The Rabbit Hole

Mareck

11:11

Hi, @troy_s @cegaton @aliasguru yesterday I was thinking I've discovered a massive Cycles speed improvement, but after lot of testing I came with 8 to 40% speed improvement.
2 peoples tested it and see no difference at all what make me think I have something wrong in my Blender.
It's if you do a render with filmic or display to none, I've got really a huge difference, can you or someone test it? The biggest gap is between log+contrast and none.

And by the way, where is the end of that hole, I can't see the bottom^^.

troy_s

14:32

Greets @tynaud

tynaud

Hi @troy_s :)

Question for you : how would you push a golden hour look using CDL node ? Did I already ask this (can't remember) ?

troy_s

@tynaud Power the compliment colour a bit.

@Mareck Worth testing on Pan's BMW scene.

@tynaud Ideally I suppose you would probably need to break your scene into regions and grade elements as required. As a one light, slope and power the compliment colour of what you are going for.

tynaud

14:49

@troy_s I have a sun lamp in the scene slightly orange, and the sky is bluish (no hdri). For now this is what I have (WIP for everything).

Mareck

@troy_s @troy_s need other tester because on my side I always have boost by disabling color management when rendering

troy_s

@tynaud For one thing, I would definitely try an HDRI just as an experiment and to generate ideas.

@Mareck That might be likely. Color management adds calculations. Can't be helped. But you also lose any ability to apply proper corrections on input footage.

@Mareck Try rendering via the CLI

@Mareck I know that when I last did heavy rendering (as in 40 hours) the CLI was significantly faster.

Mareck

No I just put the display to none during the render and go back to filmic after the render ended @troy_s

tynaud

@Mareck so this way you avoid processing each time the tiles are updated... That makes sense, but up to 40%, that's a huge difference. And if you hide the renderview while rendering ? I guess it would be the same as having colormanagement disabled...

Mareck

15:05

@tynaud it looks something like that but 2 people tested and they haven't seen any difference.
I'll try with hiden renderview.

tynaud

@Mareck Will do the test too.

Mareck

@tynaud thanks

Because if I'm the only one it means that I don't have a speed up but I have speed down at the base and recover normal speed when disabling display

troy_s

@Mareck Yes. Again, test with a timer via CLI

@Mareck Basically it would amount to a bad architecture decision in how frequently it is updating display. OCIO shouldn't be influencing by 40%, but there is an overhead.

CLI would entirely skirt around all display issues, and frankly, it is how rendering should be done once the work is finished.

aliasguru

16:41

@Mareck testing right now

@Mareck what settings / device are you using? does it make a difference for you?

Mareck

@aliasguru usually I render with blender open not CLI, and I've got different render time according to what device I use in color management.

If I put it to none it's faster by 40% than with filmic+contrast

depending on the scene and setup but min 20%

I've tried with default color management and factory settings, it's less but still make a difference.

aliasguru

@Mareck CPU or GPU? Or doesn't matter at all?

@Mareck just so I'm testing under your conditions

Mareck

GPU Blender 2.78a W7 x64

aliasguru

no difference at all for me

both renders took 3 mins 11 secs, 48 squared samples, path tracing, full HD

@Mareck surprised it makes any difference for you. The CPU should be free to do these transformations while your GPU is busy knocking out the tiles

@Mareck are you close to reaching the GPU memory limit?

Mareck

@aliasguru really strange, perhaps an addon that make blender react like that.

@aliasguru one Titan X + 780Ti so I think it's ok from this side

aliasguru

16:54

@Mareck running the test again just to make sure

troy_s

@Mareck When the work is done, I'd strongly encourage CLI. If you have a few extra computer cycles kicking around, you can break complex images or series down across machines.

@aliasguru You seeing any difference. I mean a 3D LUT is going to tax some cycles (not Cycles) but I am struggling to see a case where 40% is plausible.

@aliasguru You have me hunting for spectros again. Almost bought the ColorMunki Photo about eighteen times.

Mareck

@troy_s you're right for the CLI but usually there are lots of point to forget before lunching a render, so with CLI you have to wait till the end to see if everything has gone well.

troy_s

@aliasguru The more I have thought on it by the way, the more it is plausibly the most ideal instrument in your case for screens. It's inexpensive (in terms of spectros), behaves well with Argyll / DisplayCAL, covers screen and print (which is a bonus, but can't do artificially augmented fluorescent whiteners due to the UV Cut, which is not a very big deal)

aliasguru

@Mareck third test render just finished, Filmic Views + own color correction profile on top of it. Same render time

@troy_s shame you're not next door. I could buy one and share it :)

troy_s

@Mareck Except you should already know your render is good via previews and such. Then fire it to the farm. At least, that's what I'd endorse. You run the additional risk that on a heavy render that you could encounter a crash due to the UI, which is eliminated via CLI.

