GUIDE [WIP] DDSopt & Texture Overhauls

[z929669]

Discussion thread:
DDSopt Guide by STEP
Wiki Link

---

 

GET DDSopt:
Github Pre-release versions
Official Nexus versions (select pre-release update 4)

[z929669]

Nothing bad at all ... just maintenance overhead ;)

 

I just wanted to confirm that there are no in-game visual anomalies. The CK issue is not ideal, but if the issue is not present in game, then it would not be a problem for 95% of all modders (user side). That may be why I never see any of these issues then, as I almost never go into the CK. I hate that application on so many levels (as I have hated the CS in the past). I understand wanting to provide a perfect product though.

 

Just wondering, nevertheless: Has anyone here confirmed that 1x1 is the cause of the gray faces in the CK? Has anyone here seen the gray faces in the CK? If so, have any of those persons positively validated the issue by adding back the default face tint file and 'fixing'?

 

EDIT: Actually, this saves me a lot of time looking for these issues in-game. I am not entirely sure how to see the issues in the CK though in order to trace if possible ...

[Kelmych]

I don't believe that there would be any visual regressions. Even if it's stretched weirdly, it's monochromatic. There's not a lot to distort. The only issue would be hardware (and the Creation Kit). CK users have reported many times that if the facetints are touched, the CK freaks out and they end up with gray face. The difference between 1x1 and 16x16 is negligible, though. If using 16x16 textures potentially avoids hardware issues, I don't really see a downside to using them.

Since we don't understand the mechanism, any choice for size is arbitrary. There is no evidence that 16x16 is any different than 2x2, 4x4, or 8x8, or even for many users 1x1.

[tony971]

I've just had a breakthrough on the installation method front. The next version will get overridden by everything and be compatible with every tool. And that's without using a single plugin 

[z929669]

Still waiting on a response to my last ... I really hate to implement anything without evidence, and I really hate to gather evidence if someone else has already done it :P

[Kelmych]

It's hard to find a place in Skyrim to compare the results for testing the effects of texture collapse and the alternative textures built with the tools that alt3rn1ty created for reduced size textures such as those used by Tony971 . One of the few location possibilities I'm aware of that might be useful is to use a texture from the Skyrim Sewers mod which includes a "texturesarchitecturessewersinvisbletex01.dds" monochromatic texture. Note that it has an alpha channel, so it might not be suitable for our comparisons. This particular texture is used on the sewer grate that is on the left as you enter the main Whiterun door. It is collapsed to 1x1 A8R8G8B8 when optimized in DDSopt, and when the collapsed version is used the texture shows up as purple. This texture is not a vanilla replacer, by the way.

[z929669]

I am wondering if a similar effect is seen with other 1x1. That alpha likely makes all the difference. If the collapsed version changes the alpha to pure black, then that would account for the purple (i.e., texture is missing).

 

I optimized every vanilla texture without constraint and will test the effects in game by randomly going all kinds of places ....

[Kelmych]

The vanilla terrain textures that get collapsed don't have an alpha channel. I don't think there are any other vanilla textures, other than the face textures I mentioned, that get collapsed. I've never seen anything in Skyrim that seems consistent with a problem caused by collapsed textures.

[z929669]

Actually, there are lots of things in addition to faces that get this treatment ... head, hands and skin textures in addition to architecture, blood, clutter, cubemaps, LOD, sky, etc.The facetint stuff doesn't seem to use alpha, but I have only checke a few ...Here are some interesting examples ... alpha used for effects data:

# orange.dds ##################################################################D:Skyrim ModsMod_WorkingOptimizationSkyrim Vanilla Textures (cleaned source)texturesdlc02effectsorange.ddsprocessing: Format		   : DXT1 to A8R8G8B8 Dimensions	   : 4x4 to 1x1 - 3 to 1 levelsnotes: Turned contrast off for interface texture. Planar image detected, collapsing to size 1x1. Texture was compressed.delta:		    20 bytes less# fxtentacle01.dds ############################################################D:Skyrim ModsMod_WorkingOptimizationSkyrim Vanilla Textures (cleaned source)texturesdlc02effectsfxtentacle01.ddsprocessing: Format           : DXT5 to A8R8G8B8 Dimensions       : 256x1024 to 1x1 - 11 to 1 levelsnotes: Turned contrast off for interface texture. Planar image detected, collapsing to size 1x1. Texture was compressed.delta:            349580 bytes less

... and another ... alpha used for opacity:

# transparent.dds #############################################################D:Skyrim ModsMod_WorkingOptimizationSkyrim Vanilla Textures (cleaned source)texturesdlc02weaponsrieklingspearstransparent.ddsprocessing: Format           : DXT1 to A8R8G8B8 Dimensions       : 128x128 to 1x1 - 8 to 1 levelsnotes: Planar image detected, collapsing to size 1x1. Texture was compressed.delta:            10932 bytes less

The face textures may be a special case, since those textures are dependent upon base textures, and many titmasks do not get converted to 1x1, because they are not monochromatic. My guess is that the meshes have the restriction. But this will likely not apply to anything not being used as a texture overlay-type texture or maybe anything other than faces.

