What is your favorite workflow for keeping all DR available on MX?

[Paul Leeming]

I think one of the factors in favour of being able to go close to the edge but not over with Red, is that we have realtime indicators of exposure, allowing us to adjust continuously to maximise the top end without blowing highlights.

This is harder with DSLRs and other digital sensors that don't give you realtime RAW exposure levels.

As Gunleik said, it's two (or three) things that together help maximise DR and latitude:

1. The camera operator/DP needs to expose correctly to maximise captured latitude and DR.

2. The "developer" needs to tweak the curves etc to maximise the exported file's DR and latitude.

3. The colourist needs to utilise as much of the DR as possible and properly compress it down (via curves etc as opposed to hard exposure adjustments) to the final medium's latitude limits.

Miss any of those and each subsequent step becomes harder.

Cheers from Tokyo,

Paul

 

[Hrvoje Simic]

Hm. There is a shoulder, but it is just linear, right?

Shoulder is added to the data afterwards, the data is recorded as is. DR is fixed, shoulder is not.

Red does have a minimal highlight compression to make highlights coming from a linear capturing medium more "film like", with avoiding hard clip you'd otherwise have, but this is not the typical shoulder you are used to with film. Apart from that, shoulder in this case is not a part of the medium, but processing i.e. viewing/development of captured data, which you can affect.

With changing ISO you decide where middle gray will be, and by that what "shoulder" and "toe" will later do to your data, depending on the priorities in your scene.

You can treat Red as a fixed film stock, after testing the optimal setup to get the most similar response, and you can also harvest the advantage of digital RAW and use variability of "shoulder" as a tool for optimizing usage of DR.

I get ETTR, but now that the new generation chips have reduced the noise floor, isn't protecting the hilites paramount? It seems like protect means "don't clip" and thus avoid the right side at all costs.

Depends on scene and priorities, but yes.

 

[Gunleik Groven]

Depends...on what type of scene is and where you want to go in post.
Lowering down from the clouds is less of an issue than raising up from the dirt.

Nope... -
Well, sure...

Thing is that you can allways (at least: allmost allways...) push thing back down in the grade, but you will not be able to pull stuff "up" when it is developed "too dark".

As I mentioned, I think it is a huge issue with the current RAW graders, that they don't implement the SDK curve on RAW, and this relates directly to the DR issue and "once you have pushed it back".

Try this:
Expose an image under daylight so it just breaks at 800, while going all the way down to beautifull nuances in the lowlights (that you want to keep throughout for both offlines and online...)

Adjust the ISO/Exposure down so that you get textures in the highlights to 160-320 a bit depending on how much "on the edge" you've been exposing and color-temperature. (Check with linear @ 250 to check what is actually in the highlights...)

Ok. Now the highlights are fine, but your lowlights are crushed.

Try with the RAW grader to pick up the darks again.
Hopefully you won't manage, because that would mean that the grader digs into your metadata, and you really do not want that...
(Remember what blue-gain can do on the M f.ex.)

Way back when, I tried an early RAW grader, the primary dug into the metadata, while the secondaries worked for the grade. So that was what I had to do... RAW grading is not really what we want, we eant good RAW development and then grade.

Most likely - if your application works right You'll just get mud back up.

Then go to RCX/Clipfinder, set the exact same parameters and adjust the lowlights to be where you want them with the curve.

Export DPX.

THAT image will (if done right) be very flexible for grading.

We did a nice little experiment today with a unnamed RAW grader as to the importance of curves in the RAW developing stage, both for offlines and onlines. Unfortunately I am not at liberty to post a RAW frame, as it is a production image, but I can give you two crops to illustrate the point.

This was shot under daylight, so no red-channel clamping issues were present.

Because of the limit of 5 attachements I have to break this into 3 posts, but the examples are from the same image...

First: The highlights
1 @ 800 - redgamma (RG)
2 @ 800 redlogFilm (RLF)
3 @ 320 - linearLight
4 @ 250_RG
5 @ 250_RLF

I think you might agree that the highlights are working ok @ 250...

 

[Gunleik Groven]

Let's look at them lows...
6 @ 800 redgamma
7 @ 800 RLF

These are our reference values for how we want the image to look for offline and online...

8 @250 RG - no curve
9 @ 250 RLF - no curve

Only issue is that when the highlights look ok, the lowlights are crushed and sort of triggers the "black button" in many colorists......

