Author Topic: Tonemapping - Plz Halp  (Read 83538 times)

2020-03-11, 06:34:02

cjwidd

  • Active Users
  • **
  • Posts: 1069
    • View Profile
    • Artstation
Bit of a rant - or a cry for help - but I'm either approaching lighting or tone mapping totally incorrectly; it is absolute guesswork every time, and rarely is it 'photorealistic'.

Looking at Juraj's work, Bertrand's work, etc. - there is a consistent look and quality with each project and I'm just not convinced (for now) that they are also, just guessing; I assume they have some sort of method - a procedure. For example, when I worked in a studio, we had a *very* strict method, and it produced a consistent result, but with Corona I feel like I'm guessing once I get to tone mapping.

In Ludvik's, Time to ditch sRGB/Linear as default (?) thread, it is mentioned:

Right now, we perceive linear sRGB as the default, the start line, and we then work with some parameters to bring that sRGB close to photo-realism. We manually have to twist some knobs in order to take a picture, which by default is not realistic to our eyes, and using some controls, turn it into image that our eyes perceive as photorealistic. So why not just skip this process and have renderer(s) by default output same ranges as cameras do. [...] There's no significant reason why renderer should not work the same way. Not by having a dropdown where you can pick numerous response curves, and one of them is called photorealistic, but instead by having it defaulting to a camera, with an option to switch to a very special mode, which will make your output less realistic, but compose-able in post.

This is a very attractive proposition and it speaks to the issue I am referring to. Regardless of whether such an implementation is feasible or not, I'd like to learn how to produce images that are of the quality we expect from the Corona Gallery and I just haven't seen strong tutorials - free or paid - demonstrating such an approach.
« Last Edit: 2020-04-26, 02:30:55 by cjwidd »

2020-03-11, 10:53:21
Reply #1

Jpjapers

  • Active Users
  • **
  • Posts: 1510
    • View Profile
Its not just tonemapping that makes those guys renders look amazing every time. Its attention to detail at every stage and questioning how something actually looks in reality vs what we imagine it looks like. Lacking some minute details is enough to make your brain subconciously question an images' realism so the answer to the question "How do i make it look like [insert artist here]" is usually "Hard work and practice" when you really break it down. I get what you mean though with regard to the consistency of some people images. I think its maybe to do with everyone having a preference of HDRI for a given look, combined with repeating and refining your preferred process over time.

I dont think you need to worry though. Your renders are always awesome!
« Last Edit: 2022-03-06, 22:55:51 by Jpjapers »

2020-03-11, 10:55:43
Reply #2

maru

  • Corona Team
  • Active Users
  • ****
  • Posts: 11507
  • Marcin
    • View Profile
1. "Turn linear sRGB into digital camera product"

Quote
having it defaulting to a camera, with an option to switch to a very special mode, which will make your output less realistic, but compose-able in post.

You can get this by rendering your image at the default settings, then enabling LUT and picking one of the 3 Kim Amland's photographic LUTs. They are captured from real life digital cameras.
Then you can optionally reduce the strength of the LUT to ~0,8 (to your own taste) and play with highlight compression and filmic values (again to your own taste, to reduce highlight burn).


2. "I want to produce images like the masters"

Quote
This is a very attractive proposition and it speaks to the issue I am referring to. Regardless of whether such an implementation is feasible or not, I'd like to learn how to produce images that are of the quality we expect from the Corona Gallery and I just haven't seen strong tutorials - free or paid - demonstrating such an approach.

I am afraid this is not as easy as just following one or two tutorials. Artists have their own secrets. Some of them definitely use Photoshop or other post tools to adjust images in addition to what is done in the VFB, not only by changing the overall tone mapping/coloring, but also locally, e.g. by making some areas more or less noisy, brighter/darker, more clear/blurry, etc.

