OTOY Forums

I've been using octane for years, but two things that have always left me scratching my head:

Why does "saturate to white" affect red and blue colours 10x more than green?

Why does green render twice as clean noise-wise as purple?

boxfx wrote:Why does "saturate to white" affect red and blue colours 10x more than green?

In the scene, the red and blue objects have emission power 35, but the green object has emission power 2. The green object is therefore much darker than the other two. You can see this by reducing the exposure of the scene.

boxfx wrote:Why does green render twice as clean noise-wise as purple?

I think this is mostly explained by the difference in brightness, but there are a few possible additional contributors. In the scene, the baking texture used for the blue emitter has a lower sample count, and the colors used in the gradient maps are more saturated for the red and blue emitters than for the green emitter. In general, purple light may be noisier because its spectrum consists of two peaks - one at each end of the visible spectrum, rather than the single peak of other hues.

Well then that brings up another question, why do I need far less light to make the green object work compared to the red and blue? If I make the green emission 35 like the other two then it is significantly brighter.

boxfx wrote:"saturate to white"

I'd suggest looking into this OCIO based solution that is far more solid. I've posted results on the same page, using Octane for the rendered images. This is also worth reading.

boxfx wrote:Well then that brings up another question, why do I need far less light to make the green object work compared to the red and blue? If I make the green emission 35 like the other two then it is significantly brighter.

I don't know how much this may contribute to what you're experiencing, but the human eye is much more sensitive to the green segment of the spectrum than it is to the red or blue. In a lot of video codecs, both analog and digital, the red and blue channels have much higher compression, often 2x or more, because we just can't see the noise in them as easily as we can in the green channel. Black and white TV is essentially just the green channel of color TV. (At least in the old analog TV signal. I'm not sure about the current digital broadcast standards. It's been a while since I worked in the broadcast world where we had to be very careful about NTSC color space and such.)

This is one of the reasons you see green chroma-keying a lot more often than the traditional blue - you can get a much cleaner key off of the less noisy green channel. I doubt it's as much of an issue these days, but certainly in the 90s and 2010s, using green instead of blue for digital matte capture made a big difference. For analog film matting, blue causes fewer headaches. Most of the down-sides of using green, like color spill on skin tones, are just incorporated directly into the digital matte creation tools of today.

boxfx wrote:Well then that brings up another question, why do I need far less light to make the green object work compared to the red and blue? If I make the green emission 35 like the other two then it is significantly brighter.

This is what I get if I make the emitter powers equal, make the texture gradient colors equally saturated, disable saturate to white and decrease the exposure so there's no clipping: In that image the brightest red, green and blue parts are all close to 255. The green appears brighter simply because (0, 1, 0) is a much brighter color than (1, 0, 0) or (0, 0, 1) - sRGB green is about 3.4 times as bright as sRGB red, and about 9.9 times as bright as sRGB blue. So it was not wrong for the power to be lower on the green emitter; it's just that 2.0 was too low. 3.5 will make about the same overall brightness as the blue, but this looks a little too dark to me (even after confirming the pixel values in the final image are as expected) - perhaps my monitor isn't properly calibrated, perhaps I have become accustomed to sRGB primaries so much that my brain artificially inflates the brightness of the blue... basically, you will need some tweaking to find whatever looks best.

The saturation of the texture gradient colors is significant because in the scene the left and right objects are a red/blue light source illuminating a red/blue object, but the center object is an almost white light source illuminating a green object, so will be brighter (and less saturated) than if it had been green illuminating green.

The final important thing is that Octane does all of its rendering spectrally, so input RGB values are converted to spectral power distributions for rendering. sRGB red and sRGB blue are more saturated than sRGB green in the sense that their spectral peak is narrower, and so are more susceptible to being darkened when bouncing around in a scene. To alleviate this you may decrease the saturation of reds and blues and/or increase the saturation of greens (in linear sRGB this means negative red/blue components).

Basically, Octane excels at simulating real-world lighting and colors. It works best for physically plausible scenes with realistic colors. The only sense in which a green like (0, 1, 0) lines up with a blue like (0, 0, 1) is in the numerical sense when you write them down like that - the real world (which Octane is trying to mimic) is a lot more complicated.