@aliasguru I don't think people appreciate the cost on it. Spectros are not cheap. Have a look at even an I1 spectro, which is sort of the baseline for print work, then you get into Jeti and such. At 350$ USD, it's basically a steal.

Very cost effective.

@aliasguru Also, I forgot, there should be a method / function within DisplayCAL that lets you yank the corrections off the site / installation media.

@aliasguru There is very likely a correction that is of the sort for that style of panel.

aliasguru

17:01

@troy you mean like automatically get the Quato correction files for Wide Gamut mode?

@troy_s That'd be awesome

Mareck

@aliasguru Ok, I'll do a test with CLI when I can.
@troy_s yes right too, but still need to find what's wrong on my Blender
I'll investigate more when I'll have the time

troy_s

@aliasguru Bingo.

@aliasguru What it would likely be is a set of standardized corrections that push the filters beyond their marked xy positions. Likely to some "wider but still accurate" ranges.

aliasguru

@Mareck can you open the console window and see if any error messages are printed?

@troy_s and another can of worms open

troy_s

@aliasguru See section 3 of this displaycal.net/#quickstart

aliasguru

@troy_s I've had a look at the scene again today, to see where the huge differences come from

troy_s

17:04

@aliasguru Not entirely. It should actually be pretty painless if you can identify the type of backlight IIRC.

@aliasguru Oh interesting.

aliasguru

@troy_s biggest surprise was, that actually shaders had not (yet) been changed at all

troy_s

@aliasguru Nope. Expected.

aliasguru

@troy_s but lights were

troy_s

@aliasguru That said, I'm sure you'll see quite a difference if you do some ballpark estimations compared against albedo charts. For example... not sure if you have seen this, but there's some pretty useful information in here if you can kindly turn a blind eye to the sRGB junk.

aliasguru

@troy_s instead of the standard blender harsh area lights, Joe used huge emissive planes with a procedural texture

troy_s

17:05

allegorithmic.com/pbr-guide

@aliasguru Bingo. I try to tell people this all the damn time but not many listen.

@aliasguru I'd like to think that at this point, 20+ years into my particular industry, that I can offer at least a little bit of insight into how to light. Also see this loser: sacred-texts.com/aor/dv/0520.htm

aliasguru

@troy_s it was nothing more than ramps mapped across over each other. Gave a simple disc with very soft falloff in the end

troy_s

(Note the dimensions he references is 10 braccia by 20 braccia, which was a floating measurement of a forearm. 1 braccio = 21.344 in)

@aliasguru Big sources are the secret to effectively all aesthetic driven by the late Renaissance / Dutch Masters / Caravaggio styled lighting that still has an iron fisted hold on contemporary studio lighting.

aliasguru

@troy_s Yep. And it's so simple.

@troy_s two of those lights gave the whole impression already, the rest was only accentuating things

@troy_s finally they started putting something together for Blender to have textures on lights

troy_s

That's like the cliche really.

@aliasguru I chatted with Brecht ages ago about offering a transmission versus diffuse set of variables. In the real-world, there are literally 10000s of diffusions and every director of photography has a preference for their basic sets.

@aliasguru I am pretty sure that if you had essentially two variables, you could simulate it. It's more or less similar to SSS in thin sheets.

aliasguru

@troy_s that stuff hanging from the ceiling is worth more than the car

troy_s

17:14

Anyways... here's the documents worth looking at... I think they have albedo charts...

allegorithmic.com/pbr-guide

@aliasguru A lens set is worth about 6x as much as that car.

(Arri's Zeiss Master Primes are about 25000+ per lens, and I think there are like 14 of them now.)

aliasguru

troy_s

@aliasguru If you read that passage at Sacred Texts, do the math on the braccia, you'll see immediately that old numbnut there was using roughly a 20x sized diffusion cloth in front of his window. In close to agument the falloff.