 

[Gunleik Groven]

And finally, with a curve adjustment

10 @ 250_redgamma+curve
11 @ 250_rlf+curve

This approach was chosen for both offline and online. (Obviously not based on this single example).

For exposures stretching the DR at tungsten, 160 - 250 is used in post as the base ISO for both offlines and onlines + curve in the bottom.

And yes, we've run the different scenarios through grading sessions...

Actually the 250 RLF + curve came out smoother in the darks than the original 800 + redlogfilm in the grade, go figure...

So this is why I think it is quite an issue that the RAW graders or transcoding tools from f. ex. Quantel do not implement the SDK curve, is quite a huge issue...

BTW, the differences may not seem so big on the small jpg examples, so I encourage you to test these things out yourself to get an idea of wheter you think this might be wortwhile to implement or not. Looks pretty convincing on what we see here...

But again, these are only my 2 very limited cents on the subject...

 

[jake blackstone]

2. The "developer" needs to tweak the curves etc to maximise the exported file's DR and latitude.

3. The colourist needs to utilise as much of the DR as possible and properly compress it down (via curves etc as opposed to hard exposure adjustments) to the final medium's latitude limits.

Thanks to new tools, like Resolve, there is no need to split 2 and 3...

 

[Gunleik Groven]

Thanks to new tools, like Resolve, there is no need to split 2 and 3...

Sure? -

I didn't test Resolve today, but it's on my list for tomorrow.
So far the results have been pretty uniform with the RAW grading options.

 

[D Fuller]

Shoulder is added to the data afterwards, the data is recorded as is. DR is fixed, shoulder is not.

This is a key concept, and it is not at all film-like.

You can treat Red as a fixed film stock, after testing the optimal setup to get the most similar response, and you can also harvest the advantage of digital RAW and use variability of "shoulder" as a tool for optimizing usage of DR.

You cannot treat Red as a fixed film stock. (Here I hear Jim Morrison shouting "You cannot petition The Lord with Prayer..." but that's another story...) Not if you want to get the most out of it.

Red has a sensor with linear response. That is a very different thing from the response curve of any film stock, and to ignore that will cause you nothing but trouble. Highlights do not roll off naturally, Graeme's secret sauce rolls them off after exposure. That means that exposure is different from film in that you have to expose for dynamic range rather than middle grey or highlights or shadows. Those will come later.

I think this is what is so difficult for people coming to Red from film. In order to get the most from the camera, you have to know where the noise floor is and where you are going to lose the highlights. Middle gray just isn't as important as it is in film. You are going to add the shoulder and the knee in post. You are going to set middle gray there as well.

It's a lot like the Zone System, except that the curves are not present at exposure, they come later.

 

[Gunleik Groven]

Mr Fuller, I think you have it right...

ISO on sensors is a bad term because it creates a lot of confusion. It is better to decide what and how YOU want to work with the available latitude. This is even more so as long as you don't see a lot of the available stops on set currently.

 

[D Fuller]

ISO on sensors is a bad term because it creates a lot of confusion. It is better to decide what and how YOU want to work with the available latitude. This is even more so as long as you don't see a lot of the available stops on set currently.

This is an interesting point, Gunleik. I find many of the discussions about exposing Red (both here and elsewhere) to be a bit off the mark.

Those who come from the film world want Red to behave like film. It won't. Those who come from the video world want it to be like video. It is not.

It's possible to make it behave kind of like film by loading a look into the camera that will make the monitor show you an approximation of what the corrected image will look like, but if you rely on this, you'll give up some of what the digital "negative" can offer. Similarly, to expect Red to look "finished" like video does on set is to miss much of what can be gotten from Red.

As it has always been, the craft of the cinematographer is to capture an image that will deliver what was intended at the end of the color-grading/compositing line. With Red, that is not so different from what a cinematographer does with film, but the medium is different. Highlights clip at the sensor, they don't roll off. At the other end of the spectrum shadows get noisy and then there is no detail (this is not really much different from film: shadows get grainy and then there is only film base+fog), except that it happens in a linear fashion rather than along a "toe" curve. A curve, if you want one, is applied later. I really don't understand why this is so difficult for film DPs to understand.

I'll probably get flamed for saying this, but I think it's just laziness or lack of curiosity. If you want to just go out there with your light meter, take a reading and set a stop that's OK, but it's not the best you can do. You really do get a better image if you understand what will come later.

