Created and applied LUT in VFB looks different

[piotrus3333]

I misspoke. In fusion the LUTs do not precisely match the VFB.

I am not sure precisely what is going on. I can indeed apply a LUT to a shot with 0-900.0 and get results in fusion, but those results are not precisely the same.

Normally, we just start tone mapping from scratch in fusion. But on some recent projects made use of the LUTs out of the VFB.

If I am understanding piotrus3333 correctly it sounds like the only effect the VFB LUTs can have outside 0-1 would be gain based on the slope of the line between the last two points of the LUT (per channel?). So since we significantly rolled off the highlights we still got significant gain reduction in that 900.0 value, but linearly rather than a nice roll off.

That is exactly how it looks to me, linear above 1 if lut covers 0-1 range only. But Fusion supports "proper" cube format luts that cover as much range as you need, e.g. Linear to CineonLog can be done with cube lut with enough precision.

[Joelaff]

That is exactly how it looks to me, linear above 1 if lut covers 0-1 range only. But Fusion supports "proper" cube format luts that cover as much range as you need, e.g. Linear to CineonLog can be done with cube lut with enough precision.

Fusion's LUT support works fine, yeah. I guess the LUTs out of VFB are flawed.

Since you have done a lot of research on this, what do you think is the best "fix" that Chaos could apply? In an ideal world we could simply save a LUT out of the VFB and then apply it in Fusion (or any compositor) and the output would match.


FYI: Anyone using Fusion that wants to do Reinhard burn (Exposure->Highlight Burn) in Fusion you can use this Custom Tool node I made, which uses the same math as VRay (according to Vlado) ReinhardBurn2.zip Note the mask input triangle to the node is light grey rather than blue due to the way macros inputs work in Fusion.

[piotrus3333]

Fix from Chaos? I don't think it is needed that much. VFB is hardly a colour grading tool. It should accept some basic luts and that is it. OCIO layer would be nice and useful as now you can only do one conversion on top of layer stack.

and the linear interpolation Fusion does on lut from VFB looks like that (it was S curve in VFB with curves layer saved as lut):

[piotrus3333]

Fusion's LUT support works fine, yeah. I guess the LUTs out of VFB are flawed.

Since you have done a lot of research on this, what do you think is the best "fix" that Chaos could apply? In an ideal world we could simply save a LUT out of the VFB and then apply it in Fusion (or any compositor) and the output would match.


FYI: Anyone using Fusion that wants to do Reinhard burn (Exposure->Highlight Burn) in Fusion you can use this Custom Tool node I made, which uses the same math as VRay (according to Vlado) ReinhardBurn2.zip Note the mask input triangle to the node is light grey rather than blue due to the way macros inputs work in Fusion.

Not flawed just limited a bit. Also because of limited precision luts are not the best choice for tonemapping. Spliting hairs if you ask me but this is what professionals say.

[Joelaff]

Fix from Chaos? I don't think it is needed that much. VFB is hardly a colour grading tool. It should accept some basic luts and that is it. OCIO layer would be nice and useful as now you can only do one conversion on top of layer stack.

I hear you that it's not a color grading tool, but it is nice to be able to work with a particular look in the VFB and then get that same look as a starting point in a true compositing app.

Also, it seems like a lot more people than you or I might think do use the VFB for the final processing (or almost final). Not everyone has access to, or knowledge to use, a bunch of different software.

I think I would like the VFB to be to export a LUT that would reproduce the effects identical to the VFB in the likes of Fusion. I never used to think that much, but having used in more in a couple of recent projects it would be nice to use that as a starting point, rather than re-grading, simply because we end up showing samples out of the VFB to agency/client, and then if the re-grade is different sometimes we are chasing our tail to get it the same. Of course we could back to the old method of just always saving out EXR and loading it into comp, which works fine, but requires an extra step each time.

and the linear interpolation Fusion does on lut from VFB looks like that (it was S curve in VFB with curves layer saved as lut):

That is what I figured from what I saw... just gain past 1.0. Curious-- how did you generate that graph image? Is that actually from a LUT, or just an example of roughly what it does?

[Joelaff]

Not flawed just limited a bit. Also because of limited precision luts are not the best choice for tonemapping. Spliting hairs if you ask me but this is what professionals say.

Totally true, but I suppose it could output higher precision LUTs.

[piotrus3333]

The graph is a waveform view of 0-5 black and white gradient with lut aplied. From Fusion.

VFB grading transfered via lut to Fusion 1:1 (i'm assuming you output linear srgb):

VFB - start with full highlight burn (so simple Reinhard y=x/(x+1)) + color corrections + display correction srgb or 2.2
turn off highlight burn and save the lut.

Fusion - Reinhard + lut from VFB + View LUT matching the one in VFB

done. This is 1:1.

you can use CineonLog in place of Reinhard. Save from VFB a lut with no adjustments, load it into Lookup Table layer with Work in log.