using BMCC VFX I/O LUTs with Fusion

This is more a question for Blackmagic, but maybe someone's smarter than me

I followed Stefan Ihringers tutorial on how to set up Fusion for linear compositing, and wanted to slap some BMCC material (Prores, BMDFilm) into the comp.
Now with the Gamut-Node I can't set the material to the appropriate colourspace; I actually didn't thought I could, but not even Cineon (which I like to use in Nuke sometimes) was an option so I thought I could just use the VFX I/O LUTs that DaVinci Resolve provides.

However, I can't load the LUTs with Fusion, "error reading that file". When I try to adjust the header of the .cube I can use it, but then I get very strange values which are not even close to the ones listed in the LUT.

Here an example applied to a black-white gradiant:



The values in the colored areas are insanely high.

Would be really nice though if we could use BMDs LUTs in their VFX application, no?

Cineon conversion node is in the film tools folder called 'cineon log'

I wouldn't be using Cineon as an effect however....that's usually the raw capture gamma people will work with for some added dynamic range, but it is always corrected back to a 2.2 or DCI or other gamma before final output. Unless BMDFilm is the same as cineon and you are trying that way...

How are you loading this LUT...as a file LUT or a viewer LUT.

Most LUTs can't properly map superbrights, you may try running the gradient through the cineon log node, as that will normalize it into LOG space....the apply your file LUT after

If you linearized the gradient that would probably push the whites well above 1.0

A Viewer LUT wouldn't make much sense in this workflow I guess?

The approach is similar what you would do in Nuke. Worl in linear, every footage is linearized correctly, else it won't work properly. I don't see how this would add dynamic range in any way, it's only there to linearize the footage correctly?
Cineon isn't perfect fit but surely works better than sRGB

Of course the LUT introduces values above 1, that's what it's supposed to do. But the values are insane/much higher then they should be (reading from the .cube) and obviously they don't apply to all channels equally, what they should. The Gradient is only meant for showing how the LUT doesn't work out properly (that is AFTER altering the header, by default the LUT can't be read at all)

Can you read the LUTs manually? What values does it use for the high end?

A Viewer LUT wouldn't make much sense in this workflow I guess?

The approach is similar what you would do in Nuke. Worl in linear, every footage is linearized correctly, else it won't work properly. I don't see how this would add dynamic range in any way, it's only there to linearize the footage correctly?
Cineon isn't perfect fit but surely works better than sRGB

Sorry was not sure what you are trying to do. The post was a little confusing

Are you trying to linearize BMD footage?

Did you read where I told you cineon conversion is available in the film tools folder....not the gamut tool?

Also never suggested cineon would add dynamic range....and its there for more then just linearizing footage properly....

Hi Mike,

sorry for the confusion, i thought it was clear but obviously wasn't, haha I guess I also misunderstood your comment on the dynamic range then
I will try the Cineon Conversion when I get home tomorrow evening (not home this evening, and only Macs around me till then, haha)
Yes, I try to linearize BMD footage. Even if CIneon Conversion works fine, it would still be a lot better to be able to use their appropriate LUTs; I mean these LUTs are there, but not usable in Fusion

Hi Chad,

opening the LUT in a Texteditor, the the highest end value is 5.7661304310 (lowest end -0.0071215555)
Header reads the following:

Code: Select all

# DaVinci Resolve Cube (1D shaper LUT).

LUT_1D_SIZE 4096
LUT_1D_INPUT_RANGE 0.0000000000 1.0000000000

Not sure if I may attach the LUT, it's free within DaVinci but no idea if it's cool to upload

Resolve luts are log to rec709, not log to linear light.

If your applying the LUT incorrectly, or the wrong one you could get this overbright issue.

I still suspect this is the issue, even though it would not import properly....there's not much that can go wrong in the LUT as its just a table.

If you want to linearize BMD correctly....you may need to find the proper conversion LUT to go from BMDFilm to linear....often this may be the exact same as cineon(as it is with redLOG).....then you will still need a view LUT to see something worthwhile

But if the footage is NOT in a LOG type space, the sRGB method that tilt shows will get you quite close and with a convinient viewer LUT.

Resolve luts are log to rec709, not log to linear light.

Just for Clarification, I'm not talking about the normal "Blackmagic Cinema Camera Film to Rec709.cube" LUT, but the "BMDFilm to Linear.cube" in the "VFX IO" Folder. I do actually think that one is LOG to Linear
And now that BMD owns Fusion, I think it would be useful if the VFX LUTs worked in their VFX app

If your applying the LUT incorrectly, or the wrong one you could get this overbright issue.

I still suspect this is the issue, even though it would not import properly....there's not much that can go wrong in the LUT as its just a table.

