The Rabbit Hole

troy_s

03:23

@georges Just read your answer. I should be clear, when using False Colour, the goal is not to necessarily always bring all values into the display referred transform, but rather bring the reasonable values you expect into the view.

@Georges As with things like hot kicks and reflections, parts of those might be entirely legitimate to blow out past peak, if for example they are fresnel or kicking hot sources such as the sun or even a bright lamp.

@Georges Also, when evaluating how strong to make an HDRI there is no reliable value any more than there is some unanimous decision to expose, shutter, and ISO a photo; it depends on the damn HDRI content, how it was made, the ranges it covers, and the alignment with your scene.

By suggesting a value you are misleading people. The strength of an HDRI is absolutely arbitrary in relation to your scene and how the HDRI is encoded.

If you don't quite grasp that yet, re-read it until it is clear.

If you think in terms of grey cards, your object would be exposed as per your creative intent. Somewhere in that exposure, might be a middle grey value that ends up around 50% display referred (possibly not too!)

But that value is arbitrary with regards to where it lands. That is, it is up to your creative choice.

Wherever that value is, now you have to figure out where in the HDRI's intensity scale would your scene be anchored

And that is your HDRI alignment.

It is exactly like thinking about taking not one, but two photos and setting exposure for each. It is that arbitrary.

"How would I like the exposure on this horse? OK... Where about in the HDRI's dynamic range would I expect that exposure to land?"

This is why using HDRIs drives me absolutely batty because frankly, they are another complex facet that everyone thinks is some magic use and done sort of a thing, when nothing could be further from the truth.

If you want to really appreciate the nuance of this, take two photos with a camera you can control. Under identical lighting, take one photo at base exposure. Now take another that has a different shutter speed, a different ISO, and a different aperture that yields a different and random F-stop differential.

Now take both of those photos into Blender and align them into the same scene.

If you can appreciate the nuanced complexity of that scenario, you are ready to properly use and control HDRIs

If you can't, then you probably shouldn't be using HDRIs.

And again, guessing with what you think you are seeing is entirely not an appropriate solution.

The solution is absolutely mathematical, and for every combination of scene and HDRI, there is precisely one correct solution.

tynaud

08:50

@troy_s I am having a look at the documentation of OCIO and in their examples, it shows that the 3Dlut film desat transform happens at the end of the group transforms, and it seems to me that it is read from first/top->last/down...

I don't know if I explain it in an understandable way, my english is often wacky.

So I was asking myself that perhaps in the ocio.config it should be :

transform: !<GroupTransform>
      children:

  - !<FileTransform> {src: LMT_Power_22_sRGBScaled.spi1d, interpolation: linear}

  - !<FileTransform> {src: bassam-65-filmdesat.spi3d, interpolation: best}

in that order...

I mean, if the power 2.2 1dLut is an emulation of a cdl with power 2.2, which I think so, it should be applied before the film desat 3dLut ?

tynaud

10:03

Whoohoo ! Victory !

So, some little things that I learned, it seems transforms are applied from bottom to top. It's the opposite of the examples which are put on the OCIO documentation. If you some illegal entry block at line xxx, you must have some tabulations at the beginning the lines indicated

My editor si putting them by default, but it's often possible to change this behavior so that a tab is 4 spaces.

If you want to apply the ASC-CDL node which you have put in an xml as specified in the documentation, to make it work under the -10-+6.5 view

you have to revert the -10-+6.5 transform before applying the ASC-CDL node, then reshape to -10+6.5 and lastly apply the 3dLut film desat.

So, in my LUT folder I have this

Ha... we can upload only images here...