 

[jake blackstone]

Sure? -

Yes, I'm sure

 

[Gunleik Groven]

Yes, I'm sure

Good. Looking foreward to testing it -
I am not all that certain, but I am easily persuaded if I see that things work "right" -

Key question is how the signal comes out of the metadata-stage, what you can do with metadata before you start grading and what kind of signal you'rer grading from.

So far, the RAW graders I've been looking at have had issues in this first step, but I haven't done them all...

@ D Fuller

If for nothing else:
Digital ISO works the exact opposite way of how you would think with ISO on a filmstock.

Heighten your ISO and get MORE headroom in the top, lower it and get less. (I am not sure that saying the oposite about "protecting your lowlights" is entirely practically right, , though. Theoretically: Yes. But still in dayly use...)


Generally you would have chosen a high-iso stock when you have little light - shooting film, and low one for outdoor shooting at daytime.

I don't have a good alternative term comming up, though
It would lessen the confusion if there was some other easilly explainable term, though.

 

[Gunleik Groven]

Don't get around to it today... Next week,

But this is what I'd do.

Take some wide latitude images into resolve, tweek them so I keep the highlights with the tools in there and see what I can dig out of the lowlights.

Then I'd prepare some dpx's from the same frames from RCX/Clipfinder. Take them into resolve. Compare the results.

This is how I have seen that other SDK implementations have been not really what I want.

 

[Dylan Macleod CSC]

I think - with respect to RED at least - it should be called EI (Exposure Index) as opposed to ISO.

 

[Mark Wilkinson]

After following this thread for a while, I'm getting a sense of what Gunleik is trying to get at: Is there a way to game the system to get at all the data since the grading systems that go directly to the raw data have their own methodology - sometimes that don't seem to get all the data into the grading chain. And if you're going with the SDK, what are the settings that get you the most usable data. Do I have that right?

Re: EI - the push-me/pull-you of digital EI, noise floor and 12-bit raw ETTR is a mind bender. Especially with the low noise floor of the MX chip.

I'm going to shoot some tests... I'm not sure I'll learn anything, but it will make me feel better

 

[Hrvoje Simic]

You cannot treat Red as a fixed film stock. Not if you want to get the most out of it.

You said the exact same thing I did.

You can treat Red as a fixed film stock, after testing the optimal setup to get the most similar response, and you can also harvest the advantage of digital RAW and use variability of "shoulder" as a tool for optimizing usage of DR.

"Treating as film stock" as in deciding on the fixed contrast curve before shoot, unlike changing "shoulder" and "toe" afterwards. There are a lot of DPs who light according on what they see and can expect, with pre-set gamma curve, and don't want to fiddle with it later.

Red has a sensor with linear response. That is a very different thing from the response curve of any film stock, and to ignore that will cause you nothing but trouble. Highlights do not roll off naturally, Graeme's secret sauce rolls them off after exposure.

Again:

Red does have a minimal highlight compression to make highlights coming from a linear capturing medium more "film like", with avoiding hard clip you'd otherwise have, but this is not the typical shoulder you are used to with film. Apart from that, shoulder in this case is not a part of the medium, but processing i.e. viewing/development of captured data, which you can affect.

With changing ISO you decide where middle gray will be, and by that what "shoulder" and "toe" will later do to your data, depending on the priorities in your scene.

 

[Rob Ruffo]

The thing is, if done right, you can get a lot more than 13 stops out of MX - MX is 13 stops within a single ISO, and also without Neat Video which works extremely well when applied to 4K.

 

[Gunleik Groven]

The thing is, if done right, you can get a lot more than 13 stops out of MX - MX is 13 stops within a single ISO, and also without Neat Video which works extremely well when applied to 4K.

That would be nice to see an example of...

BTW, I don't do noise reduction. I work out hours of footage a day, not minutes... Don't need to either.
BUT
If I thought your statement was right, I'd probably consider canceling my Epic order.

Haven't done that.

 

[Pablo Villegas S.]

I'm shooting a TV series, and I've rated the camera at Iso 800 and exposed using Redcolor, then at the grading stage, using Apple color, I choose Red log and set the Iso back to 320, and start grading from there, I find out that it gives me more DR than starting grading from Redgamma at iso 800.

 

[Gunleik Groven]

I'm shooting a TV series, and I've rated the camera at Iso 800 and exposed using Redcolor, then at the grading stage, using Apple color, I choose Red log and set the Iso back to 320, and start grading from there, I find out that it gives me more DR than starting grading from Redgamma at iso 800.

That makes sense...