F5: ACES vs YRGB exposure differences in Davinci Resolve?

[Grug]

Hi Guys,


I’ve been running into some very strange discrepancies between grading in the ACES and YRGB colour spaces in Resolve, and was wondering if anyone might be able to shed some light on them?


In a nutshell, what I’m seeing is the following:


- in YRGB, images respond predictably to LUTs, but yield less subtle variation in skintones
- in ACES, images become violently overexposed when applying LUTs, and contain boosted red tones
- After correction, images in ACES appear to yield more accurate and nuanced colour reproduction


Here’s the original log image (slog3 from a Sony F5) and it’s waveform. The exposure values for her skin and the wall in the background are sitting where I’d expect them to be:

Now in YRGB mode, when I apply the LC709A LUT to the log image, exposure values are lifted to approximately where I’d expect them to go:

LC709A is designed to mimic Alexa’s Rec709 output, and does a nice job of normalising slog3 images, it doesn’t crush the blacks though, so a slight contrast curve is needed to put the darker tones where they belong:

There’s a slight overall green tint to the colour here, but it’s easily corrected out, and we have a normalised image. When apply a straight Filmconvert grade on top of the image, everything ends up right where we’d expect it to go without any further corrections or fine-tuning:

When we flip Resolve over into ACES v1.0 mode though, and apply the same LUTs/Looks, we get very different results. (note I’m not using any IDT or ODT on the footage, merely grading it in the ACES colour space). Applying the LC709A LUT in ACES results in this:

Exposure values blown sky high. Applying the straight Filmconvert look on top does the same thing, only worse:

Now with some contrast curves, we can correct the ACES image back to normalised levels:

However, there is an excessive amount of red in this normalised ACES image. Now for this scene, I had uncorrected 650w tungsten lights coming through daytime windows to simulate late afternoon light, and that would have increased the red in the image, but given the light wasn’t hitting the actress directly, it seems odd that the red levels are so boosted. With the excess red dialled out however, the ACES image yield more nuanced and appealing skintones. Here’s the corrected Filmconvert look, with the excess red removed and the exposure levels brought down to normalised levels:

Compare that to the YRGB Filmconvert image, and you can see how much more nuanced and realistic the skintones look:

And again, ACES with a straight Filmconvert look applied, brightness corrected and excess red removed:

And the YRGB version:

Again the skin looks much less wooden, and more realistic in the ACES version. The richer reds of her lips are more clearly defined from the skin on her face.


Is anyone else experiencing similar strange responses to LUTs in ACES in Resolve? And does anyone know what might be causing the discrepancy?


Any insights would be greatly appreciated.


Cheers,


Mark

 

[marvinhello]

ACES does not work with LUTs(and filmconvert etc.). You need to use specific IDT and RRT/ODT and manually do the grade, ACEScc offers a familiar starting point.

I suggest you do some research on the science of ACES.

 

[falls6]

thanks for your information

 

[Grug]

ACES does not work with LUTs(and filmconvert etc.). You need to use specific IDT and RRT/ODT and manually do the grade, ACEScc offers a familiar starting point.

I suggest you do some research on the science of ACES.

I know what ACES is designed to do. What interests me is what it can be MADE to do - which as I posted above, is yield better results from LUTs/Looks than standard YRGB processing in Resolve can (albeit with the added complexity of having to compensate for the luminance/red spikes). Which to me, seems worth exploring.

 

[Tom Roper]

That's always been my practice too, as Marvin said, the idt and odt, not luts. I'm not arguing with the result of your grade.

 

[DPStewart]

I spent an afternoon not long ago looking under the hood at how ACES does its thing. After the headache wore off it made sense, but it's a very different path.

I don't think our use of the path from LOG files to YRBG and a typical REC709 end point is any more intuitive or any less complicated than the ACES path - I think we're just used to it.

It's not that ACES 'doesn't work with LUTs' so much as it is that it requires a different sequence of the steps. Exactly like you've established.

I think the moral of this story is that you have figured out a very effective methodology. After a year of doing it this way it will feel as obvious and familiar as anything else.

Kudos to you - you win this round.