Here is the *.cc

<ColorCorrection id="tynaud">
        <SOPNode>
             <Slope>0.4 0.4 0.4</Slope>
             <Offset>0 0 0</Offset>
             <Power>2.2 2.2 2.2</Power>
        </SOPNode>
        <SATNode>
             <Saturation>1</Saturation>
        </SATNode>
  </ColorCorrection>

And here is the look to add to theconfig.ocio

  - !<Look>
    name: custom ASC-CDL transform with desat
    process_space: aces_lmt
    transform: !<GroupTransform>
      children:
      - !<FileTransform> {src: bassam-65-filmdesat.spi3d, interpolation: best}
      - !<FileTransform> {src: LMT_Shaper_to_linear.spi1d, interpolation: linear, direction: forward}
      - !<FileTransform> {src: mygrade.cc}
      - !<FileTransform> {src: LMT_Shaper_to_linear.spi1d, interpolation: linear, direction: inverse}

Georges

10:50

@troy_s I will re-check my answer and edit where it needs

But I really don't remember me suggesting that there's a magical number for all HDRIs in the world!!

So may be I just mis explained it

I was discussing that particular HDRI

@troy_s I answered my question since no one did, and I thought I gound the solution

Experimenting with many for days, I just found as a generic thing, that kicking the strength high during render, then bringing it down using slope, provided more grades in the image

That's much of a guideline, not a sacred rule

And I linked @cegaton answer for people to get the accurate info

@troy_s anyways I will re-check it and enhance it, if you have the time, please do edit it, or post an answer :)

@troy_s but tbh there are many points that you mentioned I really don't get

Yes the finall call is an artistic call

How to light the scene, but what I'm doing so for is merely testing

I'm not really trying to create an artwork, I'm trying to enhance my comprehension and skills in color management

I need to know the rules first in order to break them later

That's it

Just like with compositing an image, there are countless rules and ratios, rule of thirds, golden ratio... but they are more of guidelines

So when you say that there's only one right mathematical solution to how you light a scene!

With a particular HDRI

I mean, I understand having guidelines in this case, don't burn or heavily under expose, but as guidelines

Reading the fake color or vectorscope or waveform help seeing those limits

But as an artist, how would anyone be able to get that single one mathematical solution using a particular HDRI for a particular scene?!

Georges

11:29

@troy_s I made several tests yesterday, using several HDRIs in order to edit that question to explain how variable these values might get, again, I'm just trying to answer my question, since no one did, for particular cases

troy_s

13:07

@tynaud Order is based on to_reference or from_reference.

Matt

Oooh! Looks like neat things have happened since I was here last!

troy_s

@tynaud And no, in the case of the power, you want it applied after the shaper.

@Georges You aren't listening. When it comes to an HDRI, you have to understand it is a ratio series, arbitrary. So it isn't about lighting a scene, it is about where ratios end up relative to your scene, as I explained.

An imager has to understand the example I gave, such as aligning two photos.

It isn't a creative act, but a mathematical one.

So no, it isn't a creative act, and you aren't lighting the damn scene, but rather using canned lighting.

The only creative aspect is figuring out where your exposure is, and aligning the HDRI to it.

@tynaud You overcomplicated the configuration; no need to invert the shaper. That is what process space is for. OCIO takes the data to process space before applying the transforms. So if you want to match your CDL values, they are applied against scene referred values.

@tynaud That means you can set your process space to linear, then CDL, then shaper, and see what you had.

troy_s

13:26

@tynaud Bear in mind you can apply CDL before the shaper or after, but the results will be quite different. To match what you see in Blender with nodes, it must come before.

tynaud

@troy_s thanks for the light ;) I will try to test a few things tonight to see if I have understood it correctly. After that I will be away from keyboard till the end of the week, sadly...

Georges

13:55

@troy_s this is a part of my answer "Of course these values would work for this particular HDR image, not necessarily a rule"

Anyways I'm editing to emphasize more that each HDRI is a unique case

@troy_s edited blender.stackexchange.com/questions/56267/…

Georges

14:24

Looks like we wont need emission shader anymore lol

Only diffuse shaders here, with Look: Team Argentina Basic Desaturation

Same with no Look

False Color Basic Desaturation

troy_s

14:49

@Georges Those aren't emissions?

@Georges Diffuse with >1.0 albedo?

Georges

I know @troy_s this is why I ad the lol :)

the bottom ones are 16.292, 0.0001758,0.0001758

for blue, green and red respectively

troy_s

@Georges It is a good point to illustrate albedo versus "colour"; the albedo is reflecting more light than it is absorbing. In your case, 16x

Georges

yes

troy_s

Bear in mind that albedo is linear, and closer to stops. So 16.291 has no real 1:1 relationship with reflectance.

