Adjustment layers bit depth

[Juan]

Hello, Is there a way to know what adjustment layers work in fully 32-bit and which one will "convert down" the bit depth?

[Martin Huber]

The easiest way is to ask about the ones that interest you.

[Juan]

I would like to have a chart with each adjusment layer bit depth support. I created one with what I believe is the bit depth support of each one.

bitdepth.jpg

Juan

[shijan]

Levels works in true 32 bit if don't touch Gamma slider. (As it was explained by Martin Huber: Gamma is a exponentiation and therefore need a base larger than or equal to 0. Therefore some values are clipped before exponentiation.)

[shijan]

Guess 32-bit float support is easy to check if apply filter to some HDR EXR CG image. You need real 32-bit float sample image without build-in tone mapping. Some sample EXR test image from AMD ProRender should be great. If filter don't support 32 bit you will see some clipping artifacts. For additional check you can apply Optimize HDR filter on top, Enable checkbox "Automatic" and see if it tone maps image correctly. If output is 16 bit, Optimize HDR will not tone map filtered image in expected way.

For example from my quick test seems HUE Editor don't support 32-bit float in HIS mode, but works well in HSV and RGB modes.
Lab mode HUE Editor is strange. It generates a lot of artifacts if use Apple CMS, but looks ok with very small amount of artifacts if use Little CMS.

[russellcottrell]

Scripting can generate a gradient of 32-bit colors that are one bit apart (I think); but no application I know of has a color picker that actually reads 32-bit colors (increment would be 1/4294967296 or 0.000000000232831), and I do not know how to get the color of a pixel with scripting (in PhotoLine, that is). If it were possible, adjustments could be applied to a gradient and the colors read to see if they were still one bit apart.

[shijan]

No need any scripting. Color Editor allow to set color values in float format that is less than 0% and more that 100%.
So instead of looking some abstract and not too scientific 32-bit 3D render example, you can create black and white gradient and type in Color Editor for black color something like -200%, and for white color 200%. Instead of typing it is also possible to use middle mouse button drag to set values beyond 0-100%.

In 32-bit if you apply Optimize HDR filter to this gradient and Enable checkbox "Automatic" - it will tone map negative values back to 0% and 100% and gradient will change it's look.



In 16-bit If you apply Optimize HDR filter and Enable checkbox "Automatic" - gradient will remain unchanged.

[Martin Huber]

I would like to have a chart with each adjusment layer bit depth support.

The answer is not that simple.
Some functions internally use color management (CM) and depending on the color management system (CMS) used that may or may not support 32 bit (float). Windows CMS only uses 16 bit, LittleCMS and Apple CMS also support float.
Other functions have to clip float values to a [0;1] range for mathematical reasons.

So here are the adjustments that currently support float. Some are listed as "uses CM". These are the ones that always use CM. Other adjustments might use CM when you set the working mode to "Lab".
- Levels
- Optimize HDR
- Hue/Saturation (not in Colorize mode)
- Hue Editor
- Vibrance
- Selective Color Correction
- Threshold (but that always just generates 0 or 1)
- Color to Transparency
- Color Lookup (16 bit with 1D LUTs, float with 3D LUTs (uses CM))
- Match Colors (uses CM)
- Channel Mixer
- Gray Mixer
- Unsharp Masking
- Sharpen
- Adaptive Sharpen
- Gaussian Blur
- Soften
- Adaptive Soften
- Motion Blur
- Variable Blur
- Wipe Effect
- Exposure
- Sponge
- High Pass
- Minimum
- Maximum
- Perturbation
- Relief
- Light/Shadow
- Color Temperature (current version is 16 bit, next version will be 32 bit (uses CM))
- Chromatic Aberration
- Denoise

[Juan]

Thank you all for your replies

On my tests I noticed that these adjustment layers clips the values:
- Hue/Saturation
- Selective Color Correction
- Gray Mixer

The way I made a test is rendering an image in blender and saving it as full float color depth. Afterwards, I added a color position on a bright area in the picture and started checking what adjustment layer turns the picker value to 255 or lower.

color position.JPG

If anyone wants to give it a try, here is the original exr file.
https://www.transfernow.net/dl/20220917pgTj96Md

[shijan]

Nice test with Color Picker!
I can confirm. From Gradient test seems Hue/Saturation, Selective Color Correction and Gray Mixer are not 32-bit.
Hue/Saturation by default works in HIS mode, so same as Hue Editor in HIS mode it don't support 32-bit float.
HIS is very nice mode because it changes saturation without brightness shift.

[Martin Huber]

On my tests I noticed that these adjustment layers clips the values:
- Hue/Saturation
- Selective Color Correction
- Gray Mixer

Hue/Saturation uses HIS and that is restricted to [0;1].
Selective Color Correction also clips its result to [0;1].
Gray Mixer uses Lab and is therefore restricted to the properties of the CM.

Martin

[Martin Huber]

I can confirm. From Gradient test seems Hue/Saturation, Selective Color Correction and Gray Mixer are not 32-bit.

They are 32 bits, but are limited to a certain range of values for inherent reasons.

Martin