 

[broughtonfilm]

A lucky side effect of the added manipulation, the image changed its saturation in mids, possibly a specific color range has changed its characteristic which makes the skin look more pleasing. That's all. I'm sure it is possible to add a node or two to the YRGB version and make it look exactly the same

 

[Nate Weaver]

From an exposure standpoint, I've always gotten the same results from LC709 and ACES. For that matter, VERY similar color response as well. Definitely never seen color casts between the two.

Something is up with the way you're applying things?

 

[Nate Weaver]

In this screen grab:

This looks like you're turned ACES on, but not set the IDT to Sony SLog3.

And then, it looks like you've applied the LUT over top that, which as Marvin said above, you can't put a LUT on the tail end of the ACES chain, it doesn't work. It applies the LUT and clips highlights bad.

Edit:
I just re-read your post. You mention not applying the IDT, I missed that.

ACES doesn't work if you do not apply the IDT. Flat-out. Just turning on ACES and leaving IDT blank just yields you a random, broken image. ACES needs to know how the input footage is gamma and chroma mapped, and if the IDT isn't set to tell it, you just have random pixels.

If you like how Filmconvert or Impulz renders colors, unfortunately ACES is no longer an option.

So, um, your ACES images above are "user error", for those reading along. Sorry Mark, I just feel I needed to state that clearly for people confused by ACES and trying to follow along. No harm intended!

 

[Nate Weaver]

One other thing, it looks like you're applying the SLog2 to LC709 LUT in your first couple images, on SLog3 footage.

That Slog2 to LC709 LUT was distributed and included in Resolve about a month or two before Slog3 showed up.

That LUT footage combo would lift the footage, which yours doesn't need, it's exposed fine. It would also cause highlights to burn more than they should.

Apply the Slog3 to LC709 LUT and things will look better, it won't be lifting things.

 

[starcentral]

Mark if I remember correctly you are shooting hybrid mode of sgamut3 with slog2 which is only* possible if you shoot with the R5, or if you use the F55 in custom mode with MATRIX s.gamut3.cine "like" enabled.

Could this have something to do with your observations?

cheers,

 

[Nate Weaver]

This is ACES vs LC709.

No exposure or curve modifications.

 

[Lenilenapi]

I'm curious what people feel they are getting from the Impulz and other film emulation Looks out there.
Are they worth spending money on? Do anything you can't do yourself in grading?
What does that 50D curve look like if you raise the level a bit as it just looks a little under here.

 

[Nate Weaver]

I'm curious what people feel they are getting from the Impulz and other film emulation Looks out there.
Are they worth spending money on? Do anything you can't do yourself in grading?

You have to see them applied to footage shot in many different situations, with different colors in the frame. The grab here, there's not to much difference between LC709 and the 50D LUT. It's one of those products you just have to work with and try on lots of different material before you can see how good it can look.

Film was really bad with color accuracy. It's beautiful, but it's not even close to accurate by what is possible to capture by a modern, HQ sensor and wide gamut bayer filters.

I've mentioned this already in other threads, but the gentleman who made the Impulz LUTs made them by shooting a 32,000 swatch color chart under controlled, repeatable conditions with the films. The film was then scanned, and the numeric color values were read out by software he wrote to directly create the LUTs. It's an attempt to mathematically model real film color response, and after spending the last year with them I think he did a tremendous job. If you apply enough knowhow and resources, it's not impossible. Just a mathematical model. A table of output values for a given input.

So the answer to your question "can you do this yourself in grading?", the answer is yes and no. For a specific image, you can recreate the broad, obvious strokes of what some of the film LUTs are doing by hand, but for other images, there's just too much subtle tweaking going on to recreate in a sane manner. These are not things you can just whip up in Lift/Gamma/Gain trackballs. If you REALLY understood on a deep level what film does, you might be able to recreate in the Hue vs Hue and Hue vs Sat curves window in Resolve, or color fettle tool (sp?) in Lustre. It's important to realize the magic of the film LUTs is having all the weird and beautiful color inaccuracy done in one step. Any knucklehead can recreate the broad strokes by adding cyan in the shadows and skewing midtowns warm. A film LUT created from color charts scanned is a different animal.

