S-Log Compositing Workflow

[MANUEL_MOUSIOL]

We are shooting a no budget short movie and I have some CGI elements to composite into it. We are shooting with an Sony a7sII in S-Log3 and I was wondering how the workflow from Vray to After Effects/Premiere Pro would be including the color grading I will do in the last step.

Would I render out the animation in EXR with 2.2 gamma linear workflow or do I have to use some burned in LUT?

Would love to get some ideas from you experience film/video guys.
Thanks for any input.

[joconnell]

Shoot a macbeth chart for your footage so you can attempt to linearise it before starting to comp your 3d into it. Sony might have a lut to convert from s-log3 into normal linear space so you'd use this as a first step, comp your 3d and then put your film grade on to the result.

[im.thatoneguy]

SLog -> SLog2Lin LUT -> Linear Comp CG -> Lin2SLog LUT or Cineon Log or whatever the colorist wants.

[MANUEL_MOUSIOL]

So I make the footage linear with the Slog-LUT, then comp over that the rendering, then after that bring it back to slog with another LUT and then do the color grading?
Sorry, I never used an LUT and never worked with slog...

[joconnell]

Yep. Pretty much all rendering software and comp software works in linear these days, it's much closer to how light works so you'll get more realistic results easier. Log is a way of saving more exposure information into a very small space but you need to take it out of that space for you to work with it in compositing. As Gavin mentioned colourists are quite well set up to work with raw camera footage which is in log so after you're done comping you can ask the colourist if they can use your linear renders directly or if they'd prefer something like the original footage. If they want log you can normally apply the opposite of a lut to put it back to where it cam from.

[im.thatoneguy]

Yeah, you don't want to comp in gamma space whether that's sRGB, Gamma 2.2 or any of the Log species. The basic problem is that the math just doesn't work correctly. Our perception of the world is logarithmic. Double the volume of sound and we perceive it as a small volume increase. Same with light. Hence why cameras are measured in "stops" each stop being double of the last. So most software expresses the values in exponential values since that's how we're used to thinking (and it saves a lot of memory). But the math still needs to be done in the real world's scale which is linear: add another light and you have 2 lights not 4, double the power to your bulb and you get twice the light. If you double the exposure time of your camera you only get slightly brighter because the real world is exponential in nature. If you double the brightness of an image in Photoshop it gets way way way brighter.

Your monitor's "gamma of 2.2" in mathematical terms is "Value raised to the power of 1/2.2" or X^(1/2.2). So if you had a light which put out .18 lumens. And you added a second light you would have .36 lumens. (0.18 + 0.18 = 0.36 lumens). If you do that same addition after you've gamma corrected though you have a problem. (0.18 ^ (1/2.2) = 0.45 gamma corrected lumens). 0.45 + 0.45 = 0.9 gamma corrected lumens, even if you linearize that back to the non-gamma corrected *after* the addition you end up with (0.9 gamma lumens ^ 2.2 = 0.79 linear lumens). So it would be like every time you add a light to your set magically ending up with an additional 2 lights worth of photons flying around. In this case it would be 0.18 lumens + 0.18 lumens = 0.79 lumens instead of 0.36.

The universal rule of thumb is:

1) For each piece of footage, apply the inverse gamma/log curve to convert it to linear.

2) Do all your stuff in one consistent linear colorspace.

3) Apply a new gamma curve (if necessary) to match what your display expects.

If you've heard of ACES (Academy Color Encoding System) it follows this model. Every camera has a LUT which they call an Input Device Transform or IDT. And that converts the footage from whatever random wonky colorspace the camera manufacturer has chosen: RedGamma 4, C-Log, S-Log etc... into linear color.

Then everything is consistent and matches: red is red, blue is blue, green is green, middle gray is middle gray.

Then you apply an Output Device Transform or "ODT" when you're done which converts it into an image the display expects whether that's sRGB, rec709, Kodak projection stock, log, s-log, HDR rec2020 etc.

[MANUEL_MOUSIOL]

Wow, that was a lot of info for this small brain. thank you!
Anyway, I'll try it out. I have a few s-log3 LUTs from Sony and saw that some gave me strange over-contrast and over saturated picture and I can't really help but feel that putting that on the live action footage and then color correcting the CGI towards that would give me right results, even if returning the whole comp back to s-log3...But maybe I just have to find the exact right LUT...
Bringing back the whole thing: I'd just put the LUT onto the whole comp inversed (still have to find out how to do that) and that should do the trick before color grading everything together, right?

[Dave_Wortley]

Just convert it to a linear colour space in format like EXR or DPX and make sure you save as 16bit, then you can apply an sRGB LUT and work as you normally do in Max/VRay.

Sometimes you have to do something like convert from sLog -> REC709 -> Linear but just do it in daVinci Resolve.