"L'Appartement" (Filmic_blender + Denoise)

(SunBurn) #1

Hi everybody, wishes for a creative new year.

Here’s a new test with my apartment scene (first post this year).

The last couple of months I’ve been playing with filmic-blender and I must admit I’m amazed.
The difference in dynamic range is huge. For the first time I have realistic and powerful highlights.

I enjoyed them so much that I’ve even removed all the carpets from my original model!

I’m not an expert trying to explaining here what filmic-blender is all about, but as far as I understand it, it should be the default in blender from now on.
I was kind of disappointed hearing at the latest BF pod-cast people taking about the subject so lightly and with a tone of irony.

The biggest problem that I had so far with filmic-blender is that you can not use the usual tricks that we all use when we light our interior scenes. Tricks like AO, clamping etc. because that way we’ll counter-effect the awesomeness of filmic.

This results of course in pure path-tracing noise, that is more than prominent in almost every interior closed scene rendered with Cycles (non-hacked).

So I decided to go with the latest de-noising build of Lucas and I must admit it (almost) works. There a couple of problems here and there (some textures are blurred too much and I can also spot some splotches)but I think we’re in the right path.

I consider this test a small bite of Blender’s future.
Better color management, de-noise or some kind of adaptiveness in rendering and we’ll have an even more powerful tool in our hands.

Many many thanks to troy_s and of course the invaluable lukasstockner97.


(MZGarmi) #2

Could you create an image where we can see all versions so we can compare?

(SunBurn) #3

I’ve forgot to mention that there’s almost no post-production just a slight color balance with blender’s compositor.
Here are a few more images.

(SunBurn) #4

Technically I could but I can find any use in it since as I’ve said before it’s not a straight comparison, as I’m changing a couple of factors every time.
It’s more of a work-flow comparison. Going from a basic Idea to the finished deliverable. I’m always testing under the prism of a professional which means tight dead-lines, more than one versions, last minute changes etc. and then I post my thoughts.

Here are all my other tests so you don’t have to search
Cycles, Yafaray, Luxrender

(MZGarmi) #5

Thanks, that is good enough for me. :smiley:

(LazyVirus) #6

Yes and correctly desaturated, a linear behaviour of light and finally a bye bye to the ugly posterization.

Well kinda agree, but I wouldn’t be so dramatic. It’s a friendly chat where no one takes it very serious.
And for some reason I’ve got the feeling that almost everyone is scared to say something wrong because of troy’s dick attitude vs imprecisions. But it’s evident he wants people talk about filmic and when it happens he goes wet and puts aside some of the imprecisions.

All cheap dirty tricks that shouldn’t be used anyway. I bless every “new toy” that prevents people from using them.

I gave up hoping in some smart secondary GI cache solver (ala corona, vray etc) so honestly whatever will come is good, even adaptive sampling, gradient domain pt etc… I just fear they could be a big waste of time and effort. That’s all.
As for the denoiser, don’t take it as a saviour because its contribution on the image should be little to be acceptable. In your image, denoise contribution is huge and destructive.

As for your image, in what point have you calculated the correct exposure?
3rd pic looks kinda “generally” overexposed to me.

(Razorblade) #7

hm… the colors do look a bit washed out, but then that probably is the filmic effect to achieve a more ‘daunting’ day light, not so vivid.
But it can be like that with normal photography i sometimes put in a yellow filter just to make it more pleasurable, then people think of a more sunny happier time (yeah we had a great BBQ see the weather was OK, instead of well the BBQ was OK as it was a bit cold that day)
So hm I’m not sure what to think of it, i hope you post some more renders, to us a impression

I’m not sure how this works (does it replace the emulation of photography film types in Blender ?)
Been looking at it but not yet installed it, since i red a few things get broken and i wasn’t sure if used those :wink:
I hope something like this eventually gets in, maybe with lightgroups in compositor.
I’m not sure though if the cycles team is looking at it, or that it will stay a additional external addon.

