The Order: 1886

[L. Scofield]

PBR isn't just specular surfaces..

Pretty much all surfaces have a specular component:

http://filmicgames.com/archives/547
http://filmicgames.com/archives/557

Alien: Isolation does that.

Well, that's one. I guess it helps that it only has to render one room at a time.

Calculated what in realtime(i.e. environment, GI, direct light, etc)? I'm not sure I understand what you are getting at.

There's no GI in Crysis. The lighting consists of:

-Realtime direct light sources: directional, omni and spot.
-Realtime simplified light sources: omni-only, white-only, no specular component and no shadow casting ability.
-Realtime ambient lighting (flat color applied from every direction).
-SSAO.

I know.. but it's not old. Only within the past 2-3 years have we seen GI bounce being stored into probes and being sampled from dynamic objects. I, personally, started to see this with Farcry 3 and Crysis 3.

I'm not sure about Far Cry 3 but Crysis 3 calculates the probes in the editor, not at runtime.

 

[Shifty Geezer]

Pretty much all surfaces have a specular component:

http://filmicgames.com/archives/547
http://filmicgames.com/archives/557

Once again, you miss the argument. VFX was saying show a good looking material in I:SS that wasn't dependent on reflection (mirror-like, not specular highlights). Looking over some I:SS screenshots, when surfaces aren't wet, they are lacking realism. The Order has some better material shaders, although at times it still looks very 'Uncharted'.

 

[L. Scofield]

Once again, you miss the argument. VFX was saying show a good looking material in I:SS that wasn't dependent on reflection (mirror-like, not specular highlights). Looking over some I:SS screenshots, when surfaces aren't wet, they are lacking realism. The Order has some better material shaders, although at times it still looks very 'Uncharted'.

Which rough, dry materials look unrealistic?

 

[London Geezer]

Picture speaks 1000 words?

 

[HTupolev]

I know.. but it's not old. Only within the past 2-3 years have we seen GI bounce being stored into probes and being sampled from dynamic objects.

I'm confused. You didn't deny the validity of my example*, yet you still insisted that this stuff has only been done in the last couple years. But the example I mentioned came out over seven years ago.

*Which you probably could, to an extent; Halo 3's specular-contributing environment "probes" seem to consist of dominant-direction data (which works as long as the environmental lighting is fairly directional, but has a lot of unmitigated quirkiness like cancellation and tending toward the sky) augmented by (box-projected or similar?) cubemaps (which capture the distribution "correctly" but aren't all that dense in the environment and are treated as a minority contribution).
I won't deny that things have been continually refined.

Pretty much all surfaces have a specular component:

http://filmicgames.com/archives/547
http://filmicgames.com/archives/557

There's something bizarre about the idea that anyone would even need to justify Fresnel being useful for non-metallics.

 

[VFX_Veteran]

Pretty much all surfaces have a specular component:

http://filmicgames.com/archives/547
http://filmicgames.com/archives/557

In the real world -- of course! But we both know when we approximate materials, we have a diffuse component and a specular component. Objects that have materials where diffuse component dominates will be hard to notice any specular. Some of our artists have specular component of 0 for certain materials. So assuming Infamous:SS follows the conventional pipeline, I'm sure there are surfaces that only have a diffuse component (or the diffuse is dominate during energy conservation). Go look at the crappy building materials in ISS -- those are diffuse only. And the PBR on those look terrible (if the material is even following an Oren Nayar/Henyey-Greenstein diffuse model). They look like Lambert to me (which isn't physically plausible).

There's no GI in Crysis. The lighting consists of:

I was speaking about the environment HDRI lighting. I guess C1 doesn't have it.

I'm not sure about Far Cry 3 but Crysis 3 calculates the probes in the editor, not at runtime.

FarCry 3 does.. I took a video of it:

 

[VFX_Veteran]

Which rough, dry materials look unrealistic?

Here is a screen of a building like I was talking about.

That material looks flat and not very good texture work at all. Every piece shades the same even though they are different materials for the bricks and window seals.

 

[VFX_Veteran]

I'm confused. You didn't deny the validity of my example*, yet you still insisted that this stuff has only been done in the last couple years. But the example I mentioned came out over seven years ago.

I forgot what we were discussing. Sorry. I think I was talking about the differences between lightmaps and light probes. I'm assuming light probes are now used to sample the environment using spherical harmonics (which is way more accurate than cube maps). Which, to my knowledge, has only been used heavily recently -- not 7yrs ago.