Marcin Miodek | chaos-corona.com
Chaos Corona Support Team Lead | contact us

2020-03-11, 20:30:32
Reply #3

cjwidd

  • Active Users
  • **
  • Posts: 1069
    • View Profile
    • Artstation
@Jpjapers, thank you so much for your thoughtful and detailed response, and @Maru thank you for your advice - I really appreciate it :)

I think both of your responses more or less encapsulate the truth of the matter, which, as you say @Jpjapers - it boils down to hard work and practice. I want to be clear, I'm not grasping for a magical solution, I'm not pining for a "make this look good button", but I do imagine there are techniques - like using False Color LUTs, etc. - to help narrow the path toward an image that speaks to the quality of the 'masters'.

I would also say that I have **scoured** the forums for many of the relevant tools that have been shared here (e.g. Dubcats camera LUT dumps, fStorm LUTs, Adanmq false colour LUTs and log space filmic LUTs, specular to IOR cube (for materials - Quixel Bridge does this by default now), Dubcats ACES emulation recommendations, etc.) and gathered many high quality (?) HDRIs, including Jørgen Herland and Peter Guthrie.

I assume then, that if I am working with high quality tools, then the onus is on me to get the most from the tools which are already doing their part to help.

I'll take a look at the resources you linked @Jpjapers, they seem promising - thank you!

2020-03-11, 22:36:23
Reply #4

Jpjapers

  • Active Users
  • **
  • Posts: 1510
    • View Profile
@Jpjapers, thank you so much for your thoughtful and detailed response, and @Maru thank you for your advice - I really appreciate it :)

I think both of your responses more or less encapsulate the truth of the matter, which, as you say @Jpjapers - it boils down to hard work and practice. I want to be clear, I'm not grasping for a magical solution, I'm not pining for a "make this look good button", but I do imagine there are techniques - like using False Color LUTs, etc. - to help narrow the path toward an image that speaks to the quality of the 'masters'.

I would also say that I have **scoured** the forums for many of the relevant tools that have been shared here (e.g. Dubcats camera LUT dumps, fStorm LUTs, Adanmq false colour LUTs and log space filmic LUTs, specular to IOR cube (for materials - Quixel Bridge does this by default now), Dubcats ACES emulation recommendations, etc.) and gathered many high quality (?) HDRIs, including Jørgen Herland and Peter Guthrie.

I assume then, that if I am working with high quality tools, then the onus is on me to get the most from the tools which are already doing their part to help.

I'll take a look at the resources you linked @Jpjapers, they seem promising - thank you!

Id still say the biggest thing to always keep in mind is the reality aspect of whatever youre doing. If youre making a lamp light, do some googling and find out the lumens output of a lamp bulb. If you ever have to boost the sun intensity. Something else is wrong.  If you ever have a light set to crazy intensity to make it look realistic, something else is wrong. It sort of helps to just keep asking why when you reach those points. It often helps to set everything back to defaults to rebalance your image if you get to a point where youre making a completely unrealistic guess.

But as i say, your renders look great already! Everyone has a style. I certainly have a method when it comes to lighting but the longer you play around with things the more youre likely to develop your own style instead of trying to emulate others.  I didnt think you were asking for a magic button but it really does just boil down to understanding the software, its relation to real world values and knowing the physically accurate values and methods that you should be using to achieve realism. Yes CG is a playground and we arent bound by physics. But if youre going for realism, you should adhere to realism and the rules which apply in the real world.

As Da Vinci said "Simplicity is the ultimate sophistication".

2020-03-12, 00:49:21
Reply #5

cjwidd

  • Active Users
  • **
  • Posts: 1069
    • View Profile
    • Artstation
Thanks man, I really appreciate that, and I definitely feel you with that DaVinci quote ❤️

It often helps to set everything back to defaults to rebalance your image if you get to a point where youre making a completely unrealistic guess.

Haha, yeah I definitely find myself doing this from time to time - get too far down the rabbit hole and just have to reset.

That is a good point about light intensity, and something I should probably do more to be mindful of. I've mostly been in the habit of working with W/(sr.m^2)(?)

Also, thank you for that false color LUT you shared, I really like it! I've used a couple other ones, but I prefer the color code with that one.