(SunBurn) #8


All cheap dirty tricks that shouldn’t be used anyway

I couldn’t agree more

I gave up hoping in some smart secondary GI cache solver (ala corona, vray etc)

I’m still waiting since there’s no other real solution. I would gladly have moved to Corona (no Linux version), Vray (no Live preview last time I’ve checked) the very impressive Fstorm (only windoze and Maxx).

As for your image, in what point have you calculated the correct exposure?

I increased my exposure till the point I got clipping (white color in False Colour Look, then I set it a little Lower just to have a little headroom in the post production. Then in post I used s Color Balance switched to OPS mode (as troy_s suggested in a post) and tweaked it a little more. I remove the bluish tint that came from the hdri that I’ve used and I played a little with Power slider.
I don’t know maybe I did something totally wrong. If you have any advice please comment.


the colors do look a bit washed out

I didn’t messed up with saturation at all so I don’t know what’s causing this. I can tell if this is normal or we’re so used to “wrong” images that it looks odd?

(MZGarmi) #9

So you render your images 20 hours?

(LazyVirus) #10

It’s not an advice, I just can tell you what I find useful as a starting point to tweak lights’ energy with filmic.

I would try to use a photographic approach using a mid grey card in the scene (or Macbeth ColorChecker for color balance).
If I’m not wrong 0.18 grey should be the filmic_blender mid grey.

I would put the card facing the camera in the position where my subject must be correctly exposed.
I would start from a basic settings of light sources based on my eye, rendering with border (for fast test) the part with the card, and check color values in the Image editor with CTRL+LMB.

Left values are the Scene Referred values, while the right values are Display referred values.
(correct exposure would give you around 0.18 - 0.5).

At this point to have a quick idea of how much tweak the light energy to have a correct exposure in the point where you calculate exposure (card), a useful rough trick is to use the “False color look” + exposure control just for test. Dunno if it’s really a correct workflow but I find it useful.
By tweaking the exposure control until the card becomes grey, you’ll get the EV needed in + or - to have the card correctly exposed.

Considering that +1EV (-1EV) means x2 (x0.5) amount of light (linear), you can do the math and know how much tweak the light values.

For example if you started with energy of 10, and the card in false colour becomes grey with +3.1 EV, the new value of light energy must be 10*(2^3.1).
(And obviously you’ll have to set the exposure control to 0 again and don’t touch it along with gamma).

I repeat that I don’t know if it’s a correct way to set light energy, but it’s useful to me and time saving.

Yeah I know it’s longer to write than to put it in practice.
Since I wrote this post in multiple times and I didn’t checked it, I apologize if something weird is written, cutout, missing or whatever.

Said that in the filmic view you should set one of the contrast preset as a starting point, I guess the washed feeling could be related to the scene itself with low saturation textures.

You sure you’ve read what I’ve said?

(Razorblade) #11


I didn’t messed up with saturation at all so I don’t know what’s causing this. I can tell if this is normal or we’re so used to “wrong” images that it looks odd?

Maybe after you do some more renders, and compare it with the room you live in, then wonder if it should have more vivid colors.
If i look around at my home, i see a bit more contrast and more saturation even at a non sunny winter day, maybe add a warmer sun light ?.

(troy_s) #12

Lovely work. Keep crushing it.

I hope you don’t mind that I tweeted out the work with the #filmic_blender hashtag for other folks to see. It even picked up a retweet from Ton.

Props also to @LazyVirus for a pretty solid method and explanation to grab middle grey via the False Colour. Thanks also for helping out on the support front; it is folks like you that help to strengthen a culture. The current version of Filmic had quite a focus on getting the “aesthetic” Looks to map scene referred middle grey (pegged at 0.18) to display referred 0.5 consistently across all of the looks. This makes it useful for folks trying to mix and match other footage, ratios, etc. as well as having a firm anchor for someone looking to grade in another application.

Regarding being a dick, I probably agree; I have all the patience, time, and respect for an imager asking that seemingly “stupid” question that no one else dares to out of fear. I may be terse towards other things, and possibly too blunt in others.

I really could care less about Filmic and whether or not someone uses it. I care far more when folks pick up the torch, crush out amazing work, and engage with the colour concepts. If Filmic contributes to work or learning in some way, then it has succeeded.

