How to stop super shininess on textures?

[AluminumHaste]

Looks like shadermap also supports Multi-Angle Light Scan, so you can get better results with a setup like this:

[Epifire]

2 minutes ago, AluminumHaste said:

Looks like shadermap also supports Multi-Angle Light Scan, so you can get better results with a setup like this:

Ooookay, yeah that must help with depth processing. to have multiple light capture angles of the same object. I saw one video on SM4 where the guy was just using it like the old Crazybump. The tripod setup looks like it would do a hell of a lot of good.

For me I'm starting to branch into making my own tiling textures again. As trimsheets and just full tiled materials are actually becoming much more feasible in modern engines. As a traditional old-gen modeler, its strange to me to see a detail mesh get four different tiling materials to build it's primary details. 

This certainly is interesting territory we're getting into for games. Glad we got your main issue figured out though and thanks for elaborating on the process!

[nbohr1more]

I guess the best image-to-normal conversions I've seen here in the forums are via njob.

I am curious about this AI thing though:

https://github.com/HugoTini/DeepBump

has to be installed into Blender as a plugin?

[peter_spy]

10 hours ago, Epifire said:

The spec is basically on point with the original image (for overall tone). Really the only thing standing out to me now is that there's something funky going on with the normal map. The grout work was treated like a ridge, rather than a valley; so I'm assuming that has to do with the way the normal map was generated.

Yup, there's something off with the normal map; I actually inverted at least one channel in order to get the chipped-off edges right. The normalmap looks like just generated from a grayscale diffuse, so the white edges were convex and dark bits were concave. Isolating the bright bits, copying them to a separate layer and inverting the color would do the trick, but I thought that might be a bit too much of manual work.

Edit: I tried to do this^ quick and dirty way, https://we.tl/t-jNZdpmkJtF

And this would be the result:

Edited February 9, 2023 by peter_spy

[datiswous]

8 hours ago, AluminumHaste said:

The one where you take multiple images of the subject with lighting from all angles didn't work, there was too much noise.

I guess this method:

https://shadermap.com/tutorials/normal-map-photography/

[peter_spy]

Jeez, that's awfully time-consuming. No wonder photosourced materials stopped being a thing for quite some time now.

[AluminumHaste]

3 hours ago, peter_spy said:

Jeez, that's awfully time-consuming. No wonder photosourced materials stopped being a thing for quite some time now.

It is, but there's a note in there:

Note: This method of generating normals from photographs has been replaced as of version 4.1 with the Light Scan Node.

EDIT: but if you do it right, it has really good results as the normal map is built from the height data generated from all the images.

[peter_spy]

But in general, the big problem with photos is that you either have to have ideal conditions when capturing them, or you have to do tons of editing, like removing shadows, getting them to tile, etc. And you'll still end up with some sensor noise, and they will be in one resolution only.

Procedural generators like Substance, or other counterparts, have the advantage of being resolution-independent, and all the components can contribute to generating proper height map, which in turn is used as displacement map for automatically subdivided mesh, which you can bake normals from. All this while being tileable at all times.

[Epifire]

7 hours ago, peter_spy said:

Jeez, that's awfully time-consuming. No wonder photosourced materials stopped being a thing for quite some time now.

It's definitely still a really big deal but normally I've seen specific artists/studios have a streamlined setup for this. While I'm a Substance guy through and through right now, I can still see a benefit of grabbing accurate, real-world data to layer into a complex material. On games that are gunning for photorealism, I've even seen Photogrammetry experience listed as a plus for hiring.





 

[peter_spy]

In a long time, I don't think I've used photos for anything more than a source for a color picker, so I can get a gradient ramp, or a just a few colors to use for my base color texture. Photogrammetry is indeed useful, but in terms of capturing a photo-scanned 3d model and making it game-ready. Since most PBR materials are standardized and look similar, games tend to be slightly stylized anyway, as everyone wants to have some visual identity. Photorealism ceased to be a holy grail of games quite some time ago too, it's more of an arch viz thing (and maybe racing games, although that's mostly cars).

Edit:

Out of curiosity @OrbWeaver, how specularity works in this engine? Did TDM team use what was there by default with idtech4, or did you implement your own tweaks? From what I've been experimenting with, it's different than other non-PBR engines, as it's not just a straightforward add operation, am I right?

Edited February 9, 2023 by peter_spy

[nbohr1more]

As I recall, when JC-Denton did the HDR-lite implementation he made the following changes:

1) Increased the default specular response and added a simulated light bounce per the Doom 3 Brilliant Highlights mod by Maha-X

https://www.moddb.com/mods/brilliant-highlights

2) Added a glancing angle fresnel response

3) Added a small default specular amount to materials that have no specular so that some of the lighting effects would work better

4) Mixed in some of the diffuse color into the specular to slightly approximate radiosity effects

As of 2.11, 3 and 4 are no longer applicable but 1 and 2 still happen.

[peter_spy]

That would confirm my observations. I think that removing 1 & 2 might be beneficial as well...

1 –  because it would give creators more of the specular texture color range to use. Currently, if I go above ~ RGB 190 intensity, the specular hotspot will look overblown (with no bloom applied)*

2 – because it increases the specular hotspot saturation at grazing angles, making it harder to control. You can compare that by using the dielectric specularity trick, where you invert the color of the diffuse texture to get white specular hotspot. This is how the result looks like:

At straight angle in the center, everything looks correct:

At grazing angle:

I guess that you might want to apply that to cubemap reflections, not specular hotspot.

*That said, this is my setup only, it's all relative to light intensity, radius, etc.

Edit: would it be possible to toggle the two HDRLite features via cvar? I would be interesting to experiment with this and see how it affects making materials, for both flat surfaces and models.

Edited February 10, 2023 by peter_spy

[OrbWeaver]

I agree. It seems that artist control of specular has been removed for the sake of "cool HDR effects". If a mapper wants to make a texture with zero specular, it should have zero specular in the engine.