2020-03-12, 10:14:40
Reply #6

Jpjapers

  • Active Users
  • **
  • Posts: 1510
    • View Profile
I've mostly been in the habit of working with W/(sr.m^2)(?)

Ive been questioning which units are best to use for a while. I prefer using Lumens because lighting manufacturers pretty much always list lumens as part of their spec sheets.
In my opinion, lumens work alot better as the light output stays the same regardless of the radius, it just becomes more focussed the smaller it gets whereas W/(sr.m^2) is dependent on the size of the light.

2020-03-12, 19:28:39
Reply #7

Fluss

  • Active Users
  • **
  • Posts: 553
    • View Profile
I've mostly been in the habit of working with W/(sr.m^2)(?)

Ive been questioning which units are best to use for a while. I prefer using Lumens because lighting manufacturers pretty much always list lumens as part of their spec sheets.
In my opinion, lumens work alot better as the light output stays the same regardless of the radius, it just becomes more focussed the smaller it gets whereas W/(sr.m^2) is dependent on the size of the light.

What about directionality? I'm asking myself for a while about that... Light power does not seem to take that parameter into account

2020-03-12, 20:31:21
Reply #8

cjwidd

  • Active Users
  • **
  • Posts: 1069
    • View Profile
    • Artstation
@Jpjapers this is a great example of what I was originally posting about - I had no idea lumens wasn't affected by the light dimensions (?)

2020-03-12, 23:38:37
Reply #9

Jpjapers

  • Active Users
  • **
  • Posts: 1510
    • View Profile
@Jpjapers this is a great example of what I was originally posting about - I had no idea lumens wasn't affected by the light dimensions (?)

Well i had a few years of experience in commercial and retail lighting so i had to have a very basic understanding of real-world info on theoretical light measurements.
Ill try and give a brief and probably misguided rundown and im sure theres things here that are completely misunderstood or mixed up so ill gladly take corrections if ive got it wrong. I know Ondra and Juraj have talked about this in various threads.

First of all Each of the measurements have a specific function and arent just a personal preference thing.

Lumens is a measure of how much total light a luminaire will emit at the source (luminous power) so it isnt dependent on how big the light is because its an absolute.  So literally "How bright is this bulb?". Its required by law on most lighting spec sheets because there are minimum required lux levels across various spaces by law so this info should be available for every light you could need to include in your scene. I like working with this one because you can be sure that regardless of how big the light is, if you know the lumens value, you can be sure its accurate.

Lux The measure of light output as percieved by your eye ( measures exitance at the source and illuminance at the surface). Its a photometric measurement of how lumens are spread across a surface.

So if you run IR and set the intensity units to something like 2000 w/sr.m^2 and change the radius, youll see the scene get brighter. Whereas lumens, if you make the light bigger, the same amount of light output is spread across a larger area like when you focus and defocus a lens on a torch or a headlamp. See below. The torch doesnt get brighter its just more focussed resulting in higher lux but the same lumens.

You could have 1 really small bright light in your house and get a high lux level on your floor. Whereas if you wanted to light a warehouse with those same lights, you would need alot more of them to achieve the same lux level despite the lumens output of the light being the same. 1 lux is 1 lumen per square meter. I tend to use lux for things like mimicking daylight because its easy to measure lux with a light meter.




Watts per steridean per square meter is a radiometric measure of the lights percieved brightness and it depends on the size of your light so for every size increase, you'll get a power increase. It measures radiance which is sort of how much of the emitted power the eye is recieving. Id like to understand this a bit better.

Candelas (cd) measure luminous intensity. That is, how bright is the light at the source. 1 candela for a wax candle flame is accurate. It uses something called the steridean value which is basiccally the cone of the light. I'm not certain on this one. Again id like to know more.

