Jump to content

HumanStuff - Add-on Development Thread


Recommended Posts

HumanStuff -Add-On Develpoment Thread

MOST CURRENT UPDATE:

We have a HumanStuff Wiki NOW !

MASSIVE changes to mods file structure. I did not realize how "involved" I would become in making sure I got the added realism "just right", in 1 day the mod had tripled (3x) in size from 900Mb to 2.7Gb. But I got a 6000x6000px (6K) Skybox thats "cubic correct to the plane of the solar elliptic" from @Galenmacil's work, very bright method, I think I can use this method to create a 8K and 16K RSS correct GalaxyTex (skybox) and 1K (1024x1024) EnvMaps (I already made the 512x512 EnvMaps & 4096x4096 GalaxyTex, also correct for RSS) I quot:

PS: NOPE! Tested these "older" files, they are in someway not working out in my compression programs as "nice looking". Space Engine captures of smaller size (4K) are far nicer when compressed to DXT5 or BC7, even as DXT1. 

 

Spoiler

 

Methodology used in SpaceEngine:

1 - "Land" on the Sun.

2 - Orient the camera so that the equatorial grid is aligned (Top is North), looking through the 0h, 0° coordinates.

3 - "Disable" the Sun and planets.

4 - Adjust magnitude limit and stars brightness to adequate level. [aazard: GOLDEN INFO!]

5 - Generate the cubemap. Each of the 6 images are 6000x6000, 24 BPP, lossless compressed .TGA file.

I tried going higher than 6k for the resolution but, SpaceEngine being a 32 bit application, this is not possible. It would crash during computation...

[aazard: ^ this I can handle...this is more my area of "expertise" if this is truly the issue...its minor]

[aazard: GOLDEN INFO! VERY easy fix!! The issue with the 8K+ image creation is most likely caused by: answer from @harbingerdawn]

"The crash was probably a graphics driver timeout, not a direct crash from SE. You can change the timeout duration (time before the driver resets) in your registry. But anything higher than 4k for a skybox is excessive. You'd either need to zoom in considerably or be playing on a 5k+ screen to need more than 4k resolution."

 

^Can anyone assist with this ^

IS THIS STILL TRUE? About GaxaxyTex & EnvMaps? Same "root" image can be used on both?:
 

Spoiler

 

Name and format:

      Skybox KSP Name         // Skybox Name

      GalaxyTex_PositiveX     // skybox right face

      GalaxyTex_NegativeX     // skybox left face

      GalaxyTex_PositiveY     // skybox bottom face, rotated for 180°

      GalaxyTex_NegativeY     // skybox top face

      GalaxyTex_PositiveZ     // skybox front face

      GalaxyTex_NegativeZ     // skybox back face

      

      Visor Reflectuions

      PositiveX  // skybox right face, vertically flipped

      NegativeX  // skybox left face, vertically flipped

      PositiveY  // skybox top face, vertically flipped

      NegativeY  // skybox bottom face, vertically flipped

      PositiveZ  // skybox front face, vertically flipped

      NegativeZ  // skybox back face, vertically flipped

 

_________________________________________________________________________________________________________________________________

 

A (Slightly More) Realistic Spacesuits, Heads & Staff. For KSP v1.5+ NEEDS (as a "Dependency") TextureReplacer

Intended & Suggested For The Realism Overhaul, RealSolarSystem, & RP-1 Realistic Progression One Mod Families, Will work with just TextureReplacer

tTbyLUQ.png

(^Current "Poorly Made Hack Job Title Image" For HumanStuff^)

Forgive place holder post(s), trying to organize a lot, for me, my very 1st mod/github thing ever

________________________________________________

Plans, Progress & Issues:

*Complete the Do it Yourself "Easy Head & Suit BuilderLayered Components Workshop (LCW, for short)*

  • A set of helper PNG/DDS files to assist users in learning to make heads and faces, or all more custom options to users that may be unable to complete the process personally (aka they are limited to pre-made heads and suits) 
  • GENE KERMIN <- I am lost to why he works in 1 scene but not others....and "albino" why? Others dont have this issue unless they use gene's head (some SPH/VAB ground crew)
  • Low Resolution (1/2 the current image sized listed), Standard Resolution (Current image sizes as listed), High Resolution (twice the current sizes listed) & "All Resolutions" (All in 1) packs
  • JUST READ THIS: must Env folder (reflections) images be PNG ONLY to work? NOT DDS AT ALL??

________________________________________________

Current Project I'm Working On For This Mod:

@Aazard (thats me) has been starting at the start.... mostly making sure i'm not breaking the file structure or "blowing up" it working right.

