[RESOLVED] Linear Workflow (LWF)

[cjwidd]

Can someone tell me something about color management that I will understand?

I have already scoured the internet for articles about linear workflow (LWF) for just about every DCC application. Here are a few examples:

Understanding sRGB, gamma, color values
What is linear workflow and how can it help your renders look better?
Linear Workflow made Simple
ACES Primer, Glossary and Quick Start Guides
ACES and the ColorChecker
Time to ditch sRGB/Linear as default

I need to couch the following discussion in terms of a practical example - it is the only way I will understand:

1. The following settings are enabled in 3ds Max Preference Settings (see below). To be consistent with LWF, are these the appropriate settings? If not, what are the appropriate settings?

• Customize > Preferences > Gamma and LUT > Enable Gamma/LUT corrrection (checked)
• Customize > Preferences > General > Color Selector: Corona Improved Picker (sRGB unchecked)

2. I have an HDRI (Jørgen Herland; 'jh_int_mediaoffice_2015_8k') which is seamlessly stitched, color calibrated, white balance corrected, and has an unclipped dynamic range. The HDRI is imported with Corona Bitmap (see Figures 1 & 2). To be consistent with LWF, what are the correct texture import settings for HDR images?:

• a. Automatic (recommended)
• b. Override (1.0)
• c. Override (2.2) (*identical to (a.) if Gamma/LUT corrrection is enabled in 3ds Max Preference Settings)

Figure 1


Figure 2


3. I have a texture set (Quixel Megascans; 'Old Log Wall'). The texture set is imported with Corona Bitmap with the following settings (see below). To be consistent with LWF, are these texture import settings correct?

• ti5teabbw_8K_AO - override (2.2) (*identical to 'automatic (recommended)' if Gamma/LUT corrrection is enabled in 3ds Max Preference Settings)
• ti5teabbw_8K_Albedo - override (2.2)  (*identical to 'automatic (recommended)' if Gamma/LUT corrrection is enabled in 3ds Max Preference Settings)
• ti5teabbw_8K_Gloss - override (1.0)
• ti5teabbw_8K_Normal - override (1.0)
• ti5teabbw_8K_Displacement - override (1.0)

4. I have a Corona Camera (50 FoV) with the following post-processing tone mapping settings (see Figure 3; ACES emulation). To be consistent with LWF, what are the appropriate post-processing tone mapping settings? To be consistent with LWF, what (if any) LUTs should be applied to the camera, and why?

Figure 3

5. Test render using the combined settings described above.

Figure 4

[cjwidd]

Is linear workflow literally just referring to AOV compositing, i.e. no tone mapping in camera so that you can linear dodge (add) render elements in Photoshop and then color grade?

[Juraj]

I'll just comment on this:

a. Automatic (recommended)
b. Override (1.0)
c. Override (2.2) (*identical to (a.) if Gamma/LUT corrrection is enabled in 3ds Max Preference Settings)

Linear files are always loaded as "override 1.0", that's also what "automatic" is for them in every occassion regardless of what 3dsMax Preference settings are set to.
That's why you can just always load HDRi as "automatic" and never think about it.

Everything in newer 3dsMax and Corona (and every current version of all renderers) on market are automatically setup correctly to work, the user doesn't have to do anything. The renderer works internally 'linear', and it outputs files dependent on your file format (linear for 32bit files, gamma corrected for 8/16bit files. 16bit .exr is special situation and is treated as 32bit except having half the information).

The whole "LWF" conundrum is ancient issue from early MentalRay/Vray times transition into modern age.

Is linear workflow literally just referring to AOV compositing, i.e. no tone mapping in camera so that you can linear dodge (add) render elements in Photoshop and then color grade?

Yes, in regards to compositing post-production, it has to be done fully linear (no post, not LUT, Vray for example doesn't bake down LUT, it only uses it to preview to avoid this issue) to give the same result as renderer does internally. Color grading can be done in any gamma and color space of choice (from linear, to LogLUT, to gamma-corrected files) depending on industry and user preference. This is where ACES comes in.

[cjwidd]

The language surrounding linear workflow is pretty confusing - should just call it 'gamma 2.2 workflow'.