HA. Paint work right?

aliasguru

playing a bit with Filmic low contrast, that lighting and a tiny bit of SSS

troy_s

Is that close to what you see in reality?

aliasguru

no grading, no comp

troy_s

17:16

(I have no idea, only those amped up sRGB images... which leads me to believe it is more orange than that?)

aliasguru

no not yet.

but I didn't even intend to try that in this image.

troy_s

God. Chilling how such simple renders end up tickling the "photograph" response.

Is there subtle DOF on that?

aliasguru

so far I got closer to that damn orange, but havent nailed it

no, no DOF

and it's only 100 samples

troy_s

Interesting. I'd drop in a tiny smidge, just to take it out of the physically impossible domain.

(The little accumulation of things that can't happen in reality sum up to tremendous impact. Gleb is excellent at that, and @tynaud's effects pedal does it in spades.)

@aliasguru Is that with a PBR chain?

aliasguru

partially

base is standard diffuse added with SSS

then a pbr coat on top

basically the cynicat pro implementation

that would be the PBR Ubershader we put together

Mareck

17:20

@aliasguru isn't the Cynicat fresnel too dark at F0?

troy_s

@aliasguru That's similar to Jason Clarke's, he used some of Cynicat's work I believe.

jtheninja.wordpress.com/2016/09/18/simple-pbr-v5

aliasguru

@Mareck he had more than one version of it. One which is very flexible was able to define the amount of reflection on F0

Mareck

Ok

aliasguru

@Mareck So you had Roughness, Power, F-0

troy_s

@aliasguru @Mareck What were the core issues?

Mareck

17:24

@troy_s with F0?

The fresnel I have from Cynicat is RGB 0,0,0 for F0

troy_s

@Mareck I immediately think that is probably broken given that 0.0s are hard to find anywhere in plausible reality.

(as a general rule it's pretty safe to guess that any "maximum" or "minimum" values would never occur if we reconcile real-world models against our internal ones.)

aliasguru

Mareck

And minimum F0 for dielectric is 2 to 5%, but if it's for flexibility it's just a matter of knowing how to use.
Yes it was perhaps 0.0002 ... ^^

aliasguru

@Mareck guess you have the last setup

troy_s

@aliasguru Have you tried the bike with one of Maxime's HDRIs?

Would be very interesting seeing the shaders behave in a complex emission environment. Probably reveal issues pretty quickly too.

aliasguru

17:29

@troy_s not yet, had some client feedback on something else today

@troy_s BUT

troy_s

@aliasguru Joy!

aliasguru

@troy_s I have time now

:0

Mareck

@aliasguru @aliasguru yep this one

troy_s

I'd regale you with a wonderful story from an idiot related to a performing artist that was lecturing me on photography, but I won't...

aliasguru

@Mareck I tend to use both, but in combination with the Filmic Views, the 0.04 F0 is almost too much

troy_s

17:30

(Needless to say I politely slapped them publicly in email, only to have the idiot completely mistake my slag for some sort of concession due to his hubris. sigh)

aliasguru

@troy_s Oh please, yes!

troy_s

@aliasguru Is this because of the light levels?

@aliasguru What has been revealed is that if you generate quality materials that imitate reality to the best the models can, they hold up at all light levels. The issue is that many imagers end up stuck with the WYSIWYG thinking, and they end up cheating sRGB EOTF views because there isn't enough light and the rolloff is whack.

Mareck

@troy_s Oh no you were right about rendering with the UI, blender just crashed ^^.
But it was a bake not a render.

troy_s

@Mareck I was under the impression bakes fail miserably for photorealistic bits?

Mareck

@troy_s trying for sake of comparaison between the no cheat render of the bathroom

troy_s

17:34

Interesting...

Mareck

@troy_s what are emittance and sri?

troy_s

@Mareck I think it is en.wikipedia.org/wiki/Thermal_emittance

Mareck

@troy_s ok so not related to CGI

troy_s

Looking at a few charts is rather useful as you can get a pretty good idea where your albedo is shitting the bed.

Mareck

@troy_s yes nice ones, I haven't reallized so many materials are close to 0

aliasguru

17:44

@troy_s Albedos for different skin types? Would that pass a racism test? :)

Mareck

@troy_s sorry my mistake linear values
@aliasguru ^^

troy_s

@aliasguru The fact that there are skin types passes it. If you see just "Skin" you are uh... yeah.

@Mareck 0.05 is as low as I've seen.

@aliasguru LOL youtube.com/watch?v=KpwrNXiDGfM#t=7m45s

aliasguru

@troy_s I'm becoming an internet star ROFL

troy_s

I tweeted him and he said he was using them in the videos... but I don't see it.