Also that I'm following "all the rules" to the letter (next to the really serious letter, I'm an easy going man), in regards to the EULA, TOS and licence terms of any inclusions

  • I'm working on 1950's suits ALOT. Early USA/CCCP high altitude flight suits and Mercury IEVA suits are "mostly done"
  • GENE KERMIN <- I am lost to why he works in 1 scene but not others....and "albino" why? Others dont have this issue unless they use gene's head (some SPH/VAB ground crew)

Also I am breaking down all the dds files, "4096/4K files" looking for "8192/8K files", attempting to optimize them:

  • Env: 256x256 bit depth 24, optimization pass 1= fail (I inverted in game view when making png into dds, noobism):(.. pass 2 = I "deepened blacks" BUT see no naked eye or file attributes (size) difference... so=:cool: done? Looking into 512x512 options...
  • JUST READ THIS: must Env folder (reflections) images be PNG ONLY to work? NOT DDS AT ALL??
  • Trying @Galenmacil's RSS Correct Ultra High Resolution Unique Galaxy Background CubeMap @6000x6000
  • Heads: 1024x1024 bit depth 32, file name fixes = done, location for TR fixes = done, images "cleaned" = huge difference in "color" quality! make into compontents = WIP... Looking into 2048x2048 options
  • Suits: 4096x4096 bit depth 32, pre ksp v1.5 suits cant be reworked only remade...so = done, make into compontents = WIP.Looking into 8192x8192 options
  • GalaxyTex: 4096x4096 bit depth 32, pass 2 = I "deepened blacks" BUT see no naked eye or file attributes (size) difference... so=:cool: done? Looking into 8192x8192 options

________________________________________________

Current Project Community Members Are Working On For This Mod:

@Chubby_Hamster was reworking the SK-1 IEVA and A7L "Inner layer" IVA suits last I heard from him, he is much better than I am with this and his work is very high quality, IMHO.

There is no "confirmation" of final submission currently, forgive if this announcement is in error

________________________________________________

 

Current Community Suggestions & Authour Requests:

Please feel free to make suggestions for anything!

  • I'm looking for higher resolution (native/original not "upscaled") png or dds images for all base "template building" and for the high resolution pack, and also to "downscale" to the lower resoloution packs (mostly to have everything the same & PREVENT quality loss to highest resolutions I desire: env 512x512, heads 2048x2048, skybox/suits 8192x8192, all with bit depth of 32 hopefully
  • Cfg files... these are not new to me i've edited many and the mods I use often require (or used too) "fix by users"... but writing my own "with a goal" is NEW. anyone REALLY understand MM and TR cfgs in general that I could ask direct questions too (that has the willingness to explain/help me for like 15 minutes)

Current Poll & Featured Art/Screenshots Of The Week:
*This Weeks Deadline: 09/18/2020 at 09:01 PM EST (UTC -5)

Last Poll Results: 

There was no "last poll", below is the 1st

Latest Poll: 

What Famous Face From History Do You Prefer To be The 1ST Added to Pack in "Famous Faces"? (All 5 will be added plus more over time*)

*With-in the limits of game, Texture Replacer and my skills (not in that order...)

Next Poll Thoughts/Options: 

There are no "Next poll" ideas or suggestions, yet

________________________________________________

 

I will make reposts of info from main thread them spoil out original post, mostly to "clean up" & to save "page length" on main release thread.

Preview of my mad art skills:

Spoiler

Kerbalhead-yuri.png

Ummm how do I reverve 2nd post slot? lol

 

Edited by Aazard
1st post to open topic
Link to post
Share on other sites

16K is my "ultimate" goal (for mods "highest setting" package)

Then dropping down in "stages" to:

Intention is best image quality on a 4K monitor, playing at forced 4096x2160 AT MAX ZOOM IN, then the same for other resolutions @ 4:3 & 16:9 aspect ratios.. Aiming at:

4:3 sized 
SVGA @ H 800 V 600 R 4:3 Mpx 0.480 up to QUXGA @ H 3200 V 2400 R 4:3 Mpx 7.680
&
16:9
WSVGA @   H 1024  V 576 R 16:9 Mpx   0.590 up to WQUXGA @ H 3840 V 2400 A 16:10 Mpx 9.216
&, OF COURSE 
4K & "better" UHD

key: H V A Mpx (Horizontal Vertical Aspect MegaPix)

4K UHD 3840 2160 16:9 8.294
5K 5120 2880 16:9 14.746
8K UHD 7680 4320 16:9 33.178

key = GalaxyTxy (skybox) / EnvMap (reflections, at 1/8 size the GalaxyTex vertically flipped)

  1. 16K/2048, thats called "Epic"
    > 6 or 9 monitor "stretching" & 8K monitor intended, I have access to a 7,680 x 4,320 Dell UP3218K at work & a 5,120 x 2,880 Iiyama XB2779QQS at home to test
  2. maybe 12K (12288x12288)/1536
    > 5K/5K2K monitor intended 
  3. 8k (8192x8192)/1024 "SuperHigh"
    > 4K (2160P) monitor intended
  4. "6k ( 6144x6144)"/768 "High"
    > 1440p & 2560x1600 intended
  5. 4K (4096x4096)/512 "Medium" thats "Default Standard for Mod Pack"
    > 1080p & 2024x1536 intended
  6. 2K (2048x2048)/256 "Stock"
    > 900p & 1600x1200 intended
  7. 1.5K (1536x1536)/192 "Low"
    > 720p & 1280x960 intended
  8. 1K/128 "eSports" (I mean current "iGPU" systems like an AM4 Athlon's like the 200G to 3000G or 2019/2020 "generation" Pentium "golds" to entry level i3's)
    > 1024x576 & 800x600 intended 
  • Optional: 960/112, 720/96 & 640/72 >
    "SLOW/older" iGPU pc's & laptops down to "Min KSP system requirements" pc's & laptops
  • Optional EnvMaps: 64x64 and 32x32 separately,
    > "Min KSP system requirements" pc's & laptops that need just a little boost, but still want reflections