Well, that certainly helps to clarify my underlying question, i.e. whether linear workflow is really contemporary compared to ACES. It seems that some proportion of the industry is leaning into ACES going forward.

[cjwidd]

Came across this video from CG Bandit (2017) which is describing some sort of VFB hack(?) using a third party script to adjust gamma somehow. Is this really necessary or is there a workaround to achieve a similar result?

[Juraj]

There are no hacks necessary.

Here is primer for quick Corona workflow:

1) Install & Hit render.
Now two possible choices:

2A) Do you plan to composite essential passes in editor like Nuke, Fusion or AfterEffects or make heavy post-production that involves changing of exposure and tonemapping?
Then adjust 'Exposure' and nothing else. (Bloom & Glare is debatable, but don't bake it down ideally, but once you go to this length of process, you might as well get much higher quality B&G in post). Save as 32bit .exr or 16bit .exr if you need to save harddisk space.

2B) Do you plan to only composite creative elements ("photoshopping stuff") or basic color adjustments like in photography or illustration?
Then adjust everything you see in VFB to your liking. Save as any format you personally like, but 16bit TIFF is the most universally supported one.

[cjwidd]

Then adjust everything you see in VFB to your liking.

My issue is sort of with this idea. Given the freedom to basically dial the image based on taste alone does not provide enough constraints if the goal is to arrive at an image that emulates a photograph captured through a real camera - obviously we can spin dials until we arrive at a pleasing result, but there needs to be a method to arrive at a predictable result.

It seems like a lot of CG professionals are concerned with emulating camera response curves and the application of LUTs is a part of that process. For example, Dubcats LUTs which attempt to emulate the 'look' of various camera types. Something that I think was missed, on my part, is that any tonemapping that is applied *in addition to* the LUT is purely discretionary, i.e. the LUT is meant to describe the look of the image in its entirety, according to the LUT; if you do additional tone mapping adjustments, that is purely based on personal taste.

What is pointed out in that video (3:30) is how Corona is was combining direct and indirect light and the effect the gamma curve has on the appearance of grays in the final image. The video suggests that Corona is was combining direct and indirect light (by default) in such a way that the final rendered image is not deep (black) enough, i.e. midtones are dominate. When I look at imagery from archviz studios, the images have a sort of a deep, contrasted, volumetric appearance. I am assuming color management and LUTs are a part of that.

This video (TUTORIAL: CESSENTIAL Render Elements (1/3: Elements and Output Setup)) describes linear workflow exactly, AFAIK. So all this talk about linear workflow really just refers to (what is also known as) 'back to beauty' compositing - meaning, "don't apply any color transformation to the render elements that will break the linearity of the output, i.e. LUTs, contrast, highlight compression, etc.

I assume *that* was why there was so much disagreement in the Time to Ditch sRGB/Linear as Default thread, because some prefer to do all the look development in camera, and others want to do it in post - or something like that.

[Juraj]

Ah...that is lot of information mixed. Not sure I can answer to that.

In all these threads, terms are mixed around and given new confusing meaning. People imagine behind "LUT" or "Tonemapping" or 'Color management" something completely else than it is.
I have already given up on taking part in those discussions because they reach almost mythical levels on wishful thinking and holy grail searching.

[cjwidd]

Yes - I understand - the topic can be exhausting. Thanks for sharing your thoughts regardless, it does help clarify the situation.

[JViz]

Then adjust everything you see in VFB to your liking.

My issue is sort of with this idea. Given the freedom to basically dial the image based on taste alone does not provide enough constraints if the goal is to arrive at an image that emulates a photograph captured through a real camera - obviously we can spin dials until we arrive at a pleasing result, but there needs to be a method to arrive at a predictable result.

It seems like a lot of CG professionals are concerned with emulating camera response curves and the application of LUTs is a part of that process. For example, Dubcats LUTs which attempt to emulate the 'look' of various camera types. Something that I think was missed, on my part, is that any tonemapping that is applied *in addition to* the LUT is purely discretionary, i.e. the LUT is meant to describe the look of the image in its entirety, according to the LUT; if you do additional tone mapping adjustments, that is purely based on personal taste.