Here's a little diagram of how the measurements relate to each other




What about directionality? I'm asking myself for a while about that... Light power does not seem to take that parameter into account

I wish there was a corona light meter like the old one in mental ray. Then i could actually do some real tests because theres some stuff about Steridean I dont understand (which i think is the technical term for directionality?) and i thought based on the descriptions of the units, the directionality would affect them differently . I asked a fair few years ago about the implementation and maru had some quotes from somewhere. Hopefully one day we might see some sort of implementation when the big list of future bugs/features is running dry. Unfortunately now mental ray is gone you cant really do much.

Quote
You can paste the value which you obtain in W/(sr.m^2) from image pixel (or from your lightmeter in real life) into Corona light intensity and then change light units to Lux. But from what I understand this still requires some guessing and is more of a "workaround" than real light metering. Unfortunately, it looks like a proper implementation of light metering tools will be required for this to work.

Another idea is to use native 3ds Max photometric lights as Corona supports them. You can then switch to Mental Ray and use pseudo-color exposure controls or other tools which are built into 3ds Max to get the desired light intensity, and switch back to Corona again.

Quote
it seems that the easiest way to recreate real light in Corona (or any other renderer probably) would be to find what is the light's intensity in lumen (it should be on the box/package if this is some regular light bulb, or there should be some manual with such info if this is some more advanced/photographic light). If you know value in lumen, then you should be able to simply paste that value into Corona light. If you know the power of light in W, and you know the light's efficiency, then you should also be able to convert this into lumen. Here is an example calculator: http://www.rapidtables.com/calc/light/watt-to-lumen-calculator.htm

There are also some obstacles like light color, or light shape, so it looks like this is always some simplification.

Quote
If you are not after scientific precision, I would suggest you taking a number of reference photos so that you would know the camera exposure values (shutter speed, iso, f number, etc), then recreating these exposure values in Corona and tweaking light source intensity so that the image is visually similar to the reference photos. Unfortunately I don't think there are currently any easy ways of doing this.
« Last Edit: 2020-03-15, 23:27:02 by Jpjapers »

2020-04-22, 10:22:33
Reply #10

cjwidd

  • Active Users
  • **
  • Posts: 1069
    • View Profile
    • Artstation
@Jpjapers at the very least your summary seems like a good shorthand for how to think about these different parameters. You mentioned earlier:

If you ever have a light set to crazy intensity to make it look realistic, something else is wrong.

I was going back and forth with LightMix and realized I was boosting a W/(sr.m^2) light source to >11,000k and then remembered your comment.

Your summary post seemed to indicate that W/(sr.m^2) measurement may include a perceptual component - so there can't really be a 'ground truth' measure for a light of that kind(?)

2020-04-22, 10:59:57
Reply #11

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
1. "Turn linear sRGB into digital camera product"

Quote
having it defaulting to a camera, with an option to switch to a very special mode, which will make your output less realistic, but compose-able in post.

You can get this by rendering your image at the default settings, then enabling LUT and picking one of the 3 Kim Amland's photographic LUTs. They are captured from real life digital cameras.
Then you can optionally reduce the strength of the LUT to ~0,8 (to your own taste) and play with highlight compression and filmic values (again to your own taste, to reduce highlight burn).


2. "I want to produce images like the masters"

Quote
This is a very attractive proposition and it speaks to the issue I am referring to. Regardless of whether such an implementation is feasible or not, I'd like to learn how to produce images that are of the quality we expect from the Corona Gallery and I just haven't seen strong tutorials - free or paid - demonstrating such an approach.

I am afraid this is not as easy as just following one or two tutorials. Artists have their own secrets. Some of them definitely use Photoshop or other post tools to adjust images in addition to what is done in the VFB, not only by changing the overall tone mapping/coloring, but also locally, e.g. by making some areas more or less noisy, brighter/darker, more clear/blurry, etc.




Hey Maru, what do you exactly mean by "rendering with default settings" ? Literally all cam settings untouched, like HC 1, contrast 1, etc ?
Absolutely no highlight compression mostly burns some whites, as we all know. I know that many guys fix this in post as 32 TIF bit or EXR.
In my workflow, for example, I rarely have time for this process... and honestly I don't even like the idea to fiddle around too much in post. Feels fake to me.