*Any Ryzen 3 "G" & current 2019/2020 i5 "iGPU/APU"'s, or better, should handle "default for mod" fine

----------------------------------------------------------------

Editing/converting etc etc is getting "easy" and its pure open source apps doing the work so...wow

I also successfully generated my own "RSS correct" space engine GalaxyTex images in DXT5 BC3, Kiaser filter mipmaped 8192x8192 32BPP, 8-bit gamma integer, with 72x72 percision (even crashed my pc before I realized my GPU should compute it, not my CPU lol)

@Poodmund provide the instructions this post, I think, either way my Image is the same orientation, like exactly thanks space engine, as the 6000 cubic image, but mine is 8192 cubic.

  • I think this are NPOT (not power of two) textures as correct? For 16:9 I'm still keeping the "DIVISIBLE BY 8" rule as there is no advantage gained with other "NON-DIVISIBLE BY 8" sizes. For 4:3
  • I CAN NOT (I tried, insta crash to POST, OC'ed, normal, even under clocked just incase NO GO!!!) generate a 10K to 16K image...I just dont have the power...unless someone here is a  "Scotty level engineer" with a Ryzen 9 based system thats willing to help me generate it. 8192x8192 almost crashed it....
  • I'm going to try 9216x9216 to 9984x9984, in +128x128 size increases each try, to see what the "limit" of my pc is for this task

They take 5 min to save/export on a Ryzen 5 3600XT @ 4.85Ghz with a GTX2060 doing the lifting, each image ate about 1800mb of ram to "load" RAW also

Worked out to a few 100 Kb's shy of 90MB each... they are STUNNING! I am currently making the EnvMaps as we "speak"

Edited by Aazard
res's
Link to post
Share on other sites
2 minutes ago, Poodmund said:

You don't need an alpha channel for skyboxes so exporting and using them as DXT5 is a waste of resources.

OH PLEASE, great @Poodmund enlighten me, seriously.

Wow I cant believe you answered me lol. So what is the ideal setting it I was aiming at making a 8K (8192x8192) to 16K (16384x16384) skybox as best as possible?? detailed as you can plz

Link to post
Share on other sites

Well the skybox texture is just a flat image... why would you need the alpha channel, the shader does nothing with it so either just use DXT1 or if you don't mind it being greyscale then L8 is even cheaper.

The Unity player running in DX11 has a maximum texture size of 16384 x 16384 so go nuts. You can even disable mipmaps on your texture export so it may look like the small pixel stars twinkle as you move the camera and change the FoV... although this would mean that all 6 16384 x 16384 textures are kept loaded at all times in flight.

 

Link to post
Share on other sites
8 minutes ago, Poodmund said:

Well the skybox texture is just a flat image... why would you need the alpha channel, the shader does nothing with it so either just use DXT1 or if you don't mind it being greyscale then L8 is even cheaper.

The Unity player running in DX11 has a maximum texture size of 16384 x 16384 so go nuts. You can even disable mipmaps on your texture export so it may look like the small pixel stars twinkle as you move the camera and change the FoV... although this would mean that all 6 16384 x 16384 textures are kept loaded at all times in flight.

 

is that true to life realistic? ps answer here plz, what about a DX9 (dx10?) or openGL user? and valucan support? ps i know nothing lol

oops you did answer here lol

Edited by Aazard
Link to post
Share on other sites

True to life realistic? It's a video game. We all have different resolving power with our eyes and different focal lengths. It's all relative, realism is literally in the eyes of the beholder.

DX9 is no longer supported by KSP and Vulkan never has been officially. OpenGL should now only be used on Linux and Mac and seeing as the unified camera only exists for DX11 at the moment I have little care to worry about any other graphics API.

P.S. I knew nothing once.

Edited by Poodmund
Link to post
Share on other sites

ok so this setting:

Image: (working from the RAW 8192x8192 images, in GIMP)

RGB not RGBA,   RGB 24 bit not 32 bit, RGB Compressed DXT1 (BC1),

Mipmaps:

no gamma correction, Kaiser filter, wrap: default, do not preserve alpha?

then same for Envmaps but image "flipped vertically" as 1024x1024 to match a 8192x8192 skybox (1/8 the size is standard? or no?)?

-------------------------------------

what is DXT3 (BC2)? can i just use "RGB Compressed BC6H" or "RGB Compressed BC7" instead, the quality is better and its DX11+? no AMD/Nvidia caards after 2014/2015 are DX10 (the dx10 gpu families were retired then), I dont mind including both as optionals if its an issue to older GPU users

read: I am teaching myself as I go, I didnt expect it to be so "curved", but w/e I'll get it. easier than powershell or old CP/m

Quote

For devices with DirectX 11 or better class GPUs, where support for BC7 and BC6H formats is guaranteed to be available, the recommended choice of compression formats is: RGB textures - DXT1 at four bits/pixel. RGBA textures - BC7 (higher quality, slower to compress) or DXT5 (faster to compress), both at eight bits/pixel. HDR textures - BC6H at eight bits/pixel. If you need to support DirectX 10 class GPUs on PC (NVIDIA GPUs before 2010, AMD before 2009, Intel before 2012), then DXT5 instead of BC7 would be preferred, since these GPUs do not support BC7 nor BC6H.