Interesting.

aliasguru

@troy_s remember I gave him that tip for the mouse clicks as a reply on BlenderNation if I remember correctly

troy_s

17:51

LOL.

Ok shower, then wavelength emulation for RGB desaturation. Time to tackle it.

aliasguru

@troy_s better have something solid to eat first, it sounds heavy

anyone seen this?

3dlutcreator.com/index.html

troy_s

@aliasguru Generating canned looks is very simple. If there is a need I can demo. The BSE answer covers most of it.

aliasguru

19:32

@troy_s the canned look is an interesting topic in itself. But what we found really interesting is how much change was applied with seamingly very little artifacting

troy_s

@aliasguru You are talking about what manifests purely on the grounds of additional light?

A single light source won't reveal it, but a mixture of > 1.0 light sources will. Those ratios have a huge play on materials, in ways that aren't entirely obvious I suppose.

(That is, two or more lights with alternating ratios. A single source render at Y EV is identical to Y*X EV)

aliasguru

@troy_s confused again

troy_s

@aliasguru As I understand it in raytracing terms... assume render at X EV

Arbitrary number of emissions lighting the scene.

Assuming everything remains constant, and you increase the lights precisely by identical ratios

then adjust exposure down (that is in math terms to multiply by the value you calculate that coincides with the EV shift)

The images are identical.

Make sense?

aliasguru

@troy_s so far it does

Andrew Li

Are general blender questions allowed here?

aliasguru

19:41

@troy_s You've seen how far he shifts hues from one range to the other?

troy_s

@AndrewLi We vote on questions and execute anyone whose question fails the vote.

aliasguru

@troy_s if the math was so easy, why to other imaging applications sometimes fail so miserably on this task?

troy_s

@AndrewLi There are better places to ask though. :)

@AndrewLi I'm kidding of course.

@AndrewLi But I suspect you can find better solutions for some subjects elsewhere. Fire away?

Andrew Li

Well, I'm not sure if I should ask a question, think it'll be pretty quick

aliasguru

@AndrewLi Questions are always allowed, but answers might be forbidden

troy_s

19:42

@aliasguru Knowledge. Legacy. Corporate reasons. You name it.

aliasguru

@AndrewLi kidding too of course, go ahead

troy_s

It's great to make the snap and go "Holy shit Photoshop should be using a scene referred model!"

Andrew Li

So I have the mirror modifier on my mesh, and there's some 'tearing'

Just black areas where it didn't merge correctly

troy_s

But that's a huge leap when you realize an 18 billion dollar company or whatever with how many millions of lines of code to change and how many of those billions rely on a "legacy" compatibility etc.

It's a huge, dare I say, unsolvable task.

@AndrewLi Definitely try RenderFarm or #blender on Freenode. Too many reasons there. If it is alpha related there's more than a few that can help here. ;)

Andrew Li

Ok

aliasguru

19:44

@troy_s maybe. On the other hand, maybe not. Maybe it's a better idea to replace it with something less "broken"?

Andrew Li

Thanks :^)

troy_s

@AndrewLi This little gathering hole is mostly dark and arcane secrets and myths.

@aliasguru But you see, look at BSE...

Andrew Li

i.sstatic.net/Et4M0.png - not sure why

troy_s

If you tell your baseline average person that "Your whole concept of colour is using mixed up models that aren't effective to the idea of photography", aside from being a massive black hole of a rabbit hole, you are going to face serious pushback.

@AndrewLi That looks like a crumpled bunch of faces collapsed and gobbling up light in there, but I'm about the furthest thing from a modeller you could describe. My 3D absolutely sucks.

aliasguru

@AndrewLi try increasing the merge limit

@AndrewLi Mirror Modifier, bottom option

Andrew Li

19:47

@aliasguru Done that. Still some tearing

Looks like it's even more prominent with the suggested 1mm

aliasguru

@AndrewLi are other modifiers before the mirror one possibly deforming the mesh?

Andrew Li

Don't think so. I've applied my other modifiers already

I wasn't aware of the tearing until after I was done with the modifiers. Could subsurf modifiers be affecting it?

aliasguru

@troy_s that's one issue. If a topic is that mystified like the whole color thing, you'll meet more people who believe than people who know

Andrew Li

Or is there a way to just merge the verts? It's really ugly and it's completely around my mesh

aliasguru

@AndrewLi not if they come after the mirror in the stack

@AndrewLi have you uploaded it somewhere?

