Animated Spot starring Mantra

[dennis.albus]

Love the look and the detail you are putting into this. Keep it coming

[pezetko]

Principled Shader in H15 is not very physical and doesn't preserve energy if you are using non-metallic reflection or coat so I wouldn't be surprised if that's the reason for your problems

 

but your images look really beautiful, great work

 

I thought that Principal Shader was build according to disney model to do it finally right/correct way with correct energy preserving workflow for their PBR engine. I'm comparing it to the nonphysical hacked mantrasurface shader shader (I didn't have time to test new Principal Shader yet). So they did it wrong again?

[Kardonn]

I thought that Principal Shader was build according to disney model to do it finally right/correct way with correct energy preserving workflow for their PBR engine. I'm comparing it to the nonphysical hacked mantrasurface shader shader (I didn't have time to test new Principal Shader yet). So they did it wrong again?

 

No it's not "wrong", it's all built to spec, it's just that the Disney diffuse model already takes some kind of energy loss from reflection into account, or at least they approximate it and measure around what they've done.

 

Try this experiment and you'll see what I mean:

 

- Drop a Principled Shader onto a sphere with an Env Light in the scene, or teapot if you're that kind of feller

- Set Specular to 1.0, and give your ROP the all_comp AoV so you can see the diffuse and reflect coming out...render once and keep the image up

- Drop Specular to 0.0 and render again

- The "diffuse" component is identical in both renders because the Disney model doesn't actually do reflection masking, the reflection is simply additive onto the diffuse at all times.

 

This all kind of works out, because Specular "1.0" in the Disney model is only equivalent to a ~1.8 IoR fresnel response (8% F0 reflectivity) on a more old-school shading model...so they keep things very reigned in that way, and even though the reflection component is a straight up 'plus' to the diffuse, you never really break energy conservation all that much. But if your Base Color is {1, 1, 1} then it will fail the blast furnace test for whatever that matters.

 

The problem for me is that I really dislike the whole "metalness" workflow, and also dislike the non-literal Specular input...both for way too many reasons to get into in a short post. So I made a different recipe for the Disney shader instead with Metallic removed, a Specular slider where 1.0 means 100% reflective, and added a diffuse masking component so that you can describe any material as just Albedo, Reflectivity, Roughness.

Edited November 10, 2015 by Kardonn

[anim]

as Kardonn said it's like that by design

however what's worse than additive reflection is additive coat on top

the problem with that is as well that the coat slider is scaled by 0.25 so coat at 1 over black diffuse will be at F90 only 25% reflective, which should be 1, you can of course crank that slider to 4 to get more proper result

however if you have some reflective or metallic base and then cover with coat at 4 you will get massive energy increase since lower layers are not being compensated for that

I consider "Disney" shader more of an artistic hack for animated movies than physically correct shader for photoreal effects

and personally I have problem with architecture of Surface Model and other materials as well since many of them are averaging components when their energy reaches over 1 instead of properly layering them.

Edited November 10, 2015 by anim

[davpe]

I think having a shader that is not doing physically correct energy conservation is not necessary a disadvantage. If it does, it sounds cool but then there are cases when you'd like to get rid of it bcs of artistic reasons and you can't really do it. On the other hand, if your shader is not doing energy conservation, adding that functionality into your shop network is quite easy and it gives you more control. Only thing is that you have to know what you're doing so I agree that for people that aren't mainly light/surfacing artists it may be rather confusing not having it "physically correct" out of the box.

 

Btw. I think that all the physical correctness trend is a bit overrated. Sure, we are trying to do things that look physically correct but 1) you never (or rarely) have physically correct lighting/surfacing data so you're just eyeballing anyway and 2) even if you did 100% physically correct image, you end up changing it bcs client wants to drive the result artistically. So at the end of the day, it is good to make things physically correct as much as you can but it is not good to be constrained by it just bcs "this is how physics works".

Edited November 10, 2015 by davpe

[davpe]

posted renders look great, btw. ;-)

[Guest tar]

In defence of the Principle Shader, almost most of the new gallery presets use it, so one advantage is that we have a library of elements to start with now

[anim]

...

Btw. I think that all the physical correctness trend is a bit overrated. ...

I agree, but "physical correctness" term aside, I still prefer layered component approach to additive or averaging as the later ones make controlling influence of the components using textures a nightmare

 

In defence of the Principle Shader, almost most of the new gallery presets use it, so one advantage is that we have a library of elements to start with now

I wouldn't call it an advantage, but that's my personal opinion based on disliking the idea of principled shader in general

[Kardonn]

My thoughts on "physical correctness" is that it's ultimately a very good thing, but also agree with Tomas about proper layering and masking.

