Digital Foundry Article Technical Discussion [2017]

[iroboto]

A big selling point of PBR is that you don't have to do that. By using physically based values instead of eyeballing things materials look right in all kinds of lighting conditions. Scapes demonstrate this well.

I don't think all PBR systems are equal. I mean, many games since being released in 2013 going forward have been PBR, and not many have been able to nail the lighting for PBR correctly.
ie, look at all Bioware games with PBR, it's pretty off. Look at the metal armor in this case, quite shiny and completely out of place.

Assassin's Creed is all PBR as well, but only Unity looks exceptionally amazing, all other Asses don't look nearly as good.

The metal seems way off once again here in Syndicate.

Whereas I see way less of these issues in Unity for instance. Its PBR seems to work really well.
Anywhere in this clip, you don't see those PBR type issues with metal: (look at 9:54, the watch, looks great!)

And we don't see these PBR issues with Order 1886, Ryse Son of Rome, or GTS.
But we see these weird PBR interactivity issues with Halo 5, Forza 5+, even Gears etc. I think some would agree that certain engines, like the Fox engine for instance, has less of these issues.

So overall, I don't think PBR values on pipeline itself is enough to make things look right, it must take more than just PBR based rendering to make things look physically based. That pipeline and how the lighting interacts with those PBR values is probably what really separates games from each other, even though they are using the same features.

 

[dobwal]

I remember being impressed by the leather and some of the other cloth material in Unity. And I'm still impressed by the video, the clothing can look exceptional at times and I can't recall any games that have surpassed it.

 

[iroboto]

I remember being impressed by the leather and some of the other cloth material in Unity. And I'm still impressed by the video, the clothing can look exceptional at times and I can't recall any games that has surpassed it.

It would be awesome if they could revisit this game with 4Pro and 1X and ideally try to clean up some performance issues remaining, while providing higher fidelity.

 

[dobwal]

It would be awesome if they could revisit this game with 4Pro and 1X and ideally try to clean up some performance issues remaining, while providing higher fidelity.

Yeah like use PS4's RPM along with tressFX to fix one of the glaring issue with Unity, the awful looking hair. Everybody looked like that they wore wigs made from yarn.

 

[OCASM]

I don't think all PBR systems are equal. I mean, many games since being released in 2013 going forward have been PBR, and not many have been able to nail the lighting for PBR correctly.
ie, look at all Bioware games with PBR, it's pretty off. Look at the metal armor in this case, quite shiny and completely out of place.

Assassin's Creed is all PBR as well, but only Unity looks exceptionally amazing, all other Asses don't look nearly as good.

The metal seems way off once again here in Syndicate.

Whereas I see way less of these issues in Unity for instance. Its PBR seems to work really well.
Anywhere in this clip, you don't see those PBR type issues with metal: (look at 9:54, the watch, looks great!)

And we don't see these PBR issues with Order 1886, Ryse Son of Rome, or GTS.
But we see these weird PBR interactivity issues with Halo 5, Forza 5+, even Gears etc. I think some would agree that certain engines, like the Fox engine for instance, has less of these issues.

So overall, I don't think PBR values on pipeline itself is enough to make things look right, it must take more than just PBR based rendering to make things look physically based. That pipeline and how the lighting interacts with those PBR values is probably what really separates games from each other, even though they are using the same features.

The problem is usually the lighting, not the materials. You can grab any of those assets and put them in a path tracer or even just lit them with an HDRI cubemap (like Scapes) and they'll look great (BTW you can do both in Unity now that it has support for Octane). Now of course, PBR means physically-based rendering, not physically-correct rendering so many things will differ from reality simply because the artists want it so.

 

[iroboto]

The problem is usually the lighting, not the materials. You can grab any of those assets and put them in a path tracer or even just lit them with an HDRI cubemap (like Scapes) and they'll look great (BTW you can do both in Unity now that it has support for Octane). Now of course, PBR means physically-based rendering, not physically-correct rendering so many things will differ from reality simply because the artists want it so.

The material values are handled by 3rd parties? Or is this something that any (indie etc) can grab values and place into a PBR pipeline?

 

[Sigfried1977]

Yeah like use PS4's RPM along with tressFX to fix one of the glaring issue with Unity, the awful looking hair. Everybody look like that they wore wigs made from yarn.

Weird. I thought Unity had some of the best hair I've ever seen in a game. I was particularly impressed with the hair of assassin dude's lady friend. Certainly beat the weird physics nonsense on Lara's head. Even though it was less dynamic.

 

[dobwal]

Weird. I thought Unity had some of the best hair I've ever seen in a game. I was particularly impressed with the hair of assassin dude's lady friend. Certainly beat the weird physics nonsense on Lara's head. Even though it was less dynamic.