Andrew Li

19:50

No

Should I? Are you willing to take a look?

aliasguru

@AndrewLi would be around for a few more minutes

Andrew Li

Ok

aliasguru

@troy_s I've read through part of the Tasks on developer.blender.org that you commented on

troy_s

@aliasguru Hence why you can hear me wax lyrical in places like this or Freenode, but rarely out randomly. If a question hints at it, I'm there.

@aliasguru It is rage inducing there. I can't really comment on it. Blindingly frustrating rage here.

I always cite this one as one that I can only look at for exactly three seconds at a time: developer.blender.org/T47212

aliasguru

@troy_s one can only be patient about it

troy_s

19:53

So when you hear "Patches Welcome" I'm more like "SHUT THE FSCK UP"

@aliasguru Sorry no.

@aliasguru It is outright ignorance at that point. Wilful junk logic, junk reasoning, junk processing, junk lies.

Just shut up and do it.

It goes a little like this:

Andrew Li

@aliasguru I haven't used Blender much before, so it may be a bit suboptimal to work with: pasteall.org/blend/index.php?id=44444

troy_s

"Patches welcome"

"Does anything else do it this way?"

"We need to think about legacy"

CUT TO:

"We've completely redone how this shader works. Hope you like it!"

So I'm over it. I don't have time for petulant children.

Side note... see ramps that influence colour like this: i.sstatic.net/Glsxl.png

It's for a sun, like this: blender.stackexchange.com/questions/40987/…

Would never dive into that conversation, but you can see an interesting thing with how complex folk's thinking is by looking at the sun "colour"

aliasguru

@AndrewLi if you turn the mirror off you'll see what happened

troy_s

(and more interesting if you think about what colour the sun is as you think of it in a photo and exactly what moving parts contribute to it.)

aliasguru

@AndrewLi you accidentally deleted some faces in the front

Andrew Li

19:58

Hm

aliasguru

@AndrewLi Now, since the merge threshold is high, Blender starts to weld vertices together that should have never been

@AndrewLi this mangles your model

Andrew Li

Ok

You're right. Any idea how to restore the faces?

aliasguru

@troy_s sigh..

troy_s

@aliasguru Indeed. That bug report was the last straw for me. As well as other goofy stupid goddamn comments out of significant players.

I grew so damn tired of trying to explain to people how having an unmanaged UI was absolutely ridiculous.

Andrew Li

Also, @aliasguru - there's some subtle tearing throughout the model, almost like a seam. Sorry for all the questions, but is there a way to smooth that out?

aliasguru

20:01

@troy_s The issue really seems to be that he can't understand the larger meaning

@troy_s as there is no move to other color spaces yet

troy_s

@aliasguru Sure. And you know what? Saying that is 1000000x better than posting that shit.

But alas, hubris and ego.

@aliasguru I've actually been more or less discussing colour around Blender since well before the Matt Ebb era, and Matt was the first peep around the BI that was able to chat about things.

Back then, there wasn't even an awareness of the idea of what "linear" meant in terms of display linear or scene linear (that point will still make a majority boggle at.)

But the missing component was understanding RGB as lights.

At any rate, it was effectively this post that led to OCIO, and still folks don't know or understand why it is there really...

mango.blender.org/production/…

(See in comments. At that point I did a demonstration on how to get from the Sony encoded SGamut to 709 on the BFx list I believe.)

aliasguru

@troy_s I also remember a discussion regarding the Gooseberry Pipeline, an article you postet

@troy_s Ah, I see Gez also joined that discussion

@AndrewLi Starting point: pasteall.org/blend/index.php?id=44445

@AndrewLi but would be much better if you went back to the version where modifiers were not applied (yet)

Andrew Li

@aliasguru thanks so much. May I ask what you did?

aliasguru

@AndrewLi topology flow by the way should be optimized

Andrew Li

Ok, thanks!

aliasguru

20:09

@AndrewLi crudely removed faces in the center to create a gap all around, select the boundary, and extrude towards the center with clipping turned on in the mirror

Andrew Li

Sadly, I don't have backups setup yet, should probably do that

aliasguru

@AndrewLi this at least merged the result

Andrew Li

Gotcha, thanks again!

aliasguru

@AndrewLi you can still use your base to retopologize

Andrew Li

Yeah

aliasguru

20:11

@troy_s that discussion is a bit shocking

troy_s

@aliasguru Because Gez was about sixteen iterations advanced at that point. He's a print campaign person from having done many advertising runs, so the idea of colour management was more or less foundational knowledge.