 

Edited by Aazard
Link to post
Share on other sites
56 minutes ago, Poodmund said:

Well the skybox texture is just a flat image... why would you need the alpha channel, the shader does nothing with it so either just use DXT1 or if you don't mind it being greyscale then L8 is even cheaper.

The Unity player running in DX11 has a maximum texture size of 16384 x 16384 so go nuts. You can even disable mipmaps on your texture export so it may look like the small pixel stars twinkle as you move the camera and change the FoV... although this would mean that all 6 16384 x 16384 textures are kept loaded at all times in flight.

 

ummm can your system generate said "RSS Correct" images at 16384x16384? my system seems to dislike anything past 8192x8192 (I crash 100% of the time at 9728x9728 and up, over 9184 cause a 80% crash to post, if image completes its "screwed up" 9176 causes same crashes but if it completes it looks right), and upscaling it to 16K looks like....upscaled.

I can research almost anything but a better CPU/GPU and more RAM lol. a 3600XT, GTX2060 and 32GB-RAM cant handle it (and I'm honestly SHOCKED!)

A "master" set of 16K images would give me a "starting point" in testing further, 8192 is ok but I dont want to base all my work off it

Edited by Aazard
Link to post
Share on other sites

oh yeah..

Resize Algorithm: Kaiser (now fixed in unity) vs Lanczos3 vs Bilinear vs Mitchell.... 

Reading tells me that "Kaiser" is "Superior in every way" to all but  "Lanczos3, but only situationally", "Bilinear" is ok but inferior to "B-spline" and "Quadratic" (in that order) and that its the "default setting" for unity. Almost everything I read argues over Kaiser (now fixed in unity) vs Lanczos3 vs Mitchell, Suggestions?

resample_01.giflanczos3 is good, like wow, but kaiser is "sharper" esp on high contrast

images_contours_and_fields example code: interpolation_methods.py —  Matplotlib 1.5.1 documentation

I love manuals lol, so GL3/4 and DX10/11 "handle" 8K & 16K respectfully. Lucky DX9 handles 4096 sp it can atleast "handle" what I'm using as "default". I'll make sure to include  a 2048 max set also just incase, its only a minor effort

Note: Specific platforms may impose maximum Texture dimension sizes. For DirectX, the maximum Texture sizes for different feature levels are as follows:

Graphics APIs / Feature levels Maximum 2D and Cubemap
 texture dimension size (px)
DX9 Shader
 Model 2 (PC GPUs before 2004) / OpenGL ES 2.0
2048
DX9 Shader Model 3 (PC GPUs before 2006) / Windows Phone DX11 9.3 level / OpenGL ES 3.0 4096
DX10 Shader Model 4 / GL3 (PC GPUs before 2007) / OpenGL ES 3.1 8192
DX11 Shader Model 5 / GL4 (PC GPUs since 2008) 16384

 

https://en.wikipedia.org/wiki/Window_function

^----- super informative

Edited by Aazard
Link to post
Share on other sites

Ok I have got the manuals for: ( am reading unity now),

  • Unity <-- so i understand what I'm doing (because I only "sorta" understand as it is)
  • Notepad++ <-- I know "real" text editors well, but I like notepad++ macro support alot (or I did at one point in time, in 2012 to 2015, with geany)
  • XnView Classic & XnConvery <--- used to them for "a long time" as my foss viewer/converter for "daily tasks"
  • Irfanview + primer on what are "S3TC algorithm's" and what they are used for/examples <-- OMG i like the details, but its alot to insta absorb 
  • DDS converter <--- well...., but its functional/fast
  • Paint.net (and the "Exceed APS" plugins pack) <-- I used it before, but not at this "level of function"
  • Artweaver <--- free, nice, functional. Its good for a novice that gimp freaks out
  • AwesomeBump <---- nothing else "local" compares as foss (unless I install linux on a 3rd ssd, and...no) from my research
  • GIMP <---- I dont like it, never have. But I need to "see" the mipmap layers for image proofing, and paint.net doesnt do that (does it)
  • Krita <-- maybe with it and paint.net we can "not mention GIMP at all"
  • DarkTable <---- this is very neat, but I'm clueless at the moment with it
  • Inkscape <--- as above but seems "easier" to grasp at a glance
  • Nvidia Texture Tools < nice, like insanely nice it does BC6h/BC7..... how is ONE 8K image filling 8GB of vram, when I process to anything "better" than DXT1-BC1?
  • Scribus <-- for offical non-txt docs "fancying"
  • Space engine <--- I'm missing something, I'm sure... why the 8K "limit on no crash" or its just the limit of my personal PC

I am also reading "starter" art guides.... Like I didnt even know what "alpha" was talking about b4 or that DXT had like 10 format types with different "uses" in unity

I knew jpegs were "typical" pics and png was "pro pics", and that dds was special pgn, but xfc and psd are odd

OMG mipmaps filters... so iits basicly "default" (like just leaving it as it auto populates, normally refered to as "box" or "bicubic"??) or Kaiser (now fixed in unity) vs Lanczos3 as the "good picks" ppl talk about.

