PBR
- Thread starter VFX_Veteran
- Start date
VFX_Veteran
Regular
Another shading feature that has yet to be addressed by game devs and seems to get ignored.. Refractions..
This uses a physically plausible Beckmann distribution for refractive rays and modifies the specular lobe based on roughness.
This uses a physically plausible Beckmann distribution for refractive rays and modifies the specular lobe based on roughness.
VFX_Veteran
Regular
I think you're getting a bit held up with the implementation in The Order - other games and engines could easily go further and do PBR more... "right", if there's even such a thing. Mind you, this is no longer a thread about that certain game
It doesn't matter if it's the Order or not. I'd like to see any game do some of these things correctly.
L. Scofield
Veteran
Check this presentation:OffModEdit: On topic but I was hoping you could answer me on this one since Josh Holmes confirmed Halo 5 is using PBR.
But how can you tell a game employs PBR by just looking at some in-game screenshots? What are some things to look for?
http://www.frostbite.com/2014/11/moving-frostbite-to-pbr/
But do they fool the layman's eyes? I'd say yes.
L. Scofield
Veteran
EDIT: Damn, double post.
VFX_Veteran
Regular
LOL! I'd say no. It's only a matter of time before the tricks are going to be blatantly obvious. We want games to approach CGI.. continuing to settle for a cube map instead of real reflections is going to hold you back..
Refractions are REALLY expensive. On a current project we've had to fix issues on teeth in comp because the proper number of ray bounces would've killed our render times completely.
It's one thing to get it to work on a single asset in a test environment and another to do it right in a complex scene with a character close-up shot. I'm not a fan of relying on post for this stuff but production realities can really be brutal. We've just been asked to add a character based on celebrity talent - 3 weeks before final delivery. I kinda feel like I'm Tom Cruise in Mission Impossible
It's one thing to get it to work on a single asset in a test environment and another to do it right in a complex scene with a character close-up shot. I'm not a fan of relying on post for this stuff but production realities can really be brutal. We've just been asked to add a character based on celebrity talent - 3 weeks before final delivery. I kinda feel like I'm Tom Cruise in Mission Impossible
L. Scofield
Veteran
I doubt most people would notice except in very specific circumstances such as characters lacking mirror reflections in TO1886. Some people even thought they were vampiresLOL! I'd say no. It's only a matter of time before the tricks are going to be blatantly obvious. We want games to approach CGI.. continuing to settle for a cube map instead of real reflections is going to hold you back..
This is a way more complicated issue than you or I would think first. I think I've mentioned in TO's thread that even my experience as a modeler is good enough for picking apart any character screenshot from that game - and yet most gamers would have a hard time to notice that stuff, even if it was pointed out with paintovers and such.
Finding the right balance between what would actually work well enough and what our instinct's drive us to do is incredibly hard. We always fight with the small details following the 80:20 rule of thumb, spending the majority of our effort on stuff that would only add quality increases of a single percent, not knowing where we could stop and still sell the shot.
We can see the cubemaps and feel bad about them - but if no-one else cares, is it really worth to render with proper glossy raytraced reflections? This is a choice that will soon make or break projects, or maybe we're already there... Is it worth rendering the proper way for a week, or is it better to tweak that cubemap for 5 more days and then render in 2? And can we final that stuff with a good conscience even though we know it's all wrong... but it'll pay the bills because the client is happy?
I doubt most people would notice except in very specific circumstances such as characters lacking mirror reflections in TO1886. Some people even thought they were vampires
That's the really funny part... when people are dedicated to the final product enough that they're willing to come up with explanations for the shortcomings
VFX_Veteran
Regular
Refractions are REALLY expensive. On a current project we've had to fix issues on teeth in comp because the proper number of ray bounces would've killed our render times completely.
It's one thing to get it to work on a single asset in a test environment and another to do it right in a complex scene with a character close-up shot. I'm not a fan of relying on post for this stuff but production realities can really be brutal. We've just been asked to add a character based on celebrity talent - 3 weeks before final delivery. I kinda feel like I'm Tom Cruise in Mission Impossible
Sent a PM.
Well, shit. I can't send PMs?
VFX_Veteran
Regular
Yea, I figured it out.
What game developers care about is the impact that the paradigm has on the development process. Overall expressiveness is nice and certainly worth talking about, but it's entirely possible to take advantage of the concept while using approaches that might not be as expressive as you might like.
Similarly, relaxed standards seem to be in use for "physically plausible" compared to what you're used to, if Lambert is to be treated as somehow "implausible." (It's not perfectly precise for any real material, but it's a pretty reasonable approximation for some, and it can play fine with the usual mathematical requirements of plausibility. Mostly it's just not that expressive, failing to capture the complex effects of roughness, and that sort of thing.)
Since it seems like you're referring to TO1886, it's worth noting that in their GDC presentation, RAD explicitly mentions that the hair model isn't physically-based.
VFX_Veteran
Regular
It's just a little weird for the term to be used to such a high degree for the sake of justifying a game's graphics over another. Crysis 2 used Oren Nayar and Cooke Torrance -- both physically plausible models but they never announced it as a buzzword.
Lambert is definitely implausible as there is no material in the real world that has no roughness. To add to that, we've had several problems with materials where roughness was 0 (thereby mimic Lambertian) and placed it in scenes that had several physical lights that caused the materials diffuse portion to blow out. I used to love Lambert.. but after using in a physically correct rendering pipeline now, it's hard to ever go back to it. I've even defaulted our physically plausible diffuse shader to a roughness of 0.3 just in case the artists overlook the setting (like they do 90% of the time)..
Right. so would you call their pipeline a PBR pipe when only 1 component is actually based on a microfacet distribution and everything else isn't?
Forum warriors do that all the time; game developers less so.
That's because they're using those BRDFs for expressiveness, but it doesn't satisfy other aspects of a physically-based development.
Bungie offers a great example of terminology use; they didn't say "physically based" when they implemented Cook-Torrance specularity in Halo 3, but they mentioned that they used a physically-based approach to a partial degree with Destiny, which uses an arguably less-expressive Kelemen BRDF. The derivation and complexity of the BRDF wasn't the difference with PBR-ish-ness; it was how assets were authored, and how shading was accumulated at runtime, to allow them to produce a substantial amount of content that would look consistent at runtime under their varied real-time lighting environments.
Last edited:
L. Scofield
Veteran
I'm pretty sure Crysis 2 used Blinn-Phong.
VFX_Veteran
Regular
VFX_Veteran
Regular
Yea, maybe I was thinking about Crysis 3?
