The Rabbit Hole

« first day (301 days earlier) ← previous day next day → last day (667 days later) »

troy_s

06:23

@oblomov Greets.

14 hours later…

aliasguru

20:39

@troy_s ping(o)

@troy_s Only have a few minutes today unfortunately, but wanted to share two things

@troy_s first, the orange on the frame isn't totally due to the shader. I was mangling it A LOT in compositing to try and match the characteristics of the real world sample next to me

@troy_s it's clear to me now that when switching to filmic views, I need to trash that shader and do a proper one

@troy_s So the problem is not the presets, it's the user (ME) dealing with Blender standard sRGB Transform functions, and forgetting to measure the pixel values (which I did yesterday after all the conversations)

@troy_s an idiot going totally off the rocker in short

@troy_s second, we did the re-lighting of the next project. We had it already nailed before, but decided to do it again using filmic views. To be precise, Joe spent a few hours on it today

@troy_s in the afternoon he approached me and said this:

"I've tried these views, and after being enthusiast in the beginning, I later on thought 'Nah, it's not that different to what I had already'. So I almost wanted to go back to the earlier file. To make a comparison, I loaded the 'old' image. And it blew me away how much better the one using filmic presets is lit"

@troy_s I've had a look myself then, and was blown away myself. I said "If you show this to someone who doesn't know, he's going to believe you've used a different render engine"

@troy_s also we tested adding glow to it. BOOM! Finally glow happened where it should happen!

@troy_s finally, we compared the rendering we made to the studio photo a pro shot of the same bike

@troy_s against the rendering, the photo sucks :)

@troy_s we spent a good 20 minutes exploring highlights we never saw before and needed to fake, details in dark areas we usually got as black, light falloffs that never came out like this before, and other things.

@troy_s In short, we'll be using them

troy_s

20:54

@aliasguru Gah missed you here!!!

@aliasguru Reading back. Hold.

@aliasguru It's such a difficult topic to discuss because you need to have the scene lit accordingly, and then be looking for the details in the bounces etc.

@aliasguru Example. This is an HDRI, with no post processing, while we were mucking with paint experiments briefly...

What you are seeing there is the actual HDRI in the background. No low grade display referred plate, equal energy levels with the objects in the scene, etc.

aliasguru

good stuff

troy_s

What is interesting in that super quick (as in literally drop in object, set and go) is that the energy levels are all operating precisely in the scene.

The thing that is doing the heavy lifting is the view transform.

But what is really interesting is how that ridiculously simple energy alignment weaves those CGI elements into the scene so seamlessly.

aliasguru

What's so amazing about the filmic transforms is when you start having a lot of entirely different materials in the scene. They still play so well together

A bit like magic

troy_s

@aliasguru Yes. The albedo is a tricky slippery thing. I spoke with a render expert type, the type that really knows the math, and he said that paint can be simulated very well (especially metal flake) if you simply feed in the base as a metallic glossy, then run it through the glaze.

He said that if you did a close up crawl over the surface that you'd need to add a perlin noise via a ramp as a texture to see metallic flake particles, which is relevant if you are dealing with any metallic fleck paint structures.

At distance, the simple metallic shader to glaze is easy.

@aliasguru Note that when you add a glare / glow / lens effect, you'll also see an order of a magnitude of differnce.

When dealing with a 'faked down to display referred' lighting setup, your dynamic range can't behave as it does in a camera or bloom as such (think rear net for example)

When you have that proper light level, you can set the bloom / flare to only operate on say, the highest of values up near say 600+ (just getting near the edge of the desaturation shaper, which is 10 stops over middle grey, up to 2^10*0.18)

The net result is that the myriad of hot pings and reflections (which I'd stress I'd try to "shoot" and keep data, then grade to black and white) now bloom individually, which adds a very subtle bit of detail to the scene.

Compared against say, the 1.0 approach.

aliasguru

very excited to run more tests tomorrow!

troy_s

21:04

Anyways, hope you found that grade sort of close. That's literally a super simple shader, shot to retain detail, then graded to push the colours into the KTM range.

So Joe feels confident with it now?

aliasguru

He adapts quickly :)

the results were too convincing to not proceed with it

troy_s

As I said, it's a shorter path with folks with a good deal of experience.

aliasguru

thanks for sharing them!

talk to you again tomorrow (should have more time then)!

troy_s

@aliasguru Ciao bro.

@aliasguru Would love some private temp renders if you fire them my way.

@aliasguru Also note, I suspect there are a good number of shaders that aren't abiding by real-world values that might need tweaking. Metals for example. Definitely try Jason's PBR until the Disney one lands.

2 hours later…