Note on "lowHalo and/or NoHalo" <-- what algorithm "type" so doing the generation of the resize, same as Lanczos3 or Kaiser, or diffrent like "B-Spline"?

Why was Lanczos3 removed from so many open source programs recently???? what is alt > NoHalo?

As you can tell I dont truly have a good grip on this yet

ps if its not FOSS i dont really want to suggest it to users: 

NO  "Step 1: Ok get $400 for software"! I'd like users to have access to the same tools I'm going to use in how too PDF's/Videos.

If they get the mod, I'd like them to be able to access all the tools via links, no fees/nags and NO BUNDLE-WARE. "FREE BEER" type Open-Source perfered, or atleast "free for personal and/or non-commercial use". Looks like it shouldn't be an issue as long (I hope)

 I hate to suggest GIMP! But I cant find alternative past "a patch work of apps" and I'm not going to make a suggestion like that to users. Paint.net is missing like 1 needed feature, and Krita is new (like a few days) to me

Edited by Aazard
explained why/what
Link to post
Share on other sites

REAL (lol, oops I thought I understood 99%, I didnt) mip map generation tests

..... these are my best results out of about 50 passes in each app, trying out settings.

The images, other than master, are DXT1-BC1 with DXT10 header, best/slowest settings. no alpha anything "checked", linear gamma corrected to sRGB format < I hope thats right

examples: all 8192x8192 about 1850%  about 200, 50 area of image, all viewed in GIMP for access to mipmap layers of all, Windows sees attributes as "identical"

***added avg times***

Which looks like the "best result" to people that are in the know. My "gut" says its "nvidia tool on kaiser" is best as it has "depth" and is DARK, but I know jack.

Master image unzoomed full screen:

Spoiler

v8Hm1pk.png

Master image: (as a 256Mb dds already tga was over 2x that compressed) about 1850%, obviously has no layers of mipmaps yet

Spoiler

q4EWyn2.png

 

mm layer "1" Gimp made, Kaiser 42.6MB <--- took 5 minutes

Spoiler

cmY4Tut.png

mm layer "1" Paint.net made "Faint" (is this "LowHalo" or Mitchell?) 42.6MB <--- took 11 seconds....UMMM why? I tested this MANY times too

Spoiler

ilrRt0s.png

mm layer "1" Nvidia tool made Kaiser 42.6MB (whites are "cleaner", blacks are "blacker", "full pixel boarder" seems more emulated on mips at insane zoom 1600%+)
^----
 took 4 minutes

Spoiler

kfR3DAW.png

Nvidia tool (stand alone) created mipmaped full image, from master, 42.6mb

Spoiler

itWbElw.png

 

edit:

testing quality over and over, I go out for coffee, sit at pc when I get in and see... The tilt is not right?!

Edited by Aazard
OMG i'm a fool!... i think?
Link to post
Share on other sites

Go after a ton of reading i got them "made" how i think is right,,,,

  • I now have a "working understanding of color as its "represented" digitally, like what alpha was
  • A"pretty good" grip on DXT1 (of various BC ranks) through DXT5 (of various BC ranks) and their used or why BC6f has sf16/uf16 or how BC7 can use or not use alpha channels (but I CANT MAKE IT HAPPEN lol)
  • I relearned python, bash and powershell, brushed up on genie (switched to notepad++ as its better)
  • found all the tools/plugins i needed as MIT or "un-licence" just in case (most suggested "real programs" by me will be apache/gpl)
  • found all the texture tools i needed for upto "real working" BC7 support
  • recaptured master as "uncompressed tga" this time at 192mb each or 1.91gb total,
  • made dds BC6h-uf16, mip > triangle, filter > clamp, precision > uniform weight, compress: max, resource: CUDA... (turned a 14min cpu encode to a 40sec gpu encode)
  • I tested to work at 8192x8192 as 64mb per image or 384mb total, I saved 1/3rd file size and got a 15% or so "better looking" image (with correct tilt of milky way)... after 12 hours or so of straight read > try > read > whats sf vs uf > learn > read > whats "xzy" > try > coffee

This gives me best result for file size (best result over all, much nicer then any DXT1 or DXT5-BC3

PS why are open source texture tools with "good" DXT "past basic 1" texture tools hard to get and all command line/python driven? :(

Almost nothing supported BC7 and BC6h (that really worked with a usable image), but the effort seems worth the headache

So question is.. do I want mipmaps now or not, readin tells me they may load if anything but "full texture resolution" is selected and that would like 1/4 the quality(size) of image displayed?

or am I banished back to "the manuals room" for a day?

Edited by Aazard
made point form for readability
Link to post
Share on other sites

Ok, better grasp of converting textures for unity and some victories. Switch alot of the "new to me tools" to what seem the best options from my tests. Issues/wins mixed around.