 

[HTupolev]

I'm assuming light probes are now used to sample the environment using spherical harmonics (which is way more accurate than cube maps).

How so? They're just two different representations for the same thing; it's like saying that a fourier series is "more accurate" than a time-domain representation of an audio signal. By being stored as a series by frequency, spherical harmonics have some really neat properties, but they're not intrinsically "more correct" in a general sense.

Which, to my knowledge, has only been used heavily recently -- not 7yrs ago.

Bungie's presentations on Halo 3's lighting do describe an SH distribution, which gets used for diffuse response. They're seemingly just not using it for specularity.

(Possible that SH integration with a Cook-Torrance specular BRDF wasn't quite in their budget for environment lighting, so they split off a precomputed "dominant light" direction also used for shadow casting. Or something.)

FarCry 3 does.. I took a video of it:

Not sure what that video is supposed to be showing, but Far Cry 3 uses precomputed radiance transfer. GI responds to infinite-distance lighting (and on PC it also responds to dynamic local lights via a more sophisticated transfer setup) with bounce and occlusion and whatnot according to the static geometry, although dynamic objects do not affect the irradiance probes.

http://www.gdcvault.com/play/1015326/Deferred-Radiance-Transfer-Volumes-Global

 

[ultragpu]

Here is a screen of a building like I was talking about.

That material looks flat and not very good texture work at all. Every piece shades the same even though they are different materials for the bricks and window seals.

I'm sure you can find bad looking spots in every game but that doesn't change how well the building materials and lighting can look in other places.

 

[Globalisateur]

Here is a screen of a building like I was talking about.

That material looks flat and not very good texture work at all. Every piece shades the same even though they are different materials for the bricks and window seals.

That's a shitty re-compressed facebook jpeg with a 71 quality.

I doubt it can accurately prove or disprove anything in any advanced technical discussion.

 

[VFX_Veteran]

How so? They're just two different representations for the same thing; it's like saying that a fourier series is "more accurate" than a time-domain representation of an audio signal. By being stored as a series by frequency, spherical harmonics have some really neat properties, but they're not intrinsically "more correct" in a general sense.

Well, I shouldn't have used the word more "accurate". It does have a smaller memory footprint right? I would think the main reason most devs switched to SH is so they can get more accurate lighting by having more of them with the same memory footprint (i.e. for environment lighting) as using cube maps. At least that was my assumption.

(Possible that SH integration with a Cook-Torrance specular BRDF wasn't quite in their budget for environment lighting, so they split off a precomputed "dominant light" direction also used for shadow casting. Or something.)

Why bake the Cooke-Torrance model into the SH? What happens if you update the SH in real time? Wouldn't that pretty much nullify the computational savings of not having to brute force calculate the BRDF per-pixel in the GPU?

Not sure what that video is supposed to be showing, but Far Cry 3 uses precomputed radiance transfer. GI responds to infinite-distance lighting (and on PC it also responds to dynamic local lights via a more sophisticated transfer setup) with bounce and occlusion and whatnot according to the static geometry, although dynamic objects do not affect the irradiance probes.

It shows that the precomputed irradiance probes get looked up per-pixel (or per-vertex, I'm guessing) on the material of the car and character's hands (which are dynamic) which then get added into the final look.

 

[VFX_Veteran]

That's a shitty re-compressed facebook jpeg with a 71 quality.

I doubt it can accurately prove or disprove anything in any advanced technical discussion.

It wouldn't matter if it was a 4k image supersampled from 16k framebuffer. You can clearly see the material and how it responds to the light.

 

[VFX_Veteran]

I'm sure you can find bad looking spots in every game but that doesn't change how well the building materials and lighting can look in other places.

That image shows me a specular response from the bricks that is not physically plausible. Why would bricks be so shiny when they are dry? Furthermore, it appears the window seals have the same specular response when they are clearly supposed to be made of a different material.

Here is an image of a real world location:

There is no such specular response on bricks (unless they are coated with something or wet). Which is my point.. you can't really tell that building is using PBR at all. I would think an Oren Nayar BRDF would be used on the bricks with very very low fresnel so that you don't get hardly any specular or the roughness would be so high that you'd not really see a massive amount of energy at the ideal reflection vector.

 

[djskribbles]

The Infamous shot ultragpu posted clearly shows that some of the bricks are wet, hence why they're shiny.