 

EDIT: Following is an example of what I am absolutely sure is a problem texture after optimization, but it is unclear if this is due to downscaling the texture or conversion, but I suspect the former:

# skyrimcloudsfill.dds ########################################################D:Skyrim ModsMod_WorkingOptimizationSkyrim Vanilla Textures (cleaned source)texturesskyskyrimcloudsfill.ddsprocessing: Format		   : DXT1 to A8R8G8B8 Dimensions	   : 32x32 to 1x1 - 6 to 1 levelsnotes: Planar image detected, collapsing to size 1x1. Texture was compressed.delta:		    692 bytes less

 

[tony971]

Creation Kit specific issues can be ignored. The (soon-to-be) updated installation method makes the CK a non-issue.Edit: One visual regression that has been noticed, however, is in pine trees. They become less full after optimization. 

Edited October 16, 2014 by tony971

[z929669]

you want to use "raise foliage-map opacity each mip-level" and set that to some higher value in the DDSopt submenu under behave. that will fix such issues.

[z929669]

@Kelmych et al

Please take a look at the current state of the DDSopt Guide on the wiki. We need to better consolidate info contained within the FAQs and use examples under the Appendix section. There is a !#%&-ton of information on those pages and much of it is redundant. If someone could consolidate it somewhat, I will take care of optimally organizing it all and incorporate it better into the overall guide structure.

 

Also note the info in yellow text ... this is stuff I am uncertain about, so any info we have that I may be unaware of might need to be added to rectify or embellish.

[lexo1000]

Hi! I have read every single lines in the guide but some stay mysterious for me

1 - What is the general rule for contraints tab? Sometimes the guide says using DXT compression and another time R8G8B8.

2 - I want to clean some textures mod to remove all useless blank alpha channel. But I try many time to process these with DDSopt but its never remove this unnecessary alpha..

3 - Some color map have white alpha channel but with just some black pixel in it. I think its a bad conversion of the texture, can you confirm? It will make just some pixel transparent, that is seem useless. Can I remove this alpha with no risk?

Edited December 15, 2014 by lexo1000

[Aiyen]

1: There is no general rule.. it all depends on what you want to accomplish. I will try to recommend that you study up on compression and their formats if you want to know all the intricate details of it. If you want to simply have the most general answer, then simply convert everything to dxt5 and call it a day. It is the best performance/quality format there is. The only time you would ever want something else is for certain normal maps where you do not want any compression at all. 

 

2 & 3 : How do you know the alpha channel is useless ? ;) 

Just because you have a fully transparent texture, perhaps the model or effect it is attached to requires an alpha channel to work and behave properly. So as a general rule you will always want an alpha channel. It does not kill all performance to have it, so you might as well leave it if you are uncertain. If you do not want one for just a specific texture, then DDSopt is not the proper program to use. Use GIMP or Photoshop to manually fix individual textures, to your liking.

Can you can remove it/them? Well go ahead and try it out then fire up the game and take a look... that is the only sure way to make sure.

 

In general it seems like you expect more of DDSopt then what it is meant for. It is a general purpose batch converting program.. for when you do not want to open up every single individual file in GIMP or photoshop and fix it manually.  

[lexo1000]

Thanks for answering ;)

 1 :  The vanilla DDS are all convert to DDS so I could understand that's the normal compression method for everything. I have read lots of threads about this question but it seems that there is no rule, except for _msn body texture, which must stay uncompressed. But if a dds doesn't contain any alpha mask, the DTX1 should be ok without lost in quality no?

 

2-3 : A pure white alpha channel on a normal map is useless, yes. Its take up space and its useless, since its only inform that the texture is totally opaque, which is the default for a color map without any extra alpha channel.

Edited December 15, 2014 by lexo1000

[Aiyen]

Normal maps alpha = their specular map which depending on the mesh or effect is relevant. 

 

If there is no alpha there is no loss in dxt1 correct.