Well I can't apply the LUT at all. That's the original issue the LUT won't load with the File LUT node. It's only readable when changing the header inside the .cube, and then it gives the strange values for overbrights.
The LUT applies the same values to all color-channels, so a shift like in the screenshot shouldn't happen at all; yet the values of those areas aren't the ones from the LUT either.

Maybe the LUT is made "incorrect" for Fusion to understand, I have no idea, that's why I was asking I guess you can't do much wrong there, I just load up the file

often this may be the exact same as cineon(as it is with redLOG).....then you will still need a view LUT to see something worthwhile

It's definately not the same as Cineon, sadly Cineon goes up the 13, not only 5 in linear values. I mean it will be a usable workaround but somehow I feel like the right conversion would do us all a favor
For Preview, I use the approach of Stefan Ihringer from his Linear-Working tutorial

If you use AE, there is that free magic bullet LUT buddy that can do some conversions...
http://www.redgiant.com/products/magic- ... lut-buddy/

rather then hacking the header...worth a try

Good idea with LUT Buddy
Unfortunately loading the LUT into LUT Buddy crashes AE completely.

It baffles me a bit that BMD seems to have no interest to make their LUTs usable anywhere except for DaVinci.
I tried to apply the LUT inside DaVinci and then re-save it as .cube, only gives jibberish.
LUT Buddy crashes.
Fusion refuses to load.
Nuke loads with altering the header, but there's not much use of it there (can't use it for importing footage e.g.)
Briz LUT Converter (I tried that some weeks ago) can't also solve that.

I reported that to BMD months ago (back then regarding Nuke), and I know for a fact that The Foundry tried to contact them too, to implement the LUT as a proper linearization.

It seems like BMD doesn't really care for their cams to be properly integrated into a VFX workflow, that baffles me a lot especially after the purchase of eyeon. Sigh.


Sorry for the rant, I just feel a bit disappointed.
The Cineon Converter works, better than nothing for now. Thanks Mike

OK this should work....

1.Use the Fusion LUT Cube creator, create the cube image....render it out as an exr or tiff float 32 image

2.Import into DaVinci, apply the VFXio LUT there onto the cube creator image...render out a float 32 image (no scaling..important)

3.Load the rendered image into Fusion....use the LUT Cube analyser back in Fusion on the image..save out your analysed LUT (no rendering needed....just write file)

Unless I am missing something, this should definitely work

These tools are in the LUT folder in Fusion

I would probably use the rectangular size, and maybe up it to 128...if it is an unexpected size the analyser wont accept it so keep the image dimensions the same in Davinci as Fusion. If you don't choose rectangle in the creator settings it makes a very long thin image...not sure if Davinci would like it

This way you can translate any Davinci LUT to Fusion (and others)

CUBE.jpg (84.35 KiB) Viewed 8559 times

as a sanity check...


take the un-rendered cube image (no davinci LUT applied) load into fusion...apply the LUT created by the analyser using the file LUT node

import the rendered image from davinci and compare the 2 using a difference merge

There will be some difference for sure...but you may not be able to see it without hitting normalize on the viewer...hover your mouse over the bright spot and see the difference.
Gives you a sense of the accuracy...

Hi Mike,

I tried something similar with Nuke yesterday, thank you for that input it's a great idea.
One problem of this method though is that you get 3D values, yet it's only a 1D LUT... not a biggie but still

Actually this seems to work quite well with the BMDFilm->Linear LUT. So we actually can linearize the footage correctly.

Now the more or less funny aspect is that we are linear, but can't go back. Means, this method doesn't work at all for Linear->BMDFilm.
So we are stuck in linear and can't render it out appropriatly. Oh well

I am surprised the LUT's are not available somewhere....but I think you can still accomplish this if you need to...

If you can load the same frame of a BMDFilm image and a linear version you can run them through the color curves tool and 'match curve'...then instance this tool and feed in the LUT CUBE creator, which will apply the same change...then pipe that result into the LUT Cube analyser. That leaves you with a Linear-BMDFilm LUT.

Obviously the sample frame will have a huge impact on accuracy...it would be best to use a color chart or something with a lot of info to draw from.

When you apply this LUT (LINEAR-BMDFilm) is the situation where you can run into the overbrights...where a LUT applied to linear footage will not work properly and you have to normalize by converting it to LOG using the Cineon LOG node....and of course you would then need to create a LOG-BMDFilm LUT. Then a viewer LUT is also a problem....because the transform from liner-log-BMDFilm should happen at the end of the flow before the saver...but you still need a viewer LUT to work with the footage.

The image below is the simple case of matching the curves to create a Linear- BMDFilm LUT.
I think this is the proper workflow, but those smarter then me might have some tweaks to add to it.

I am just hacking my way through this by learning from years of mistakes...

