Hello everyone.
i have been using ACES with maya since some times now and decided to test it on Blender.
I have been hearing and seeing a lot of negative comment about colormanagement in Blender so i decided to run a small experiment to see how different could my render be, compare to the Arnold ones.
TECHNICAL ASPECT:
(to prove the consistency between the 2pics)
The scene: traditional cornell box.
2 cube - white material - 0.5 specular weight - 0.5 roughness - 1 diffuse weight
1 sphere -metal material - 0 roughness
3 planes - textured
1 Camera (focal length 35mm)
I had to add a plane with emmissive shader for blender to simulate the light - i disabled all ray visibility except camera
Material:
Diff Weight-1 SpecWeight-0.5 SpecRough-0.5
White material : R.G.B-0.570 (non colormanaged) // R.G.B-0.701 (colormanaged)
Green mat: R-0 G-1 B-0 (non colormanaged)
Red mat: R-1 G-0 B-0 (non colormanaged)
Texture mats: DiffWeight-1 SpecWeight-0
Textures:
Colorcharts from https://github.com/colour-science/colour-nuke
Left chart is .exr : Utility linear-sRGB
Right chart is .exr: ACEScg
Middle pic is BobRoss .jpg : Utility-rec709-camera (looks closer to the pic but this probably not the good idt)
Light
I didn’t really know how to convert Arnold exposure value to Blender power value for the light. Using wikipedia (https://en.wikipedia.org/wiki/Exposure_value) i managed to find 15.221 which seems to work.
But i guess this is not the good value.
Arnold: arealight
color: R.G.B-1 Intensity: 1 Exposure 16
Other settings are default
Cycles: arealight
color: R.G.B-1 Power: 1W Strength(node): 15.221
Render settings:
-Arnold-
Sampling: 4-2-2-2-2-2
Clamping: Indirect-250
Raydepth: Total-10/Diff-5/Spec-5/Transmis-12/Volume-0/Transparency-8
-Cycles-
CPU -PathTracing
Samples: 512
Lightpath: Total-10/Diff-8/Glossy-8/Transparency-8/Transmis-12/Volume-0
No refle/refra caustics
ViewTransform: sRGB
Image converted in nuke to png.
IDT: Acescg
Viewtransform: sRGB
RENDER
ARNOLD
CYCLES
Also did a second arnold version with change in red and green color. Can’t remember what I changed , something with color managed color picker and i had negative value.
The result of this render is much closer to the cycles one so that’s interesting.
CONCLUSION:
So even if some render difference could be explained but the difficulty of choosing a color and the conversion of the light we still have a big difference on the BobRoss picture, the only 8bit texture.
In fact I don’t have enough knowledge to interpret the results of this experiment so that’s why I am posting here.
All i can say is that I try my best too keep the scene in both render similar but we still have difference.
And theoretically the Arnold render is the " good one" as Blender color management might be broken.