1 hour ago, Benjamin J said:

I can now sort of see the challenge in making PBR materials if you haven't been doing it from the start... Because all Youtube videos I've been watching simply take a material and apply it as a whole to an object. But, of course, when we model a terminal building, the glass, the metallic frames etc etc all sit on the same diffuse map, which probably complicates the entire process. Instead, if you could just paint the materials onto the model, which is what Substance seems to allow to do, it makes everything a lot easier... Again, did I get that right?

You can also allocate things inside Alpha channels, this will allow you to have texture metal parts, with roughness and other PBR attributes inside the metallic map.

Unity engine allows you to use this technique, were via the Metal texture maps you can use the RGBA channels to define PBR attributes, each games has it owns parameters for example  R=Metal, G=Roughness, B=Empty, A=Occlusion, etc.

Using the Red alpha channel to define metals, black colours will mean is not metal, were white colours means it is metal, and middle colours would mean a level of metal depending of the quantity of white.

But as Umberto pointed out, better results come when you have different textures for different materials as you can set individual parameters achieving better results, having said that, anything converted to PBR will react way better to ambient light using much less computing processing than the old technique.

Regards,
S.

 

 

Here's a nice website (it's actually a support topic for a product) that discusses PBR with examples:

https://help.sketchfab.com/hc/en-us/articles/204429595-Materials-PBR-

Autogen looks so dull... maybe PBR textures will mean shiny windows (and other metallic/glass surfaces)?

Thanks for the additional answers all! Some very helpful resources shared here. Definitely have a better understanding of PBR now, but I have one more point that confuses me. It's all about the application of those materials onto the models

So, up until now, we would generally map a model with textures made from photographs or that are hand-drawn, typically combined onto larger texture sheets. Of course these are not PBR materials, as they do not include any of the required maps to make it so. Like others have said, one could make these maps on the basis of the diffuse map (and that's what Materialize as well as Materialing Alpha, I think, both do), but it wont look as good as when you use actual PBR materials, made with the express purpose of PBR.

And that's where I get confused. My question, specifically: How does one map PBR-aware materials onto models, on the basis of photographs taken from a site? Given that these appear to be mutually exclusive...

To go with an example: say we model a window with a metallic frame next to a wooden door. What's the best way to map the textures, and retain the different material properties? Do we:

1. map as before, unwrapping the model to a single diffuse map, on which we combine PBR-aware materials (diffuse maps edited to represent the real-world counterpart, by changing color, for exmaple). The different maps (metallic, roughness etc) sport differential greyscale values to make sure that the door appears 100% rough, but the metallic frame does not.
2. or do we use 3 seamless PBR-aware textures that describe the separate materials? This approach will then not allow for baking of ambient occlusion and such things, as such effects are inherently not seamless.

I'm guessing approach 1 is still the way things are done. I hope the crux of my confusion is coming across and that I'm making some sense here?

3 hours ago, Benjamin J said:

1. map as before, unwrapping the model to a single diffuse map, on which we combine PBR-aware materials (diffuse maps edited to represent the real-world counterpart, by changing color, for exmaple). The different maps (metallic, roughness etc) sport differential greyscale values to make sure that the door appears 100% rough, but the metallic frame does not.

I'm guessing approach 1 is still the way things are done. I hope the crux of my confusion is coming across and that I'm making some sense here?

Yes, that's what you normally do, and it's what products like Substance Painter are designed for: to create multiple layers of different PBR materials, which use *masks* to map cover different parts of the object polygonal mesh. It's not so much different than having multiple Photoshop layers, each one with a mask (to create a composite) just that, in PBR, you are not just setting "colors", but also height, normals, metalness, roughness, opacity, emissive, all at the same time.

Approach #2 is still useful, for optimization. For example, if you have a transparent material you use only on a very small part of an aircraft/scenery, it's usually best to separate it and make it a different material and have it assigned it only to those polygons that are supposed to be trasparent, so you are not calling the more expensive shader (the transparent shader is more complex to draw) for the whole object, just because you have a small transparent window.

Edited November 29, 20187 yr by virtuali

Thanks for the helpful explanation Umberto, this helps a lot! If you don't mind, I sent you a PM with some additional questions.

Double post... Sort of. I've been looking into Substance a lot since the last post. It helps that there are beginner tutorials on YouTube, with some excellent information. It strikes me that the way to go about 'painting' in Substance Painter is really similar to what I've always been doing in Photoshop. Using material layers, you can mix and blend materials of all sorts to get exactly what you need, and then mask parts to make sure that you are not overlapping materials for walls with materials for windows frames and glass. A very powerful and, frankly, intuitive piece of software, me thinks. Finally, Substance painter can then export the 'flattened' texture (to use Photoshop terms) for use in P3D. I gotta say, at 150$ for a perpetual indie license, I'm sort of thinking it may be a worthwhile purchase, given the quality of work I'll be able to get out of it for, hopefully, years to come!

The one thing I now need to wrap my head around is how textures derived from reference photos play into this. I've gotten somewhat accustomed to e.g. copying a texture of an entire door from a reference photo and pasting it into the right position in Photoshop, then tweak it in ways that would make it fit in the texture (adjusting the color, for example) but this approach does not seem to be... available (or desired?) in this 'new' approach to texturing.

Anybody have any insight into this?

 

On 12/1/2018 at 12:39 AM, Benjamin J said:

The one thing I now need to wrap my head around is how textures derived from reference photos play into this. I've gotten somewhat accustomed to e.g. copying a texture of an entire door from a reference photo and pasting it into the right position in Photoshop, then tweak it in ways that would make it fit in the texture (adjusting the color, for example) but this approach does not seem to be... available (or desired?) in this 'new' approach to texturing.

Reference photos can still plays some part, by sticking to their name, and be used as a reference only. The old practice of sticking a photo on a 3d object it's showing more and more its age in the days of physically-correct rendering, it's really a legacy of the old days, were the graphic engine could only do very limited lighting, so you had to put as much information possible in the texture, to overcome the engine limitation. This was already an issue with FSX, but with PBR (and the availability of PBR painting tools), photos shouldn't be used in the actual texture. Just put them in a window on the side of the monitor, and look at them, to get a feeling of what *material* you should use. That's the only usage for photos I see now. A photo used as a texture is like a broken clock: it shows the correct time only twice a day...

Getting the hang of when using actual modeled geometry, or artificial detail added with normal/height maps is also key to efficient modeling. With "efficient" meaning both things: from a performance point of view, but also to how your time is spent.

And, with all PBR painting tools, the trick is trying to be as procedural as possible, and reduce manual painting as much as possible. Your goal should be doing 90% of the work done with pre-made materials (just learn to tweak their parameters to fit your project) and pre-made procedural generators like wear, scratches, dust, etc. and paint only the finishing touches, like spot corrections, decals, signs. 

Excellent, thanks for the detailed post Umberto! Then my confusion is resolved.