What is pointed out in that video (3:30) is how Corona is was combining direct and indirect light and the effect the gamma curve has on the appearance of grays in the final image. The video suggests that Corona is was combining direct and indirect light (by default) in such a way that the final rendered image is not deep (black) enough, i.e. midtones are dominate. When I look at imagery from archviz studios, the images have a sort of a deep, contrasted, volumetric appearance. I am assuming color management and LUTs are a part of that.

This video (TUTORIAL: CESSENTIAL Render Elements (1/3: Elements and Output Setup)) describes linear workflow exactly, AFAIK. So all this talk about linear workflow really just refers to (what is also known as) 'back to beauty' compositing - meaning, "don't apply any color transformation to the render elements that will break the linearity of the output, i.e. LUTs, contrast, highlight compression, etc.

I assume *that* was why there was so much disagreement in the Time to Ditch sRGB/Linear as Default thread, because some prefer to do all the look development in camera, and others want to do it in post - or something like that.

I agree with Juraj on many points here. all you need is either "linear" 16 bit tiff image out of the VFB, or you could compress the highlights in the VFB and maybe add a bit of contrast, then save as 16 bit tiff. what you do next is up to you, it really is taste from this point forward. a good way to understand how the look of an image you like was achieved is to just open it in photoshop and look at the histogram, check the blacks, whites, the spread of the values etc. there are no specific HOLLY GRAILS but each artist have a way of doing things and that's why each artist's work is unique.

I quite like dubcat's 0.7 2.4 -0.03 1.0 1.0 setup but it's just a start, you have to use photoshop's power. the only way to clarify it further is taking a look at the work of professional interior photographers, it's not about realism, there is no way a scene would look like some of those images in reality, they add so many layers of processing on top of (photographs) until they look how they want them to look. or they just choose great compositions and add a bit of contrast and that's it, it really is taste.

my view is that visualisers need to stop wasting their time inside of 3ds max tweaking at stuff. leave things at default (mostly), add nice light, angle it nicely, play with the shadows, and you will get a flat looking image that looks like puke, save it as tiff 16 bit take to photoshop and prepare to be amazed. unless some deep level compositing is needed, I have no idea why anyone in the archviz world would need to save a 32 bit image.

[maru]

Color management, tone mapping, linear workflow - all of this can be pretty confusing. But based on my experience (which is not that great in this regard unfortunately) and mostly on what people more knowledgeable and smarter than me told me, the LWF craze is a thing of the past, and we should just forget about it. Modern software like Corona and newer versions of Max have everything set up correctly by default and you just don't have to think about it. Older software usually had a single checkbox that you would need to toggle depending on whether you were saving to exr or doing post in the VFB or AE, so that wasn't a big deal either. :) 

So basically I can confirm everything Juraj wrote in this thread.

[cjwidd]

I was just really confused about what exactly was encapsulated by the linear workflow when I wrote this post. From what I gather, it just means rendering out AOVs with no post so that you can linear add them all back together in PS.

I don't know why that was so confusing at first, probably because it implies gamma correction, and that's when things get pretty squirrelly.

[JViz]

I was just really confused about what exactly was encapsulated by the linear workflow when I wrote this post. From what I gather, it just means rendering out AOVs with no post so that you can linear add them all back together in PS.

I don't know why that was so confusing at first, probably because it implies gamma correction, and that's when things get pretty squirrelly.

Yep you're right. Glad it's all clear now. Just render and find a unique editing style and that's it really.

[romullus]

Not really. What you save and how you do it, has nothing to do with so called linear workflow. What you feed to the renderer and how it performs its calculations internally, that's what important. But as others already said, you don't have to sweat about it novadays, everything is done automatically. I'm sure many still remember that chaos when industry was transitioning to LWF :]

[JViz]

Not really. What you save and how you do it, has nothing to do with so called linear workflow. What you feed to the renderer and how it performs its calculations internally, that's what important. But as others already said, you don't have to sweat about it novadays, everything is done automatically. I'm sure many still remember that chaos when industry was transitioning to LWF :]

Forget the terminology, he's just saying that he only needs to care about the image being linear when he wants to do compositing in 32 bit, which is correct. Let's not confuse the guy even further, jeeeez