Hoping to get the wide gamut rendering and HDR curves out soon.

Thanks again…

(SunBurn) #13

Thank you for sharing you method. I gave it a quick test and it works perfectly.


compare it with the room you live in, then wonder if it should have more vivid colors.

I try to compare with similar photos since human vision/ perception can play tricks :spin:, but I’ll keep that in mind for my next test.

Thank you for sharing my test on twitter and thank you for helping me learn and understand the subject deeper.

(Lane) #14

Excellent work Sunburn, its excellent to see the difference on different render engine too.

(MZGarmi) #15

I don’t understand this technique with 0.18 grey, can someone explain it for dummies?

(troy_s) #16

In the scene referred domain, depending on lighting, the values will range from some low value to infinity. This is very similar to looking outside of your window when taking a photograph; there are merely a range of values, with no magic numbers like “1.0”.

When you attempt to capture the scene with a camera, either physical or virtual, the camera will have a range of values that it can “record”. There is a minimum and a maximum value that the camera will “see”, with values outside of this zone being ignored. We could call the entire range of values that the camera records a “dynamic range”.

To a standard observer though, the range of values is uniquely nonlinear. That is, the range of energy values are bent and distorted. Where one might expect the “middle-ish” value to be around halfway between the minimum and maximum value, it is actually much lower. Historically, photographers used a grey card to locate roughly this middle value. The value was roughly 18% reflectance, which after a standard observer adapts to the lighting, represents roughly the middle 50% of the perceptual field in terms of intensity.

As a convention, 0.18 has been carried on and used as a middle grey point in CGI and digital imaging. Most transforms will typically map somewhere around that value to around the middle of the display / output range. In the case of Filmic 0.18 scene referred (zero to infinity theoretically, in Filmic it ranges from 6.5 stops over middle grey to 15 stops) is mapped precisely to the 50% mark of the display referred output (minimum to maximum, typically 0.0 to 1.0)[1].

So what does this mean and how can it aid an imager? 0.18 can be lit to and make it easy to know where values land in your image. That is, if you wanted a pillow to represent the middle range of exposure in the output, you could place a diffuse grey card of 0.18 albedo in the scene next to the pillow, flip to the False Colour view, and adjust lights until the grey card lands in the grey band range. This corresponds to the pillow being lit for middle grey, and its value will land precisely at 0.5 display referred. Likewise, the identical trick can be used if you want an object to be darker or lighter, say three or four stops above or below the middle grey point.

[1] The savvy reader will realize that CGI raytracers generate inherently visually linear energy output. They likely also may know that LCD displays are also inherently linear. Why then, doesn’t the output get dumped directly, in a somewhat (plus or minus aesthetic bending) linear form for the perceptual systems to interpret? That’s probably a discussion for another day, but needless to say, it is largely historic, and adds another layer of complications unrelated to the perceptual system. It is frequently confused as such however.

(mareck) #17

That’s how I recognize someone who has better knowledge than me, by his way of being, and if I don’t understand something he tells me, I know I must do efforts and researches to reach the point where I can exit the mist.
Filmic is perhaps The better tool that breaks the gap between Cycles and other well known renderers, it’s not the only, there is more to improve, but the light is such an important topic for realism. Troy make half the job, the other half is our knowledge so let’s do our job.

And, well done @SunBurn, really good images.

For the ones who want to grow up there knowledge: http://colour-science.org/posts/the-importance-of-terminology-and-srgb-uncertainty/

(MZGarmi) #18

Thank you man, now i understand it. :yes:
But one question. 0.18 albedo means color with R 0.18 G 0.18 B 0.18 value?

(Lsscpp) #19

yes. I just put a plane/cube with R 0.18 G 0.18 B 0.18 diffuse shader

(LazyVirus) #20

Undeserved :o
All credits goes to you.

I have very basic knowledge about color management, but I’ve read ages ago that it’s phisically impossible for CRT monitors to get a linear input from pc and output correctly that signal without loss. This explains the gamma correction of softwares and the standard transforms set when CRT ruled. I can’t figure out how much time it’ll take to switch to linear signals as standards.