You know you are a proper graphics nerd when you look at a texture like that and get warm fuzzies for how pretty it is…
I keep thinking that there’s got to be a better DXT1 compression (besides the known DXT5 variant) for Normal maps… something about color-space-axis mapping taking into account the post-pixel read normalize applied to the values…
I had no idea that Unreal engine actually biases towards not compressing Normals and instead just half-res sampling them - same space used, and they assert the quality is higher. I may experiment with adding a mode for that to the SHADERS.MAP processing…
I suppose I should clarify: I’m referring to my question here: EXAMPLE - Ships #1
To which you replied:
And then EvilleJedi said:
Sooo, what I’m doing now is that I’m blacking out the seams, ending up with a primitive black and white bump map (the diffuse is just a template) and just converting it to normal map with PS NVidia normal filter. Because the 1999 look of the Xnormal’s UI put me off and because it didn’t even seem to do what I needed (it seems to want highpoly meshes?)…
(That’s why I asked about white and grays, not blues.)
I didn’t know Ndo2, but your 128,128,255 and some exaples I just googled made it clear you’re actually drawing into the normal map itself. That’s interesting. Maybe I should look into it?
BTW that Hiigaran BC normal map is beautiful.
EDIT: Looking at your normal map it just struck me, there’s no base level. It doesn’t really make much sense to convert bump/displace map, when the nature of normal map is just each pixel having a normal information based on its color channel. There’s no higher or lower place, you can stack panels on panels to infinity, it’s all just local.
So yeah, I should stop this black and white thing…
As I mentioned, so far I only baked them from high poly meshes to low poly meshes in Max, i didn’t realize this…
Xnormal is UUUUUUUUUUUUUUGLY - but works great, has tons of features. Cole would know it better than I do - it does (as one option) just bump->normal, no mesh or anything else required.
With a normal map there is no ‘base’ level - aside from 128,128,255 - aka 0,0,1 - aka ‘normal away from surface’.
With a bump map there is also no firm base level - because if you set it at 128 (aka 50% of whatever 255 is in terms of offset) - then all of the local deltas are zero - Xnormal would return a normal map with 128,128,255 for the whole thing. You can, from there paint up (->255) or down (->0) and emulate what Ndo2 is allowing an artist to paint directly. Me, personally, as a non-artistic ‘math’ person, I could emulate what Cole did in Ndo2 via a bump-map quite quickly - having ‘light’ panels layered on top of the 50% grey base with feathering to make ramp edges (and controlling the contour via the feather falloff curve), inset edges/lines with ‘dark’ overlay segments, etc.
There are many, many ways to approach this problem.
My only question would then be, with a bump -> normal map, is there any difference between the one with a white panel on 128 gray and the one with a white panel on black? Does it then mean that one wouldn’t be “shallower” than the other? Is that all just about the local difference too?
I’ll have to try that an compare. It’s the last piece of puzzle for me.
I think in Xnormal you can define a ‘scale’ to the depth - with the same scale your 2 examples would be fairly distinct… otherwise, with a scale correct for each respectively, the output would be (nearly) identical. It really does all come down to screwing around a bit to get your bearings
Exactly this. Sometimes I’m bad at remembering that only a tiny group of people were privy to the experimentations and theories behind the scenes. I guess we’ve never really talked about it. This isn’t a direct correlation to what you’re doing since you’re kitbashing but it might be useful to you or others.
What we did was take the original diffuses and scaled them up to the desired resolution. (All these examples have been truncated and reduced in size to fit the forum guidelines so you’ll have to imagine).
A snippet of the rear of the battlecruiser, original texture:
Drawing into the normal map:
Usually working from the outside in, determining what is a panel, what is a groove, if something is sticking up or is recessed, etc. Pay attention to how things are connected and how deep your cuts go and how thick you make your lines. Line width, density and how it relates to the scale of your ship are all things that you get better at with practice.
Once the normal map is done, independently generate AO, cavity on a 50% grey background:
(this whole image looks super dark in my preview pane so I’m monkeying with it. It’s not quite right but close enough to demonstrate)
Discarding the original and using the above as a base, repaint the entire diffuse from scratch to rebuild the original:
Paint between the AO/cavity and the base so that the lines are always on top, recreating and interpreting the original detail. Your diffuse should be far less contrasty than the original due to HWR’s lighting model and the interoperability of how the team/spec/refl/paint all affect your ship-understanding this comes with a lot of practice and experimentation.
Ok that is the ultra fast explanation, there are a lot of nuances in each step but that at least gets the basic workflow out there and might give you some options you hadn’t considered.
It’s pretty close to what is going in this video:
But with more Marquee and Polygonal Lasso and frequently straight up drawing on it with a brush.
Scole, this is some amazing stuff. Yeah I’m shamelessly kitbashing your textures (I hope it doesn’t offend you), but the AO trick can still come handy one day. And even if not, it’s really nice to see how it was done. This, after the ship examples, is the second best moment presenting the secrets of Remaster process. I just love it.