Time to ditch sRGB/Linear as default (?)

[Njen]

Compositing is much more complicated than simply adding together a bunch of layers. What if you needed to reduce the specular component on a specific object, or desaturate a specific object, or boost the SSS of a specific character? Or, what would you do if you can't render every object in a single pass due to memory limits, or combining passes from different renderers? All of which happens on practically every single shot, on every single show I've worked on. You would be amazed by how much detail is scuplted into the final comp. All of this is possible by rendering out necessary layers.

Saying that grading individual passes and combining the result looks worse just proves my point that photorealism is subjective. As while you think it might look worse, the director thinks it looks better.

Also, many places render out deep images, hence bypassing the need to render masks, of which the individual components become even more important.

If photorealism wasn't subjective everyone who takes the following test would get a 100% pass to the following:
http://area.autodesk.com/fakeorfoto

Trust me, photorealism is subjective, highly subjective.

[Ludvik Koutny]

For the 3 years I worked in PFX VFX studio, it was only on very rare occasions that we ever needed to do things like boost specular or boost SSS in post. Actually, if you conform to PBR rendering, then doing such things is simply breaking the shaders. I almost never had any supervisor, director or client requesting something to be boosted specifically in post, as it almost always looked worse. We got away with it on some very cartoony things, but in photorealistic scenes, things looked always far better when tweaked right in 3D. Even simple boosting of reflective layer caused it to mismatch between directly visible surface and reflected surface. Also bounced light then usually did not add up.

I would also recommend this article from Blur, who are known for their high quality work when it comes to fully CG projects: http://www.creativebloq.com/blur-studio-elder-scrolls-online-cinematic-2123047

A few quotes from the article:
"To streamline production we strive to render as many elements together as possible because it's the fastest way to achieve realistic results with V-Ray. This minimizes the risk of inferior results trying to reconstruct various passes during compositing."

"Five or six years ago we could produce elaborate breakdowns of simple shots into hundreds of passes all cleverly composited back together. It's still somewhat true when we deal with cartoony or stylized projects, but I find that the key to a realistic project like Elder Scrolls is to spend most of your time inside the 3D package."

"Ornatrix and V-Ray can now render hair strands as splines at render time, which allowed us to render our characters and their hair in the same pass and using the same GI lighting solution."

Regarding Fake or Foto challenge, I wanted to bring it up myself, as like you, I too consider it to proof of my point. Point that photorealism is hardly subjective and that vast majority of people find the same few images picked by staff photoreal. The fact, that it is hard to get 100% result in this challenge, even for a trained eye, shows that it's generally quite common for CG images which fool majority of the people to be produced. The challenge is not about people voting how much photoreal do they consider a particular image to be, but rather if they can distinguish CG imaged scattered across actual photographs.

[peterguthrie]

DO IT!

no question whatsoever in my mind.

[romullus]

Rawalanche, do you have some specific values, you'd like to become as defaults? If so, can you share it with us? I'd like to try and see if it will magically makes my renders more photoreal :] Now i have to tweak tonemapping for each scene individually. If your proposal will take place, then it will mean that i will have different starting point, but i will still have to tweak those controls. Would that change anything?

[Ludvik Koutny]

Rawalanche, do you have some specific values, you'd like to become as defaults? If so, can you share it with us? I'd like to try and see if it will magically makes my renders more photoreal :] Now i have to tweak tonemapping for each scene individually. If your proposal will take place, then it will mean that i will have different starting point, but i will still have to tweak those controls. Would that change anything?

Not yet, I haven't even tried it yet. I am asking Dubcat to do some of his LUT magic to try ACES in Corona.

That being said, you will always need to tweak something on per scene basis, in the same way most photographers still take their photos in Photoshop to fine-tune them. You would not necessarily get super pretty image right out of the box, but you would get something that's a lot closer to what would happen if you converted your 3D scene into real world and actually took a picture of it with a digital camera.

It also makes eyeballing material properties from photos and translating them into CoronaMTL settings a lot easier :)

[Ondra]

I was thinking about this lately a lot too. We will almost certainly do it for 1.6. Now the question is, what should the defaults be. And would you consider taking this further with perhaps auto-exposure/autocontrast? That is the real difference that cameras make

[Njen]

Rawalanche, I'm telling you, on all of the projects I've worked on, whether it's X-Men, Wolverine, Hunger Games, Game of Thrones, etc., and whether the objects conform to PBR or not, the comments about what looks real, and what doesn't look real, all have been subjective to whoever is commenting on whatever submission is being commented on.

I use and have used many renderers professionally in my career: PRman, 3Delight, Mantra, Arnold, VRay, Mental Ray, Maxwell, and more, and I really do hope that Corona doesn't break with what the enitre CG/VFX industry puts out by default. It will just cause unnecessary confusion.

[Ludvik Koutny]

Rawalanche, I'm telling you, on all of the projects I've worked on, whether it's X-Men, Wolverine, Hunger Games, Game of Thrones, etc., and whether the objects conform to PBR or not, the comments about what looks real, and what doesn't look real, all have been subjective to whoever is commenting on whatever submission is being commented on.

I use and have used many renderers professionally in my career: PRman, 3Delight, Mantra, Arnold, VRay, Mental Ray, Maxwell, and more, and I really do hope that Corona doesn't break with what the enitre CG/VFX industry puts out by default. It will just cause unnecessary confusion.

I'd suggest giving this a watch:

I still don't think we understand each other. You won't lose ability to render linear. It will still be there. It's just that it won't be default, it will be one click away.

[lolec]

Saw that video today, I was disappointed to remake the test and see the "bad " behavior on light saturation. I'm a total idiot regarding color but felt like the video was into something, I was unsure on what to even say or propose, I'm so glad someone else took the time to post it and Ondra actually recognizes it as something worth doing :)

Thanks!

[Njen]

You won't lose ability to render linear. It will still be there. It's just that it won't be default, it will be one click away.

I'm fully aware of ACES CG, I've helped rollout the implementation of it in two studios. FYI, ACES CG relies on linear images being input for best results.

I'm saying linear should stay the default output, and have whatever custom colour operations be the 'one click away' option.

[Dionysios.TS]

Very glad to read this post this morning! Just do it guys and go forward with it...

Thanks for your efforts as always.

Dionysios -

[lacilaci]

I was thinking about this lately a lot too. We will almost certainly do it for 1.6. Now the question is, what should the defaults be. And would you consider taking this further with perhaps auto-exposure/autocontrast? That is the real difference that cameras make

But what would be the standard for auto contrast/exposure? Auto exposure in dslr's can be pretty shitty at times...

I would agree on some default filmic tonemapping with an option/checkbox to export as linear(in render settings also in vfb maybe)

Also (autofocus/pick focus in vfb when in interactive rendering??)

[pokoy]

Why fix something when it's not broken? What's the point of having a few parameters set to arbitrary values when it's a matter of seconds to set them by myself?
I'd rather control this by myself, and honestly I don't want to lose any of the parameters of the VFB post

Do we really need another change in default behavior for 1.6? And then yet another 'legacy'-checkbox in order to achieve the same result like in prior versions when re-rendering a scene? Do you realize  that with every point release we get a new default somewhere and it takes effort to make sure the result is exactly the same when re-rendering an old scene and introduces uncertainties?

[burnin]

true... also, more of you should just give Blender a spin :P

[pokoy]

I should add that having an auto-exposure / auto-levels button wouldn't hurt. But I really would hate if any of the defaults that are in place right now would change and would be yet another thing to keep in mind when trying to match an older render.