You could probably figure out the LOG-BMDFilm LUT workflow from here.

compare.jpg (54.09 KiB) Viewed 8495 times

This is why the Linear workflow with an sRGB viewer LUT is so much simpler....

Unless you are really in love with a particular custom gamma, its far easier to just work in sRGB and color grade to taste after the fact.
Going from linear to a custom gamma requires a few more steps.

This is why the Linear workflow with an sRGB viewer LUT is so much simpler....

Unless you are really in love with a particular custom gamma, its far easier to just work in sRGB and color grade to taste after the fact.
Going from linear to a custom gamma requires a few more steps.

sRGB isn't a large enough gamut for general purpose use though.

Also, what extra steps are involved? Seems to be the same to me.

- Chad

i am assuming Stephan's sRGB view shader maps the overbrights better then standard LUTs...so I may be incorrect. I never actually tested.

But say in a red R3D workflow if you wanted to maintain the Redgamma....you would bring in linear and do your CG....but to reapply the RedGamma, you first have to run LIN-LOG....then apply a LOG-RedGamma LUT.

No problem there but, the veiwer LUT while you are still working in linear is problematic because a simple LIN-RedGamma LUT will not map the brights properly. That means needing a viewer LUT that is a LIN-LOG-Redgamma chain...

Like I said this is an assumption that the sRGB and rec709 veiwshaders Stephan provides are more accurate then simple LUTs. In that case it is just a simple veiwer LUT and a gamut tool at the end.

If you had a HDR LUT that could map the overbrights it would be much simpler.

Does that make sense? I know you have far more knowledge about this then me....I have went through this all before trying to maintain RedGamma in linear workflows, and this was the best I could come up with.

Well, the lin-log-gamma can be condensed from 2 operations to 1, you can even do the display LUT in there too. Just loading one 3D LUT in the viewer should be enough.

Idea being that if you are doing this on more than one or two shots you'd want to have a single LUT for the viewshader anyway.

Your right of course....

I just made a macro that does the log conversion and applies the LUT and loaded it into the LUT directory.

Sounded more intimidating then it was.

Stephan had suggested this to me earlier but I never got around to trying it.

My sRGB view shader simply applies a 1D gamma curve (see here). I haven't tested this but I'm assuming that this is the curve that is applied by Fusion's Gamut tool if "Add Gamma" is enabled. It should also match the OCIO LUT or Nuke's sRGB setting.

It doesn't do any gamut transformations. So this means that it's only correct if your linear footage is in an sRGB gamut but to be honest... the main idea of that viewer LUT is to have some kind of quick preview for an easy linear workflow.

Your right of course....

I just made a macro that does the log conversion and applies the LUT and loaded it into the LUT directory.

Sounded more intimidating then it was.

Stephan had suggested this to me earlier but I never got around to trying it.

Making that macro into a LUT would be faster, because it would run on the GPU.

If you can load the same frame of a BMDFilm image and a linear version you can run them through the color curves tool and 'match curve'...then instance this tool and feed in the LUT CUBE creator, which will apply the same change...then pipe that result into the LUT Cube analyser. That leaves you with a Linear-BMDFilm LUT.

I will check this in the evening, but I think this won't work either. The problem is that all these methods never care for the input range, but only for values.
The BMD LUTs integrated in DaVinci heave the appropriate range written into their header, and using the LIN-BMD LUT right after the BMD-LIN LUT doesn't change the source image, exacly as it should. The problem is that no other application picks up the 2nd LUT correctly, I guess because the don't (or can't?) pick up the input range.
The Cineon Converter from Fusion is super cool here, because I just hit "lin2log" and the back-conversion is done.
Nuke needs the right curve and I just set it to that for input and output, done.
This LUT system that makes totally sense for DaVinci is just not usable in any other environment, that's what I think.

This is why the Linear workflow with an sRGB viewer LUT is so much simpler....

Well the Viewer is sRGB...
But I guess you meant linear workflow with interpreting that footage as sRGB?

I just think a linear workflow doesn't make too much sense when I don't linearize the footage properly, then I can just comp "as always" and have the same results.

Thanks for all the input guys, really appreciated.

What I will do now is the following.
This evening, I'll read out everything "usable" that I can using DaVinci and Nuke. When I have constant LOG values for the white and black point, I can type those into Fusions Cineon Converter and compare it with a pre-rendered log2lin image from DaVinci, eye-matching the rest of the values (gamma etc). This isn't really accurate of course, but saves the issues with going lin2log again.
Then I'll hack that into a Macro so I have it easily available, of course I can share that with anyone interested

Things that BMD should do:
Add the BMDFilm (also for the 4K) curve into the Cineon Converter (Alexa, sLOG and C-Log are already in), and also hand out that damn formula for that curve to The Foundry so they can implement it into Nuke

I was able to create a Curve ASCII for the BMDFilm2lin LUT that is usable in Nuke, you can utilize it there. As it is based on a table, it's of course not super accurate for now "close enough"

Your right of course....

