Trying to get a clear idea about grading SLog2 and using LUTs

[wilecoyote]

I've reading and testing during the last few days about the oh-so-celebrated SLog2 in my new A7S, trying to wrap my head around it.

First of all: my only previous experience with log formats was with the Cinestyle preset in Canon DSLRs. In those days, I just added a contrast curve at the end of my grade, and graded "before" it; for example, in After Effects I'd use the Master Curve in Colorista II, or I'd add a adjustment layer on top of my footage with my curve. Then, when I started using Resolve, I'd add a node with just my contrast curve at the end of my grade, *after* all the other nodes.

This seemed to be the recommended way to work, according to Stu Maschwitz and also this thread:

https://forums.creativecow.net/thread/277/22579#22579

Where Alexis von Hurkman seems to imply that the contrast curve or LUT is better applied after all the other nodes.

(Note that, when I talk about LUTs here, I mean 1D LUTs, i.e., contrast curves, not 3D LUTs).

So now I get my Sony A7S, I shoot Slog2 tests with it and I import them into Resolve… except that in it, the only SLog2-to-709 LUT that I find is in the "Input LUTs" section, which suggests that the way to work with it is to apply the LUT *before* start grading, i.e., in the first node, and then proceed as if you were dealing with a regular 709 video file. Note also that, in the previous link, Mike Most says that SLog2 is actually designed for that:

"… In fact, one of the design criteria for both Panalog and Sony's Slog was the ability to be corrected directly in a video environment with "normal" video scaled controls,"

…Except that, when I do that, I can't seem get any of the supposed advantages that SLog2 has, specially the higher dynamic range in the highlights.

For example, consider this Slog2 clip:



(The book is the closest thing I had to a human face in my house :cheesy.

After I apply to it the SLog2-to-709 LUT built into Resolve, I get this: highlights completely blown.



I then add a node after it and try to recover the highlights, while keeping the book (the "human face") in the 40% range in the waveform monitor. And I do this with the video color controls, not the log ones, as per the quote above...

…Except that everything is a mess. I need to bring down the gain to ridiculous levels to recover the highlights, and then I have to stretch the difference between lift and gamma so much, that all the noise in the shadows shows up:



Look at the bottom of the blue couch. The noise is even worse in motion. The impression I get, while playing with the image, is that there simply isn't enough information in that part of the shot (between shadows and midtones), and I'm almost tearing it apart when grading it.

So, what am I doing wrong? Are you supposed to use LUTs at the beginning or at the end of the grade? Is it better to design your own curve instead of using the factory-supplied one in Resolve? How is everyone else doing it? Or am I asking too much of the camera, and there's simply too much dynamic range in this image?

 

[dhessel]

Luts should almost always be the last node in your grade with all correction happening before, since they can and do destroy/clip data. The only time you would not is if using the 2 step film emulation process for the impulz luts. One reason why it is blowing out so much is the S-log2 profile in Resolve is not made for S-log2 from the A7s. It is different in the A7s that other sony camera's in that the A7s is in video range and the other are full range, so it add much more contrast than it should. I believe there are some luts from Alister Chapman floating around that may work better...

 

[Andrew Stone]

Um sorry to be blunt...

But STOP shooting into windows unless you are going to put up ND gels to bring the sunlight down a few stops. And everything in your shot is rear lit from the window! Your problems have nothing to do with you using log gamma curves. Sorry.

Here is a good place to start on lighting techniques:

http://lowel.tiffen.com/edu/

 

[Terrence Wilkins]

Slog-2 has a lot of grain to it, to get the cleanest results out of camera try exposing to the right by about 2 stops. This should give you the most noise free images.

 

[cyork]

In that scenario, yeah, you're going blow some highlights but the noise will be far better if you expose up a 1-1.5 EV and bring everything back down in post. Also, that curve is pretty hardcore on the bottom end. You could lower the mids a bit and get smoother rolloffs.

 

[wilecoyote]

Um sorry to be blunt...

But STOP shooting into windows

No worries, no offense taken :cheesy:. I have to clarify, however, that I do in fact know how to light a scene (or at least, enough to not backlit a subject as extremely as I did in my screencaps). This was only a test; I wouldn't think of doing this in an actual job with an actual human being. It was just that, after hearing so much about the A7S/Slog2's famed dynamic range, I decided to test it using the most extreme example I could find. Judging from your response, I'll file this under "I was asking too much of the camera"...

 

[wilecoyote]

Luts should almost always be the last node in your grade with all correction happening before, since they can and do destroy/clip data. The only time you would not is if using the 2 step film emulation process for the impulz luts. One reason why it is blowing out so much is the S-log2 profile in Resolve is not made for S-log2 from the A7s. It is different in the A7s that other sony camera's in that the A7s is in video range and the other are full range, so it add much more contrast than it should.

Thanks for your response; I guess this is what was tripping me up. As for the LUTs, after I posted this I remembered a blog entry I read some time ago:

http://mikemost.com/?p=251

Where he definitely confirms that LUTs should come last. However, this leads me to another doubt: doesn't Resolve work in 32 bit float? And isn't one of the advantages of that environment that no clipping occurs? If you apply an extreme grade in one step, you just have to apply an equally extreme one in the opposite direction… and that's it. And yet, I tried applying a Curves node before a lift/gamma/gain node, and some highlights were definitely clipped.