However, lately I notice that "correct tone mapping" gets more and more in Corona users´ minds... since, also in my opinion, it is THE key factor for realism.






2020-04-22, 11:11:08
Reply #12

Designerman77

  • Active Users
  • **
  • Posts: 507
    • View Profile
Its not just tonemapping that makes those guys renders look amazing every time. Its attention to detail at every stage and questioning how something actually looks in reality vs what we imagine it looks like. Lacking some minute details is enough to make your brain subconciously question an images' realism so the answer to the question "How do i make it look like [insert artist here]" is usually "Hard work and practice". Personally, I think photorealism boils down to basically one rule.

I agree only partially with that. Of course stuff in your scene should be realistically modeled , textured, etc.

BUT: if your render engine just does not get that sweet spot of light calculation and the correct proportions in color grading, etc... even the most qualitative scene will look fake.

Lately I watched Johannes Lindqvist again. And as he mentions... the basic part of the realism in his work is due to FStorm and its visibly great tone mapping.
Quote Lindqvist: "even a stupid can generate realistic images in FStorm without tweaking on buttons."

And of course, as you mentioned: on top, his modeling and textures are 99% insane, which makes his images look insane.
As for post, I saw that he often does just a minimal tweaking... well, that´s clear with renders that are naturally tone mapped & lit from the start.



I will give another example: look at real photos. No matter how shitty, blurry or badly exposed they are, you will 99% see that it is a natural pic, not a render.
Why? because the proportions in contrasts, colors, etc. are "correct / natural.
« Last Edit: 2020-04-22, 11:36:28 by Designerman77 »

2020-04-23, 04:09:31
Reply #13

lolec

  • Active Users
  • **
  • Posts: 179
    • View Profile
This subject has surfaced over and over since Fstorm seems to produce great results out of the box.  I wanted to try.
I found a good Cornell box taken with a Nikon D70, I believe it was used to showcase Maxwell 1.0 back in the day.

So, this is how it looks when we compare it to Corona default settings:



I agree that it looks a little dull and uninspiring. Maybe even "not realistic" compared to what you get from a camera.

Now, this is what you get with Fstorm in the Default settings.


I used the exact same RGB values, maybe there is something I don't know about Fstorm, so I tweaked the RGB values to match a little bit closer.



That's better... so yea Conclusion: Fstorm produces more photographic images in the default settings.

However....

Look what happens when I enable LUT, choose Kim Amlan 02 and Enable Bloom and glare (it takes 1 second to do all of that)





Well, to me, that looks pretty photographic. I don't think it would be possible to correctly choose Fstorm or Corona in a blind test.

Now, if you like the original Fstorm look. Just increase Saturation to 0.2



Now, I'm not suggesting that just because it is easy it shouldn't be extremely easy.

My proposal is to include a "Magic" button (just like the iphone has) that automatically enables the most photographic settings. Maybe even enables a new pannel with limited options that all produce "ready to use" images.

I do not think you will be able to produce Johanes level images just like that. But I do remember how much my images improved when I discovered LUTs, and how long it took, because I didn't have that information in my brain, there was some friction.

I would also say make the photographic LUT the default, since everyone who doesn't want LUTs, know what they are and how to disable them, but people who WANT LUTs, might not know they exist.

But that's probably too much of a change.

Anyway, hope this helps!
 


2020-04-23, 10:59:21
Reply #14

Nejc Kilar

  • Corona Team
  • Active Users
  • ****
  • Posts: 1101
    • View Profile
    • My personal website
That is a very interesting observation and post Lolec. Thank you for your time :) I am a huge fan of those Dubcat's LUTs. They are amazing.

I suppose I'm wondering, could you share that Cornell box with us? I'd be really interesting in seeing how Octane handles it because they have a fairly specific "look" themselves.

Thanks again for your hard work, man!
« Last Edit: 2020-04-23, 12:44:38 by Nejc Kilar »
Nejc Kilar | chaos-corona.com
Educational Content Creator | contact us