I just made a macro that does the log conversion and applies the LUT and loaded it into the LUT directory.

Sounded more intimidating then it was.

Stephan had suggested this to me earlier but I never got around to trying it.

Making that macro into a LUT would be faster, because it would run on the GPU.

I guess that could be done through the LUT cube and analyzer as well....

The longer I stay in this business, the dumber I feel

The linear-to-logarithmic step is hard to do in a LUT because of superbrights. Do Fusion's LUTs or its LUT cube analyzer support values larger than 1.0? And even if they did... how high to you sample?

I just tried making a LUT using the CUBE and analyzer running the LIN-LOG-RedGamma setup and comparing it to a RedGamma source

the LUT generated from the analyzer was up to 30% difference in the brights

the macro setting LUT was at most 0.1% difference

either the LUT or the analyzer lacks the precision

The macro LUT however functions near perfect. It would be nice to have it accelerated however, but at least it's correct.

I checked some files and compared A to B to C and still I can't get it right with the Cineon Converter.
Also, Nuke tricked me too with the LOG-values it shows so lots of my observation was wasted time

However, I found some values that work to some degree. It's not perfect, but maybe a start till BMD (hopefully) adds a native support.

Code: Select all

{
   Tools = ordered() {
      BMDFilmConversion = CineonLog {
         CtrlWZoom = false,
         Inputs = {
            Depth = Input { Value = 5, },
            RedBlackLevel = Input { Value = 40, },
            RedWhiteLevel = Input { Value = 710, },
            RedLAD = Input { Value = 481, },
            RedMidValue = Input { Value = 0.275, },
            RedFilmStockGamma = Input { Value = 0.8, },
            SLogVersion = Input { Value = FuID { "SLog2", }, },
         },
         ViewInfo = OperatorInfo { Pos = { 825, 214.5, }, },
      },
   },
   ActiveTool = "BMDFilmConversion",
}

CineonLog Node
Log Type Cineon
Black Level 40, White Level 710
Softclip 0
Film Stock Gamma 0.8
Conversion Gamma 1.0

All values are, obviously, approximate. If someone gets closer, feel free to share

You can not apply a 2.2 gamma straight onto a scene linear image. Like mentioned above
you first need a lin2log transform and then apply the right LUT to get you to your rec709 or
sRGB output.
But again this is not a simple gamma but an S-curve if its a 1D LUT.
I am using a .csp LUT that has a lin2logC prelut and then a log2rec709 3d lut for Alexa footage.
It works as a viewer LUT in Nuke or 3d Software but for some reason it does not work as a viewer
LUT in Fusion

I have attached a screen capture. On the left is the .csp LUT loaded as a OCIO File Transform ViewLUT
and on the right its loaded with the OCIO File Transform node.
All the values above 1 are clipped in the viewer.

Hi,

I've never dealt with .csp LUTs... but my first guess would be that the viewers are set to 8 bit in the preferences (don't remember where... "tweak" section maybe?). That would cause the viewer LUT to receive clipped values while the tool has no problem working on superbrights...

Stefan

hey, I've been following this thread, I got a BMPCC myself and was interested in how to do a propper linear conversion for compositing.

I was really hoping that for Fusion 8 it would have been added in the Cineon- or Gamut tool.

too bad...

Hi,

thanks for the reply, I was suspecting that and set them to 32bit float and it still clips.
(I set it in Preferences > Frame Format)

The viewer has the full float information. I can just apply a build in Lin2Log LUT and
all the detail in the highlights is there.
So maybe it only expects 3d LUTs between 0-1 in the viewer when you load them through
the OCIO File Transform?

Just adding a voice here to BMD, it would really help everyone if not just the right tools, but also the correct procedure is provided to accomplish this in as few steps as possible... So BMD camera footage to resolve to fusion and back. It would also be great if the viewer LUT in Fusion could additionally incorporate grading info from resolve.

So to clarify, get your BMDfilm footage in Resolve, grade it how you want, export the original footage in linear to fusion, load a viewer LUT that will let you see the identical image, ideally with the preliminary grade incorporated (either by stacking 2 LUTS or by exporting a combined LUT from resolve), and return the renders to resolve and have them match with what they were before (hence, with the rest of the film)...

+1
On why doesn't BM Fusion support LUTs for it's own hardware?

The linear-to-logarithmic step is hard to do in a LUT because of superbrights. Do Fusion's LUTs or its LUT cube analyzer support values larger than 1.0? And even if they did... how high to you sample?

I had a version of the LUT fuses which allow HDR including negative values. Not sure if I can dig it out now but it is possible without much trouble. As far as I can remember not all LUT formats can handle HDR though.