@aliasguru I did. I also mentionned it recently... "Uh... Like why not aim at REC2020 so you will see the work demoed on every 2020 display and HDR capable display everywhere?"

CUT TO: Crickets chirping.

@aliasguru The reason crickets chirp when you ask that, is because again, the idea of RGB is such a basic banal concept to anyone that has ever touched digital imaging, and that banality gets ever lower the longer you do it.

You don't even think about it.

But when something as simple as "What does changing an RGB value do?" is asked,

it is remarkable that very, very few imagers have even asked that

Yet at the core, there is the root of all evil conjecture. :)

aliasguru

if it was only RGB... But there is that A appended to it as well...

and those bits...

8, 16, 32,...

Andrew Li

Is it a good idea to always apply modifiers before sculpting?

aliasguru

@AndrewLi most of the time you have to. but keep a copy of the mesh around to be able to start over

Andrew Li

Ok, thanks for all the advice!

aliasguru

20:16

@troy_s With "changing RGB value", you mean the chromaticity or the intensity?

@troy_s Gez comments this: What you need is to map colors from S-Gamut into a more reduced space, sRGB in this case. For that a “rendering intent” has to be used. A rendering intent is a strategy to decide where to put all the colors present in the larger gamut space into the sRGB space.
Each value has to be re-mapped. For instance, the green shade of the greenscreen background has a “coordinate” in S-Gamut and a different one in sRGB.
Applying no remapping, S-Gamut’s green stays out-of gamut and instead a duller shade is used.

@troy_s on the question why the images became so dull

@troy_s but i really have to say, those AdobeRGB reds look a lot more fresh than the sRGB ones LOOL

troy_s

@aliasguru Shup you!

LOL

@aliasguru In what context?

@aliasguru Because your statement is both true and false.

@aliasguru False in the sense that AdobeRGB's red is identical to 709's red

@aliasguru True in the sense that, given a scene of mixtures of colour, that given AdobeRGB green is significantly different than 709's, that the sense of yellow (compliment) may contribute to your experience of the AdobeRGB red as "more red"

aliasguru

@troy_s I'm just kidding. I remembered the top rated post you sent me regarding exactly this discussion by that photographer

troy_s

@aliasguru Duller is not great here either. Less saturated mixture of light is more accurate, that is, more compliment is added. That of course holds true in any space, but the ratios mean totally different things of course.

aliasguru

@troy_s with the color samples

troy_s

@aliasguru Yes. God yes. That nightmare.

@aliasguru "Here is sRGB's red! Here's an AdobeRGB encoded image of that red, dumped without any transform!"

Ugh.

It's face slappingly nightmarish.

But where do you start? I don't. I just wait for someone to come along with interest.

aliasguru

20:25

@troy_s that must lead to an article from some other guy who's doing the same thing upside down, drawing the opposite conclusion

@troy_s uh, one question I forgot to ask

@troy_s slipped into my mind when you sent that link to cynicats tutorial on black and white point round

troy_s

@aliasguru Yes?

aliasguru

@troy_s when I set them in Blender, do the value ranges get crippled? Like what happens in Photoshop if you apply levels. Know what I mean? Or does math precision save me?

@troy_s in other words, I'm not sure when the curves are applied. Before the View Transform? After? Before conversion to sRGB space? After?

troy_s

@aliasguru I'd call this a third level bit of thinking. You've grabbed onto scene referred values, you've interpolated that concept and applied it to white balance etc...

@aliasguru So shorter answer:

1) Everything in Blender is post display referred transform to the best of my knowledge. Absolutely everything.

2) That means that everything operates on scene referred linear values.

3) When selecting white and black, that is more or less making the graph cover a larger range, although I seem to remember having to turn off clipping and actually moving the graph to actually adjust certain values.

aliasguru

@troy_s that should be good news

troy_s

Now... what unfolds from this...

@aliasguru Q1) Let's assume you leave everything exactly at 0 and 1.0

What happens if you grab the 50% left to right area of the curve and pull it down?

aliasguru

20:33

I shall darken the "mid greys", a bit like a power function

gives me less space for dark shades to actually be mapped

I'd expect details to be lost there

when mapping to display referred

troy_s

@aliasguru Now interestingly... read what I said above.

