don't fill the texture margins with data... keep them empty... so the dilation filter does not clamp the nearest neighbors.
To reduce the seams more you can also fill the background with the (0,0,1) tangent-space normal RGB888 value = ( (0,0,1)*0.5 +0.5f )*255 = (127,127,255)