By the way, the color "science" of Arri, Red, Blackmagic and LC709 are all in one way or another based on film LUTs. Arri in particular applied what they learned with their film scanners to the Alexa, and that's part of the reason it looks so nice (I assume you have to have a good understanding of the finer points of negative film to be able to make good scans. Lots of lookup tables in the scanning and film output business)

What does that 50D curve look like if you raise the level a bit as it just looks a little under here.

Looks great. The LUT is just modeling the lower mid gamma of 50D compared to the others. Nothing you can't tweak back on your own.

 

[Lenilenapi]

So do they go on top of your 709A LUT or instead if it ?
If it's easy , could you post some examples of slog3 709typeA alone vs a good Film LUT ?

Also any suggestions for which inexpensive package works well w FS7 ?

 

[starcentral]

I have the impulz collection and a few others and still I tend to use the Sony F65 beta slog2-to-Rec709 3D LUT, or for slog3 the free Kodak & Fuji ones (now included in Davinci v10 and higher).

In general I'm not a fan of instant "looks" because when they might work for one set of clips, they may not work for another within the same project. So I think its more important to just have a general film base to work with and grade from there to whatever style or look you are going for. To me the sony, Kodak, Fuji ones will allow you to do just that.

Spend some time at www.liftgammagain.com and guarantee you'll learn some interesting things.

If you want to download the Kodak/Fuji LUTs here they are: www.starcentral.ca/forums/LUTs/kodak_Fuji_LUTs.zip

Another set of digital effects filters I've had A LOT OF FUN with are the Tiffen Digital Effects ones.. it has a lot of "instant" film stock looks that you can quickly preview and not to mention just a ton of filters that Tiffen is known for. It can blow you away playing with it. The trial is free for a limited time.

Cheers,

 

[Grug]

One other thing, it looks like you're applying the SLog2 to LC709 LUT in your first couple images, on SLog3 footage.

That Slog2 to LC709 LUT was distributed and included in Resolve about a month or two before Slog3 showed up.

That LUT footage combo would lift the footage, which yours doesn't need, it's exposed fine. It would also cause highlights to burn more than they should.

Apply the Slog3 to LC709 LUT and things will look better, it won't be lifting things.

Ah! Quite right. In all the madness switching between colour sciences and raw SLOG2 and compressed SLOG3 timelines I applied the SLOG2 LC709 instead of the SLOG3 version by mistake, that explains the overly elevated blacks in that YRGB LC709 frame.

As I mentioned earlier, I know this isn't how ACES is intended to be used, but try applying your Impulz 50D LUT on your sample frame in ACES, and just pull down your exposure and the excess red to a normalised level and I think you'll see much more lively skintones than you're getting in YRGB. It really does make a noticeable improvement to skintones with Sony footage without having to resort to more complex secondary corrections.

Mark if I remember correctly you are shooting hybrid mode of sgamut3 with slog2 which is only* possible if you shoot with the R5, or if you use the F55 in custom mode with MATRIX s.gamut3.cine "like" enabled.

Could this have something to do with your observations?

cheers,

These samples are all SLOG3 ProRes4444 recordings, so I don't think that's what's in play here. I'm working solely in Sgamut3.Cine these days, so I only shoot SLOG2 with raw (where I can then move into Sgamut3.Cine in post), all compressed recordings are SLOG3 so that I can access the nicer gamut.

I think Nate's posted samples highlight quite nicely how ACES just seems to pull out more red specifically in skintones (which is why I think it translates to less wooden looking skin when you apply film LUTs/looks in ACES and then correct the oddities). Look at how her lips and the shadowed side of her face just seem to have more blood in them than the YRGB LC709A sample (but the red on the cars in the background looks the same).

It's that specific improvement in skintones that interests me, particularly when combined with the lovely midtone contrast and colour response you get from the film-emulation LUTs/looks.

 

[Nate Weaver]

but try applying your Impulz 50D LUT on your sample frame in ACES, and just pull down your exposure and the excess red to a normalised level and I think you'll see much more lively skintones than you're getting in YRGB.

It used to be that applying LUTs over ACES in Resolve clipped all highlights at 60IRE, which was a bad strange look. Haven't tried it lately. My experiments in color have trailed off quite a bit since I jumped ship.