So I kept crashing trying to generate a "larger" master in SE,

I found my issue, I think.... vram

 

Watching my cpu/ram gpu/vram in a system monitor I see the following,:

System doing the work:

Spoiler

OS: v2004 Win10, B550m chipset, 3600xt on 3rd party air, 32gb ram, a 2060 with 6gb on NSD v452.06 and 16gb, max setting for B550 as "shared", nvme ssd, and a 3tb hdd. EVERYthing is basicly just idling but GPU maxed out, watching it try to fill vram it returns a compression error on load

SO.... I believe this limits my "highest" successful (like no crash OR "artifacts" in capture) capture in SE to =  8960x8960, which is an odd "size/number". 

But I can process this image into anything, at its original size or "smaller", for a format that should work, and it does, so thats a victory! The DXT for color maps/morals seem pretty simple to research other than the "BC6/7" types.

 

blah blah,

So I have a master image, its not perfect but its divisible by 4 and it looks good.

I can even scale the resize up with little difference at "under 40x40 pixel areas", like single points of light or color on the image, BUT:

IMPORTANT PART: >> Seems I can not use my system (due to vram limit) with more advanced "rescaled to 16k" (16834x16834) image sizes D...namely with BC6h_sf (although I dont think I want that for my use) & BC6f_uf (which oddly use the most resources to work with, I would have guessed the higher number = more ram space and time/procession needed). I hit a vram "near cap" of like 100% "full", at 16380x16380 but the images are fine, but thats ALOT of missing pixels off when you do the math versus 16384x16384. 

BC7 any mode I can handle, at 16384x16384 BC7_uniform (both no alpha component modes) I dont have a vram issue, which I dont really get, is the "alpha" component of image really that "big in memory" or maybe its how its compressed?

The BC7 image is larger file size at same image size, AND they look identical to my eyes in every way.

BC6h says it can do HDR though, where i only get RGB or RGB(A) (as a "color map" used in this case) out of BC7 or under.

So am I geting an advantage with BC6 & BC7, for my uses, over DXT4/5 or even no alpha type DXT1? I get that DXT1 will load with least memory used for image size.... BUT what format gives me the best "ratio" of file size to quality of result.

_________________________________________________________

PS:

I can make various versions now, from the masters, at least with "great" results, compared to my earlier ones, upto 8k... so I might hold off on the 16K image until I get a GPU with 8GB+ vram.

On mipmaping, I have the images with both mipmaps and with out. As the image becomes "larger" it seems to be better with out? or do I just not have the "PC/gpu power" to view my own work lol.

Sorry to treat this as a "my own thought/learning" post area the last few days... but maybe one day this thread could help someone else as clueless as I am lol

Edited by Aazard
spelling, pointed out "my points/quests" in my mad rant
Link to post
Share on other sites

Ok, Anyone wanna check my work out, here are some files

The non-mipmaped dds files 1536 to 16384 (except 2 of the I mentioned last post), all in DXT1-BC1-U, DTX5-BC3-U, BC6h-sf, BC6h-uf, & BC7U-M6, NO mipmaps versions. 
Note:
 if "U" is extra, its for "uniform precision", if "M" is added its mode

https://www.dropbox.com/home/TTL-MW-SB_Files?preview=DO_NOT_HumanStuff_TTL-MW-SB_No+mips.7z

The max size masters, lossless tga compressed, at my "max capture" of 8960x8960, already flipped  (so my batch file doesn't need it included)
Note:
 Gimp sees this as "sRGB default", but picking any "srgb" settings causes WILD alpha bleed

https://www.dropbox.com/home/TTL-MW-SB_Files?preview=001_NO_TOUCH_TGA_MASTER_FLIPPED_TTL-MW-SB.7z

feedback? Are my results even close to other ppl's, in all your opinions

Edit:

ok testing I dont take any "huge" ram use hits, DXT1 uses the least as I expected from forums/research. Reading the unity manual  (for Unity 2019.2.  as it seems ksp 1.8+ uses Unity 2019.2.2f1) I see: I have removed non-WIN/Mac/Linux PC OS's cause the rest of info might be confusing:

 

 

Spoiler

 

The following table shows the default formats used for each platform.

Platform Color model None Normal quality (Default) High quality Low quality (higher performance)
Windows, Linux, macOS, PS4, XBox One RGB RGB 24 bit RGB Compressed DXT1 RGB(A) Compressed BC7 RGB Compressed DXT1
  RGBA RGBA 32 bit RGBA Compressed DXT5 RGB(A) Compressed BC7 RGBA Compressed DXT5
  HDR RGBA Half RGB Compressed BC6H RGB Compressed BC6H RGB Compressed BC6H
WebGL RGB RGB 24 bit RGB Compressed DXT1 RGB Compressed DXT1 RGB Compressed DXT1
  RGBA RGBA 32 bit RGBA Compressed DXT5 RGBA Compressed DXT5 RGBA Compressed DXT5
Android (default subTarget) RGB RGB 24 bit RGB Compressed ETC RGB Compressed ETC RGB Compressed ETC
  RGBA RGBA 32 bit RGBA Compressed ETC2 RGBA Compressed ETC2 RGBA Compressed ETC2
iOS RGB RGB 24 bit RGB Compressed PVRTC 4 bits RGB Compressed PVRTC 4 bits RGB Compressed PVRTC 2 bits
  RGBA RGBA 32 bit RGBA Compressed PVRTC 4 bits RGBA Compressed PVRTC 4 bits RGBA Compressed PVRTC 2 bits
tvOS RGB RGB 24 bit RGB Compressed ASTC 6x6 block RGB Compressed ASTC 4x4 block RGB Compressed ASTC 8x8 block
  RGBA RGBA 32 bit RGBA Compressed ASTC 6x6 block RGBA Compressed ASTC 4x4 block RGBA Compressed ASTC 8x8 block
  RGBA RGBA 32 bit RGBA 16 bit RGBA 16 bit RGBA 16 bit