Notice how the dry bricks have no sheen?

 

[VFX_Veteran]

The Infamous shot ultragpu posted clearly shows that some of the bricks are wet, hence why they're shiny.

Notice how the dry bricks have no sheen?

So are you implying PBR = specular? When I look for PBR, I'm looking for a frame representing a scene that clearly shows different materials that look physically plausible. Taken from AC:Unity --

The floor has a different response than the rug, which is different than the "cooke-torrance-ish" golden statue, which is different from drapes, or the chairs, or the chandeliers diamonds.. I can count several different material settings (or types) in that frame. I can't do the same thing in ISS. Just saying..

 

[Nesh]

That image shows me a specular response from the bricks that is not physically plausible. Why would bricks be so shiny when they are dry? Furthermore, it appears the window seals have the same specular response when they are clearly supposed to be made of a different material.

Here is an image of a real world location:

There is no such specular response on bricks (unless they are coated with something or wet). Which is my point.. you can't really tell that building is using PBR at all. I would think an Oren Nayar BRDF would be used on the bricks with very very low fresnel so that you don't get hardly any specular or the roughness would be so high that you'd not really see a massive amount of energy at the ideal reflection vector.

There are bricks that are "shinier" depending on the material and if they are painted. But I understand where you are coming from.

 

[MJP]

Bungie's presentations on Halo 3's lighting do describe an SH distribution, which gets used for diffuse response. They're seemingly just not using it for specularity.

(Possible that SH integration with a Cook-Torrance specular BRDF wasn't quite in their budget for environment lighting, so they split off a precomputed "dominant light" direction also used for shadow casting. Or something.)

I think that they actually did use the SH data for specular in the shipping game. Their course notes from SIGGRAPH certainly go into detail on how to do it (they refer to it as "area specular"). They basically pre-computed Cook-Torrance specular projected onto SH for a set of viewing angles with the camera located on the XZ plane, and stored the resulting coefficients in lookup textures. The pre-computation also depends on fresnel intensity and roughness which gives you 3 dimensions for your parameterization, but they factor out the fresnel in such a way that you can reduce it to 2D. They also drop the SH coefficients that depend on Y, since they assume the specular reflection to be symmetrical with regards to the Y-axis. Then in the shader they just had to come up with a local coordinate frame where the camera is lined up with the XZ plane, and then use that to rotate the SH coefficients that they pull from the lookup textures. It's pretty cool stuff considering how long ago it came out.

 

[L. Scofield]

I was speaking about the environment HDRI lighting. I guess C1 doesn't have it.

Yeah, IBL was introduced in Crysis 2. Though then again, Crysis 1 was later ported to consoles and that version does support IBL.

FarCry 3 does.. I took a video of it:

Well, the video doesn't show anything that static probes can't do. However, I investigated and Ubisoft released a paper on the lighting:

http://www.gdcvault.com/play/1015326/Deferred-Radiance-Transfer-Volumes-Global

The probe data is semi-precomputed. Only static objects contribute to the indirect lighting.

Here is a screen of a building like I was talking about.

That material looks flat and not very good texture work at all. Every piece shades the same even though they are different materials for the bricks and window seals.

OK, that does indeed look lackluster, lol.

I'm sure you can find bad looking spots in every game but that doesn't change how well the building materials and lighting can look in other places.

VFX_Veteran is right on this. Those bricks look like they're made of plastic.

 

[VFX_Veteran]

MJP,

I noticed that you are a programmer for Ready-at-Dawn? First off, congrats on the game! It looks to be the best looking PS4 game(IMO) to date. A few questions:

1) You guys are using Cooke-Torrance for your lighting model. Why not GGX? It has a much better tail and we recently switched to it on our end.

2) What diffuse model do you guys use?

3) Do you think there is enough power in the PS4 to utilize full dynamic GI (including reflections) in a game like Order?

4) I'm really annoyed that only a few light sources are "tagged" as shadow casters. And then even among those, not all objects are shadow recievers. What's the cost on doing this for consoles, at least for light sources that move (i.e. lamps, flashlights, etc..)?

5) Finally, I hear about ray-traced solutions in real-time all the time (i.e. SSR, or HBAO to name a few). But are they all done in 2D? I am eagerly awaiting ray-tracing in 3D dimensions, or even path-tracing for true bi-directional importance sampled BRDFs and lighting.