Some characters didn't bother me but some characters were really distracting especially close up with really thick strands.

https://goo.gl/images/mzmCZJ

Other than that though I was really impressed with a lot of what Unity did technically, art and effort wise. I'm planning to load it up again and see how it stands up against more modern titles.

 

[zed]

It's average visual acuity for humans.

I'm not sure if you're serious 20/20 is NOT average vision acuity
eg http://www.richmondproducts.com/files/8213/1550/0515/ETDRS_Instructions_082108.pdf

Dr. Colenbrander also emphasizes that, contrary to popular belief, 20/20 is not
actually normal or average, let alone perfect, acuity. Snellen, he says, established it
is a reference standard. Normal acuity in healthy adults is one or two lines better.
Average acuity in a population sample does not drop to the 20/20 level until age 60
or 70. This explains the existence of the two lines smaller than 20/20: 20/15 and
20/10.

It is amazing how prevalent this myth persists about 20/20, have a google you'll see even on some optometrists websites often they claim 20/20 as average or even perfect vision WTF they don't even know a simple basic fact about their own profession

 

[OCASM]

The material values are handled by 3rd parties? Or is this something that any (indie etc) can grab values and place into a PBR pipeline?

There are charts like the one bellow floating around the web:

I suppose most indies take charts like that as a start and extrapolate and experiment from there. If you undertstand the basics and test with a variety of lighting environments it's easy to stay within the bounds of reality.

 

[Graham]

^^^ that chart is actually hugely misleading.

One of the great problems of PBR references is using sRGB or perceptual colours can very easily mislead.

Take two of the darkest materials in that chart: coal and 'dark soil', which are listed as intensities of 50 and 67 respectively.

This sounds like a pretty minor difference, so you don't need to be that precise... But these are perceptual values in sRGB. That seemingly tiny difference is actually around a 2x difference in terms of linear light. That's a *huge* difference in the amount of reflected light.

This is why I said earlier that material balancing is damned tough. It's so easy for people to paint in arbitrary details (especially shadows, creases, etc) and completely break the material, while thinking they've improved it. You need to treat it as a mathematical problem and author in a way that removes the human perceptual element (which is a hard thing to convince artists to do).

You also get nasty knock on problem: where one thing is too dark/bright, and other things get balanced to it...

 

[OCASM]

^^^ that chart is actually hugely misleading.

One of the great problems of PBR references is using sRGB or perceptual colours can very easily mislead.

Take two of the darkest materials in that chart: coal and 'dark soil', which are listed as intensities of 50 and 67 respectively.

This sounds like a pretty minor difference, so you don't need to be that precise... But these are perceptual values in sRGB. That seemingly tiny difference is actually around a 2x difference in terms of linear light. That's a *huge* difference in the amount of reflected light.

This is why I said earlier that material balancing is damned tough. It's so easy for people to paint in arbitrary details (especially shadows, creases, etc) and completely break the material, while thinking they've improved it. You need to treat it as a mathematical problem and author in a way that removes the human perceptual element (which is a hard thing to convince artists to do).

You also get nasty knock on problem: where one thing is too dark/bright, and other things get balanced to it...

That chart is actually very old. I think it's what the developers of Remember Me used internally. Nowadays every "Intro to PBR" guide tells you to author everything in linear space and to remove all lighting information from albedo textures. Big studios also don't rely on 3rd party charts. They build their own databases based on their own measurements.

Do this things right and that last problem you mention is gone.

 

[Recop]

No mention of HZD ? Their PBR is really good, at least for the clothes.

 

[iroboto]

No mention of HZD ? Their PBR is really good, at least for the clothes.

I was only bringing up where I saw PBR lighting issues, relative to the idea that values were enough to ensure a realistic look. It wasn't meant to be a comparison/best of PBR thread.

 

[Clukos]

Assassin's Creed is all PBR as well, but only Unity looks exceptionally amazing, all other Asses don't look nearly as good.

Syndicate ain't all that bad:

Although I agree that Unity is still better.

 

[Dictator]

BTW, I asked on twitter before, but dragon age from this gen is not utilising the PBS introduced into frostbite. The first title to do that was Battlefront.

It did probably use a normalised pipeline though, hence why the metals are more convincing than last gen frostbite titles.

 

[London Geezer]

Ah the curse of AF hits the X too. Titanfall 2 has lower AF than the Pro. It’s a curse.

 

[iroboto]

Ah the curse of AF hits the X too. Titanfall 2 has lower AF than the Pro. It’s a curse.

@Globalisateur :: I hoping from this article that you will see that when something is wrong, DF will call out shit.

 

[Nesh]

Is DF planning to make an update on their F7 analysis since it got an update with enhancements?

 

[iroboto]

Is DF planning to make an update on their F7 analysis since it got an update with enhancements?

lol I hope so.
Man these guys are going to be so busy haha. No vacation for these guys for a long time.