 

[wilecoyote]

About overexposing Slog2: I've tried that too, but in that case, all the detail in my highlights is lost, even before I apply the curve. I guess the test situation I used above was too hardcore.

 

[stevedocmaker]

About overexposing Slog2: I've tried that too, but in that case, all the detail in my highlights is lost, even before I apply the curve. I guess the test situation I used above was too hardcore.

If you're exposing for the scene outside the windows, then you're already overexposed by several stops. If you're exposing for the book, then it looks about right, which of course is really 2-3 stops under optimal for Slog2. This is why you're seeing noise in the shadows. If you're going to use LUTs, you have to have the clip exposed properly to begin with.

 

[dhessel]

Thanks for your response; I guess this is what was tripping me up. As for the LUTs, after I posted this I remembered a blog entry I read some time ago:

http://mikemost.com/?p=251

Where he definitely confirms that LUTs should come last. However, this leads me to another doubt: doesn't Resolve work in 32 bit float? And isn't one of the advantages of that environment that no clipping occurs? If you apply an extreme grade in one step, you just have to apply an equally extreme one in the opposite direction… and that's it. And yet, I tried applying a Curves node before a lift/gamma/gain node, and some highlights were definitely clipped.

It is not Resolve, it is the Lut. Resolve is indeed 32bit float and manually removing the log curve would not clip data. Luts are normally set up to only work in the range from 0 to 1 where black is 0 and 1 is white, so Luts in general force any resulting values outside of this range back into 0 to 1 when they are applied. This is perfectly acceptable behavior if the Lut is applied last.

After re-reading your post I may have mis-interpreted it the first time. Are you seeing cliping without using any luts in between those adjustments?

 

[wilecoyote]

After re-reading your post I may have mis-interpreted it the first time. Are you seeing cliping without using any luts in between those adjustments?

If you mean my last post, yes. And I wasn't even using a LUT, but just the Curves tool in Resolve. I added a node with Curves (with a S-Curve), and then I applied another node and tried lowering the gain control. If you did that without the curves, you could see more highlight detail as you lowered the gain level; with Curves, however, you lower the gain and all you see in the Waveform is the typical flat line that appear when something is completely clipped.

 

[dhessel]

Ok so I think it is the way that you are performing these corrections is what is causing this behavior. If you add some gain to the first node pushing the highlights into super whites, then apply another node with an s-curve followed by another node where you gain the footage back down it might appear like it has been clipped by Resolve but that is not really the case. The curves allow adjustments in the 0 - 1 range, any values outside of this range will be adjusted by interpolating the curves outside of their adjustment range. When you gain up you push the highlights outside the 0-1 range, applying an s-curve by its nature will flatten out the top of the curve and resolve will continute this flattening out into the super whites. Now when you try to gain down after the curves little highlight data is recovered since the previous curves just flattened all that data down. You basically just clipped the data yourself when you flattened out the top of the curve.

If you do this process again and create a single point in the middle of the luma curve and pull that point toward the bottom right corner of the box you will create a curve with an an increasing slope that never flattens out, gain down in a node after that and you will find that the highlight detail can be recovered as you expect showing that Resolve and using curves do not clip data as long as you don't clip it yourself.

 

[theprestige]

Theres very few steps to give yourself a great base with SLOG 2. A guy at the Sony Centre in Pinewood Studios taught me as he had been grading SLOG 2 for some time.

This is strictly for SLOG 2 not anything else, mind.

Step one: Crush your blacks as much as you can.

Step two: Clip your highlights

Step three: Increase saturation by however much you want.



That's really all there is to immediately getting a great image. Obviously theres a few refinements to be made, but that's a great start.


I'm not sure if the same can be applied to SLOG 3.

 

[jurny]

rotfl what a stunning piece of advice.
simply restore the highlights with highlight recovery or use a secondary to bring the desiredred luma range down

 

[caseywilsondp]

the first issue i would guess in the workflow is using the slog2 lut in davinci. it's an slog2 lut for what i wonder? slog2 on the a7s and the f55 are very different.
i don't use davinci, but in my typical workflow, the lut is the last layer. if i'm just doing a quick grade in premiere with slog2, i'll put the lut i made for the a7s slog2 in, then if i want to make adjustments i'll add layers above (before) to do what I need.
so for this shot I simply plugged my lut in, and added a layer above (before) to pull down just the highlights. normally i would put a power window around the window and adjust the highlights, but for the sake of time i just pulled all the highlights down.
this is all just based on the jpeg i pulled from your post, i imagine more detail could be resolved from the actual file.


so anyways thats my workflow. use my lut as last layer/node, make adjustments above/before

 

[theprestige]

rotfl what a stunning piece of advice.
simply restore the highlights with highlight recovery or use a secondary to bring the desiredred luma range down

Hey this is advice from a professional grader and he showed me himself and the results spoke for themselves. We applied it to 2 different clips. Don't knock before you try

 

[paulcurtis]

There are two approaches to grading log footage.