@aliasguru It in fact will adjust value 0.5 down.

Does that adjust middle greys?

aliasguru

it adjusts energy values

wait

you mean, mid gray is at 0.18

I'm fiddling with linear values

right?

troy_s

@aliasguru You are indeed.

So middle grey isn't at 0.5, but rather 0.18 in Filmic (could be anywhere really, but in filmic we put it at 0.18)

Have you ever tried to actually adjust shadows using the curves in Blender?

Go give it a try... from middle grey down. :)

I dare you.

aliasguru

@troy_s And I always thought something is broken :)

@troy_s I could file a bug report on this and make some devs happy :)

troy_s

@aliasguru The problem is, you file a bug and the short solution is... "Make the scale "nonlinear!"

So you go and make 0..1 nonlinear according to what? sRGB?

Oh it's broken again.

Worse, now when you want to tweak your normals map using a curve, it's nonlinear.

Oh god.

See the rabbit hole suddenly open up?

(There is a solution that solves all of these design problems with regards to the audience and needs in this particular instance.)

tynaud

20:42

@Mareck pasteall.org/blend/index.php?id=44417 I had posted this on BL, have a look, it's from cynicat.

aliasguru

@troy_s knee-deep in it

troy_s

@aliasguru (The actual solution is to permit transforms on all UI elements, such that an imager can calculate what they are attempting to achieve in any given context and apply the correct transform)

tynaud

@Mareck No difference in render time with two different scenes, using filmic + contrast, no transform at all and masking the window...

aliasguru

@troy_s clean but a huge task

@troy_s with 2.8 currently running, and all the ambitious targets on the list - not seeing any color topics on them however - that's a while away

troy_s

@tynaud This is positive!

@aliasguru Which is also why when you sort of dive into the rabbit hole you can begin to see how important tabling these things is.

Again, the reason they aren't in everyone's view is largely because of a complete underlying misinterpretation of what RGB is.

It's accepted as some sort of universal constant.

But newsflash, my insisting on UI being managed? Think about every Apple device in the modern era

The UI dumps values to display, no transform.

How do those UI elements appear? :)

(I have a longstanding idea that I should rotate all primaries so that when people go willy-nilly ignoring colour transforms or trying to select colours, they get an immediate and visceral "Whoa this is wrong" sensation)

It would be smarter to rotate all primaries and have people become acutely aware of the issues in the modern imaging era than have things that might appear to some as being sorta kinda roughly weirdly sorta close.

@aliasguru My issue with being saddled with drastic changes that don't consider these things is that for the foreseeable future, we are going to be stuck with the decisions.

aliasguru

20:56

@troy_s until things break fundamentally and need to be tackled

troy_s

@aliasguru You mean like curves?

;)

@aliasguru I don't think that is an accurate statement.

aliasguru

@troy_s I mean like rotated primaries

:)

troy_s

@aliasguru Yes, that would make it clear.

(It is also how I troubleshoot things on the actual colour front.)

(as an aside, think about various reference spaces you can have in visual effects or rendering and how you use a colour picker... :) )

aliasguru

I found a picture of the rabbit hole. If you look at it from outside, you'll see the entire truth: de.123rf.com/…

@troy_s allright, time for bed

@troy_s hope color pickers are not chasing me tonight

@troy_s in my dreams....

troy_s

@aliasguru Night bro.

Chat soon.

Sleep well.

Mareck

21:30

@tynaud thanks for the test, and I'll have a look at the shader.
@troy_s you are brave, really. Thanks to still be here explaning us why we are blind.

troy_s

22:11

@Mareck Bah. It isn't like I am delivering some magic to all the peasants in the farms. I've always been about imaging and anything that can help an imager's work I'm all for.

Mareck

It's exactly for that I thank you
And by reading the conversation with aliasguru I'm undestanding what you are doing since some time now, and how it's difficult when you have a truth but not a lot can see it.

troy_s

22:34

@Mareck I find most things to be contextual. In this case, it would certainly be helpful if more folks around the culture were up to speed so that they too can see the issues. Remember, we aren't discussing an esoteric edge case but RGB, the actual media of the output!

Mareck

@troy_s It's what is crazy and why the red pile comes to my mind, but at that time I wasn't seeing the half of what I see now.

I still need to learn more about rgb triplet light, the xyY space (if it's a space)... but not tonight, I'll go to bed.
See you @troy_s

Transcript for