Georges

you started talking that foreign language that I really have to learn :)

So 16.291 has no real 1:1 relationship with reflectance

troy_s

14:53

Nope.

Think of it this way, if you made a 100% reflectance and the thing you were reflecting 200 units, and you put a perfect reflectance angle so that you are looking at it, what value would you see?

You need to force yourself to stop guessing and twiddling knobs, take a step back, and think about the "reality" of your scene. It will bring a clarity.

Georges

200 units

troy_s

So a value of 16 merely means you are reflecting 16x the incoming light, which depends on a variety of factors in terms of how much is output.

There is no relation / reason to use 16.291 as though it were a magic value for albedo, as it is the wrong model and wrong concept.

Georges

Yes but there are no lights here, these are only diffuse shaders, no lights

troy_s

There is some light.

Otherwise what is it reflecting?

You have a world value set.

Set that to black and see how much reflects. ;)

Georges

I set that to 0.0001758 so yes there is some

background set to 0.0 :)

lol

but there's ambient occlusion :)

troy_s

15:01

You will have to take it at word: if there is no light, there would be no reflectance.

But the key point that 16.291 is not a magic number here. 16.291 has only relevance on the scene referred values.

Georges

world 0.0001758, no AO

Yes I got that @troy_s

No magic numbers

TBH @troy_s this one day will become standard, until then, I understand why many wont mingle with it, people stick to what they now better, people using the old way for a long time, adaptation isn't an easy step

So 16.291 represents the range from no light to what used to be white in sRGB?

sorry range from 0.0001758 to 16.291

since in real life there's no such thing as 0-1

troy_s

15:20

Correct!

Many imagers are insecure with what they learn because the knowledge is built on poor foundations.

When an imager builds up on solid foundation, they are much more willing to use techniques because they are in control.

Feeling out of control is not going to help someone. So hammering the foundational knowledge is critical. Really processing what is happening is key.

Once an imager grabs on, thus far it seems, they run with it.

Examples: You with the HDRI experiments, @tynaud with the CDL per shot looks, etc. Imagers aren't idiots, but frequently crippled with bad foundational concept models.

(Further, the bits of the puzzle start to connect across all things, as opposed to feeling like a million small little tricks, cheats, and do-this-then-that.)

Georges

Yes I agree, BUT, these should happen gradually, no one becomes an expert overnight, we learn from our mistakes, and we need to find a way to implement this in real application, I agree that the most important thing is the theory itself, but the application is not irrelevant

And there's the technical part, where we are used already to use hacks to get the effect we need, we are used to it, it's saved in our start up files :)

unchecking all these hacks takes time

Like AO, clamps, ...

I hate clamps buy the way, they're gone for good :)

and the fact that we're using cycles, which is a hack in a way, right?

it is biased, it doesn't calculate things as cameras or our eyes would, we don't have a checkbox for reflective and refractive caustics in our eyes

When I'm rendering a test, I have to keep all these in mind, besides the new knowledge, learn how to get rid of them one by one, in order to see how that affects my results, testing isn't bad at all, this is how I learned almost everything I know

On the other hand, I always felt something was wrong with the old system, this is why this new LUT grabbed my whole attention

And it is the right way to do things, because it is built on physically correct foundations

Georges

16:05

I mean how would you get this using the old 1D LUT using diffuse shaders, a sun lamp and a sky texture?

Georges

17:01

Anyone knows where can I find documentation about the updated light path node?

oblomov

19:30

@ Cegaton Did you make a text typo?
"Middle Grey = Gray Scene Referred Linear value 0.018009142."
Shouldn't that be 0.18009142. ? :p

cegaton

20:14

@oblomov yes, one zero too many

oblomov

@cegaton :)

cegaton

@oblomov @troy_s can you revise the Color Code Values? github.com/sobotka/bassam-test @oblomov spotted an issue here.

troy_s

20:48

@cegaton You have commit access.

troy_s

21:02

@cegaton I thought I had the numbers close

Transcript for