But first think about what you are doing. You are grading a range of light that is greater than the display of your monitor or output device. The light out of that window is brighter than your monitor could ever show.

If you record a 'normal' profile on camera then the camera is compressing that range of light to 0 to 1 using a gamma curve of some description. It is compressing by reducing the tonality of the highlights and shadows in favour of mid tones. This is basically what footage on your TV looks like. This is display referred.

When you record log what you're actually doing is asking the camera to store all the light distributed fairly evenly from shadow to clip. This is scene referred.

So my point being is that the two approaches are

You keep the 0 to 1 flat curve of the log and you add contrast and colour to it and grade it to your taste. You are still 0 to 1 but you are manipulating the range in that. A lot of people do this, it isn't technically correct as you loose out on what log is but whatever works for you. In a larger pipeline with other vendors then no, you can't do that. When you are grading this, you are doing what the camera would do in a simple profile although you have a bit more control over it.

The second approach, the technically correct approach is when you add a shaper LUT in any decent grading package what you're doing is usually expanding that 0 to 1 back to something which relates to the light that was there on the day. So for Slog2 you'd usually get something like 0 to 8, sometimes 0 to 12. When you view that on your monitor you will see the highlights totally blown out unless you take your exposure down on your grade. Mid grey is normally represented at 0.18. So light doubles, 0.36 is a stop over, .72, three stops and so on. So 0 to 12 is about 6 stops over mid grey. Which is about right for the SLog2 container.

Why do this? Well there are all sorts of operations on images that work better when they are using scene referred light. Blurring, motion blur, filtering etc, work properly and the basic math for colour correction is simpler when you have 'real' light. However not all grading packages show the values of linear light, i think resolve still likes to scale it all back to 0 to 1, but i could be wrong. But underneath that is what the values are.

But what this does mean is that at the end of your pipeline you would have to shape that massive range of light back down to 0 to 1 in order to display it. Of course once HDR displays are common place that last step probably isn't needed.

This is why something like ACES includes a RRT and ODT, those last steps are about shaping the 'real world' light into something that can be viewed on different devices. A monitor vs projector would be able to shaped differently.

It is complicated to get your head around but i think the key is that final step. That shaping of the light back into display referred space, that's what the soft clipping or other operations are. For example a film print LUT expects 0 to 1 range so before going into one of those you need to shape your light accordingly.

You need to experiment with your own personal workflows.

But as i mentioned above, you can just eyeball the display if that works for you. Just add gamma to the original log file and ignore these shaper LUTS

cheers
Paul

 

[Samuel H]

It still sounds terribly silly to me, no matter who it comes from. I shoot slog2 because I want to have as much DR as I can, clipping shadows and highlights as steps 1 and 2 doesn't make sense to me. And shooting with low saturation just to increase it in post is inefficient as well, you're throwing away part of the already limited color space the codec can use (I solve this by increasing saturation in camera).

If you're doing it by hand (as I always do)... Highlights are already where they need to be straight out of camera. Shadows have to be brought down but there's no need to crush anything, just bring down the black level without crushing anyehing. Then add a curve to get better contrast, and in this step, yes, crush highlights and shadows if that's what you want, but it's a content-specific decision in which you have to be aware of what you actually want to do (I hardly ever crush anything: I reduce contrast in the shadows and highlights, and increase it in the mids, but always trying not to lose detail.

 

[jurny]

Hey this is advice from a professional grader and he showed me himself and the results spoke for themselves. We applied it to 2 different clips. Don't knock before you try

So am I. That's why i find this 3 step program total bs. You grade for emotions. You help the scene. imho a7s slog2 is so versatile i have problems selecting the actual grade. So many options...

 

[paulcurtis]

It still sounds terribly silly to me, no matter who it comes from. I shoot slog2 because I want to have as much DR as I can, clipping shadows and highlights as steps 1 and 2 doesn't make sense to me. And shooting with low saturation just to increase it in post is inefficient as well, you're throwing away part of the already limited color space the codec can use (I solve this by increasing saturation in camera).

There's a wider view though depending on where your work is destined for or if you're working with multiple people. If you are shooting for projection, for HDR, for theatrical release then the benefits of a true full log workflow are clear. That final shaping step can be switched to match whatever deliverable you want. If you are shooting for yourself or for the web only then the extra overhead of a log workflow probably isn't justified either.

The log formats come from mimicking cineon scans and film workflows.

The other side is if you're doing post work, comping and vfx, then a full log workflow is needed there too - things just work better. Adjusting exposure, power windows etc,. look a bit better in full scene linear.

Disconnecting your work from your deliverables is needed when projects may get delivered to various sources. From a film out, an web trailer, a 709 master, a P3 projection or HDR display. Your work is mastered at the widest color and gamma space possible and you shape it for each deliverable.

But the vast majority of projects don't require that and the effort of a true log workflow is silly. But i think it's important to realise *why* we have log workflows in the first place and where they originated.

For the most part any picture profile holds around the the same range as a log version, just distributed differently.



cheers!
Paul