 

Various types of RGB(A) Compressed BC7 & RGB Compressed BC6H are listed as highest quality when the options of DXT1-BC1 or DXT5-BC3 are the "normal or low" quality option. But they, BC6 & BC7 need DirectX 11 to function or "fall back" to RGB half & RGBA 32bits, respectfully. <<--- is that worse than just having them as a "dx10/openGL loadable" format instead?

I got fall back info from the supported formats in unity part of manual also.

Spoiler

 

All supported Texture compression formats

The following table shows the Texture compression format options available on each platform, and the resulting compressed file size (based on a 256px-square image). Choosing a Texture compression format is a balance between file size and quality; the higher the quality, the greater the file size. In the description below, see the final file size of a in-game Texture of 256 by 256 pixels
.

When you use a Texture compression format that is not supported on the target platform, the Textures are decompressed to RGBA 32 and stored in memory alongside the compressed Textures. When this happens, time is lost decompressing Textures, and memory is lost because you are storing them twice. In addition, all platforms have different hardware, and are optimised to work most efficiently with specific compression formats; choosing non-compatible formats can impact your game’s performance. The table below shows supported platforms for each compression format.

Notes on the table below:

  • RGB is a color model in which red, green and blue are added together in various ways to reproduce a broad array of colors.

  • RGBA is a version of RGB with an alpha channel, which supports blending and opacity alteration.

  • Crunch compression is a lossy compression format (meaning that parts of the data are lost during compression) on top of DXT or ETC Texture compression. Textures are decompressed to DXT or ETC on the CPU, and then uploaded to the GPU at run time. Crunch compression helps the Texture use the lowest possible amount of space on disk and for downloads. Crunch Textures can take a long time to compress, but decompression at run time is very fast.

Texture compression format Description Size
for a 256x256 pixel Texture
Platform Support
RGB Compressed DXT1 Compressed unsigned normalised integer RGB Texture. 32KB (4 bits per pixel) Windows, Linux, macOS, PS4, XBox One, Android (Nvidia Tegra and Intel Bay Trail), WebGL

Note: With linear rendering on web browser that doesn’t support sRGB DXT, textures are uncompressed at run time to RGBA32.
RGB Crunched DXT1 Similar to RGB Compressed DXT1, but compressed using Crunch compression. See Notes, above, for more on crunch compression. Variable, depending on the complexity of the content in the texture. Windows, Linux, macOS, PS4, XBox One, Android (Nvidia Tegra and Intel Bay Trail), WebGL

Note: With linear rendering on a web browser that doesn’t support sRGB DXT, textures are uncompressed at run time to RGBA32.
RGBA Compressed DXT5 Compressed unsigned normalised integer RGBA Texture. 8 bits per pixel. 64KB (8 bits per pixel) Windows, Linux, macOS, PS4, XBox One, Android (Nvidia Tegra and Intel Bay Trail), WebGL

Note: With linear rendering on a web browser that doesn’t support sRGB DXT, textures are uncompressed at run time to RGBA32.
RGBA Crunched DXT5 Similar to RGBA Compressed DXT5, but compressed using Crunch compression. See Notes, above, for more on crunch compression. Variable, depending on the complexity of the content in the texture. Windows, Linux, macOS, PS4, XBox One, Android (Nvidia Tegra and Intel Bay Trail), WebGL

Note: With linear rendering on a web browser that doesn’t support sRGB DXT, textures are uncompressed at run time to RGBA32.
RGB Compressed BC6H Compressed unsigned float/High Dynamic Range (HDR) RGB Texture. 64KB (8 bits per pixel) Windows Direct3D 11: OpenGL 4, Linux.

