Ive got a 32bit exr sequence , project is set to 32 but batch comp node still is only 16bitFP , (I can see the warning triangle pop up on node connections from the exr) is there a way to set batch or rather the comp node to 32bit? (image node seems to be doing 32bit for example
I kind of doubt it. Most nodes seem unwilling or unable to do 32 bit–even Action requires that you turn on a Batch pref to do it (so maybe you can use the Action node in lieu of the Comp node?)
Out of curiosity, why do you need 32 bits?
I am prepping a tutorial on substractive CG compositing in flame , there is quiet a bit of a error if I compare it to the pure 32bit render to the image after the whole chain of operations which are all just substract/add and multiplies.
Same operation is clean in nuke because its all 32bit internally , and there is a reason resolve uses 32bit image processing, but yes, I am probably overthinking this, comes with the job I just kinda assumed batch is completely 32bit Float.
Usually I get 16bit FP renders so I never noticed but I did 32bit from blender/cycles this time to get cryptomattes working as blender does not support different bit depth per channel as vray does.
I mean its nothing you can “see” but It just feels weird/bad to do simple multiply and plus operations in 16bitFP , rounding error on import into the node then the multiply is beign done on that rounding error in 16 half float getting me a even bigger error e.t.c
half float really isnt that much precision, its just 10bit for everything between 0 and 1 and then even worse going higher so I much rather have the 23bit of precision of Full float
I did however have some clients that requested us to send 32bit EXRs back which doesnt seem possible then with flame unless you are really careful and limited in the tools you use. (like photoshop).
I will try that with action though good idea.
In my experience, there’s only a couple of nodes that will even take a 32 bit image and work on it without bringing it to 16. Matchbox is one, but I’m fairly certain the output will be 16 bit. I’m not in front of a machine right now so that’d be an interesting test. If you can output a true 32 bit image from a shader then you could just use one of the LogicOp ones to do all your operations.
On a recent project I needed to grab a very small slice from position and depth passes. Every node would result in hardcore stepping so I ended up writing a shader to set the in/out which gave me the same as Nuke.
oh yea I wanted to take a look at getting a fine matte from a position-reference pass , really need 32 bit when dealing with hair and fur… thx for the shader!! LogicOP sounds like a good idea as well.
Practiacally I would probably just stay in Nuke for this CG stufff but normalizing them sounds like a good workaround. Cheers!