I tried restoring the Mr. Elephant test image for Blender 4.0 Cycles (the Specular shader that’s commonly used doesn’t work for some reason, so I had to switch all of those over to an equivalent Principled shader) and rerendering with all the bells and whistles (Path Guiding and Light Trees, though for time constraint reasons I didn’t go with volume rendering)
The two light sources in the fireplace I completely turned off and, instead, made the fire plane itself an emitter.
In an attempt to get more reasonable colors (as the fire texture used already is intensely yellow when it shouldn’t be) I hacked together a material that is temperature based. Unfortunately I got very disparate answers for how hot a fireplace fire actually burns so I’m not sure the colors are “accurate” but in some ways they ought to be closer than the photo texture.
I also took the time to switch all the other lamps over to a blackbody color.
Note how, in the very low exposure version (-6 stops), Filmic turns the fire into what looks like a commercial for McDonalds or something. Strong red and strong yellow and little in between.
The AgX version gives something much more reasonable with reds, oranges, and very reasonably pale yellows that suggest it’s bright.
The bookshelf in the background also greatly benefits from AgX.
Maybe the blackbody emission strength does not match in the shader itself (to strong).
For this reason i think some reference footage with different lightsources and its knowing lumen would be helpfull to compare the lightsources in cycles.(including hdr,blackbody whatever missing emitting sources)
I think it does not have to be a Alexa footage.Even a new I phone that has the raw footage not overexposed could be used for that.
Yeah the blackbody node as is simply puts out, afaik, the normalized color, but gives no intensity information what so ever. It’d be nice if it had a second output for that so you can scale the temperatures as they are supposed to be scaled. Hotter parts burn brigther.
In the above render I did attempt to consider that, though I just did a simple linear interpolation: Hotter parts are linearly brighter than cooler parts. I’m not sure that’s actually correct.
I also didn’t know the absolute brightness. And it’s an incredibly simplistic fire model (literally just a transparent glowing plane) when in reality it would of course be volumetric, and the wood would also glow. (In the scene it was set to pure black, and I left it at that even though it presumably ought to look mostly like the other visible logs with some charring and some glow right where the flames are)
Should note that even this depends on how precisely the camera used decided to map the sensor values to colors, including, say, what whitepoint was chosen, but this is a good test that very clearly demonstrates that AgX actually does better with fire than Filmic does and people simply don’t realize just how skewed colors tend to be.
It’s particularly bad whenever fire and skin tones interact. (Though Wega found, that, to handle that correctly in a renderer, you practically must use spectral rendering because skin models based on three channels simply do not work well with the various hues of flames, especially colder ones that emit more reddish light)
PS: although these are all final images with all sorts of color management applied to them, showing yellow fires, pinkish/red fires, or orange fires, they all share one thing, “the main body of the fire is white” similar to how AgX behaves and contrary to filmic trying to paint the main fire body as yellow.
Comparing AGX and Filmic, the main difference is the greater visual depth of the scene and some color distortion when using AGX. I agree.
It’s a pity that development of Luxcore stopped; it provided even greater depth and did not distort colors. And not many will agree with this
Not there yet…
That "Merge branch ‘main’ into oidn2" means that he updated his local branch "oidn2" with the latest changes from the "main" branch.
What you want is the other way around. Something like: "merged #108314 into main". If that happens, then the stuff will be in the alpha…
Currently, in 3.6, there is an EEVEE setting used to make SSS translucency, relatively to thickness of mesh.
The thickness setting should open ability to make more shading effects relative to thickness, for EEVEE next.
So, currently, it was added to Material Output node, without restriction to a render engine.
But is has no impact on a Cycles or EEVEE render.
I decided to have some fun with Cycles. I made this original render last year of a dragon sculpt used for a school project for making custom scale textures. Back then I didn’t have Cycles X, AgX, nor Principled BSDF 2, so I wanted to see how different the renders would turn out.
Only did a few minor adjustments to the lighting (they were too bright) and subsurface scattering. The left is the original and the right is with all the improvements.