Note: BC6H Textures are uncompressed at run time to RGBA half on the following platform configurations:
- macOS with OpenGL
- Platforms with Direct3D 10 Shader Model 4 or OpenGL 3 GPUs.
RGB(A) Compressed BC7 High-quality compressed unsigned normalised integer RGB or RGBA Texture. 64KB (8 bits per pixel) Windows Direct3D 11: OpenGL 4, Linux

Note: BC7 Textures are uncompressed at run time to RGBA 32bits on the following platform configurations:
- macOS with OpenGL
- Platforms with Direct3D 10 Shader Model 4
Platforms with OpenGL 3 GPUs.
I removed the formats not supported on: Windows, Mac OS Linux and Other "non PC" Systems, to avoid confusion
RGB 16 bit 65 thousand colors with no alpha. Uses more memory than the compressed formats, but could be more suitable for UI
 or crisp Textures without gradients.
128KB (16 bits per pixel) All platforms.
RGB 24 bit True color, but without alpha. 192KB (24 bits per pixel) All platforms
Alpha 8 High-quality alpha channel, but without any color. 64KB (8 bits per pixel) All platforms.
RGBA 16 bit Low-quality true color. This is the default compression for Textures that have an alpha channel. 128KB (16 bits per pixel) All platforms.
RGBA 32 bit True color with alpha. This is the highest quality compression for Textures that have an alpha channel. 256KB (32 bits per pixel) All platforms.

You can import Textures from DDS files, but only DXT, BC compressed formats, or uncompressed pixel formats are supported.

 

 

So I guess the question I  want to know for the users of the texture file in question, the skybox... what is the "ideal balance" of size vs quality on this?

Edited by Aazard
Link to post
Share on other sites

preview, of DXT BC1 image "PositiveZ".  8mb @ 4096x4096, as a png image used to upload to gallery:

still working on turning dds compression to reduce "wash" at rifts dark core transitioning to a bright area

Spoiler

nkBL49a.jpg

same image as 16mb BC7

Spoiler

MjfuoCI.png

in game:

Spoiler

jV0DSqB.png

 

Edited by Aazard
Link to post
Share on other sites

Good news!:D  Not 100% original but it was a great corpse to build Frankenstein's monster from

We have a HumanStuff Wiki NOW !:confused: <-- All original "still usable" information compiled by:

 

May you rest in peace @CaptainKipard :ph34r:, surely eaten by an @Tiberion.... so sad we have lost you guys...tooo 2015/2016 all over again @RoverDude where are u?  @Yargnit are you still around? @linuxgurugamer and @NathanKell (I truly enjoy your YT content btw) are still around I think. @sarbian blesses us still, may Kerbol shine upon him

What happened to all the MASTERS, new students are climbing the mountain, seeking you

Most tutorials are old and/or links are dead, software suggested is vastly outdated, from the era when lots of "awesome teaching content" was posted alot has changed, and many that supplied them have left us.

I'm "interested" but seriously lack the skills or know how to recompile all this data.

Sorry to ask, but I am really at my current skill limit here. I'm learning "what I can, as quickly as I can", but my "skill cap" isnt going to hit "guru level" or anything. I'm straight up adding content as I learn it

Is there a current resource past the "older doc" in the locked thread?

Edited by Aazard
added a plea
Link to post
Share on other sites

Hi, we received a redundant pull request to add another copy of this mod to CKAN.

You don't need to do that, everything works fine if you upload new releases to the original mod.

This extra copy should probably be deleted:  https://spacedock.info/mod/2529

Link to post
Share on other sites
13 minutes ago, HebaruSan said:

Hi, we received a redundant pull request to add another copy of this mod to CKAN.

You don't need to do that, everything works fine if you upload new releases to the original mod.

This extra copy should probably be deleted:  https://spacedock.info/mod/2529

Which copy will show up, AS COMPATIBLE (like if CKAN settings are set for mod compatibility with KSP v1.8.1 only), for users of RSS, RO, RP1 (whom are using KSP v1.8.1 for RP-1 Realistic Progression One v1.6 for KSP 1.8.1 compatibility)

They are identical, but one is versioned for KSP v1.8.1 and one for KSP v.1.10.1

Thats the intended "main user group" the mod targets.

Edited by Aazard
added missing "CKAN" & highlights
Link to post
Share on other sites
30 minutes ago, Aazard said:

They are identical, but one is versioned for KSP v1.8.1 and one for KSP v.1.10.1

Given that they are identical, you could add a so-called version file as defined by AddonVersionChecker (AVC) to the zip, specifying 1.8 as KSP_VERSION_MIN and 1.10 as KSP_VERSION_MAX. CKAN can read this file and gather its compatibility information from it, allowing RO players to install this mod on 1.8.1. This is the usual and preferred way to specify multiple compatible KSP versions, and it's probably also easier to maintain (both for you and for the CKAN team).

There's some more (although partially outdated) information about .version files with an example here, or if you prefer a more technical definition, here's the schema.

Feel free to ask me or @HebaruSan if you need help setting up a .version file.

Edited by DasSkelett
Link to post
Share on other sites
1 hour ago, DasSkelett said:

Well, I've decided to take the work off your hands did a PR to include such a .version file in your repository: https://github.com/aazard/HumanStuff/pull/1

You are awesome, merged!

Sorry to be a noob after all the years here (Purchased: Jul 21, 2013 @ 11:41amKerbal Space Program - $13.79 USD) but 1st time modding for others... ummm next steps needed? New release with merge in it and push "1" version to space dock set to "v1.10.1" or for "v1.8.1"?

forgive delay html markdowning the github wiki )...blah...)

Edited by Aazard
Link to post
Share on other sites
This thread is quite old. Please consider starting a new thread rather than reviving this one.

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...