 

The philosophy I go with is that I want my materials, lighting, and renders to be as close as I can possibly get to being physically correct. My aim is always to recreate the same image you'd get if you shot the scene with a DSLR, and properly linearized the image off the sensor...that should always be the goal reference while still working within your 3D package of choice.

 

After that, just the same way you start with a Camera RAW file and begin grading it, you do the same with the renders. They get the same color matrices, distorts, lens fringing, mild aberration, grain, and vignette that a film plate would have, and once that's all lined up, the color grading and LUT pipeline is no different between renders and film.

 

You can grade with power windows, you can paint in local adjustments, do spot exposure, whatever you want...but if the initial render going into this process isn't a very faithful reproduction of the Scene Referred Data, then you're just never going to reach your full potential.

[Kardonn]

A couple more! Gramps and the dog.

 

Needs some raytrace bias adjustments for the newspaper to render right.

 

[Netvudu]

Awesome! And I like a lot the fiber job.

Is it a spanish client? because the butter package is indeed in spanish...I know it because I´m spanish, you know 

 

 

BTW, and without trying to derail the thread, I see the Principled shader doesn´t work for transparent/dielectric objects, does it? 

[magneto]

as Kardonn said it's like that by design

however what's worse than additive reflection is additive coat on top

the problem with that is as well that the coat slider is scaled by 0.25 so coat at 1 over black diffuse will be at F90 only 25% reflective, which should be 1, you can of course crank that slider to 4 to get more proper result

however if you have some reflective or metallic base and then cover with coat at 4 you will get massive energy increase since lower layers are not being compensated for that

I consider "Disney" shader more of an artistic hack for animated movies than physically correct shader for photoreal effects

and personally I have problem with architecture of Surface Model and other materials as well since many of them are averaging components when their energy reaches over 1 instead of properly layering them.

 

Can you please elaborate on your idea of perfect shaders for photoreal work? It would be interesting to hear your ideas for people who value it

 

I thought both Disney and the old mantra surface shader was meant for photorealism.

[Kardonn]

Work in progress for the large environment establishing shot. I sure love Houdini for doing big layouts. Matte painting take over of course beyond the middleground hills, but I'm thinking of just extending the full CG layout to pepper in trees all the way to the horizon to help anchor the MP and the CG together more.

 

[Kardonn]

Finished neighborhood establisher! More of an afternoon feel to it now, was a little too graded towards sunset before to work with the rest of the spot.

 

 

And the finished shot of gramps and max catching a nice aroma. In a perfect world I'd have done a VDB gradient solve on that stupid dog tag in the fur...

 

 

 

 

Finalized pack shot.

 

 

Shot of all the kids on the couch

 

Edited December 23, 2015 by Kardonn

[vfxhound]

This is an awesome thread! Fantastic work Kardonn!

[Mandrake0]

nice job.... congratulation to the team..... :-)

[eistan]

holy shit! its really amazing! looks like Pixar! especially girl and father on the first page! and the grass is godlike!

[timmwagener]

Looks awesome! (....although I cannot look at butter packshots anymore like I could in the past  )

[Kardonn]

Butter really does make everything better.

 

Had a chance to dance with DOPs on a cream pour. Really happy with how it came out, though it was pretty difficult getting a very stable mesh on that thin and slow moving cream residue. Ended up doing some serious hackery with SOP Solvers after the FLIP sim where I'd delete all points that were further than 0.005 units from the collision object, and merge every frame together that way in order to have a more stable contact point.

 

Fixed a ton of issues with the way the cream pours out of the bottle via SOP Solvers in the DOP network as well. Seems like such a simple sim and shot, and yet with the way the bottle moves I was just getting way too much sloshing and ugliness in the way the pour looked. Ended up making a ring around the inside of the bottle with the bottom section of the ring cut out. Extruded it in, converted to a VDB, and then in the SOP Solver I was just deleting any FLIP particles that ended up inside that VDB.

 

Worked quite well though all in all, and luckily I only had to RotoPaint fix two tiny spots in Nuke after all was said and done.

 

DG060_comp_v001_.mov

 

Was amazed at the speed the FLIP sim ran, especially with such a dense collision SDF (I think in the neighborhood of 15M voxels). 15 minutes for this sim, 3 minutes for the SOP Solver, 12 minutes to mesh it. This entire sim and render was literally done the night before the due date, and needed a lot of iterations to get the performance right...so thank you SESI.

 

Edited January 19, 2016 by Kardonn

[Kardonn]

Here's a few images for print campaigns as well.