Converting Red to how Alexa handles saturation

[Hrvoje Simic]

Mixed lighting do screw up things for any camera, but Alexa seem to manage it much better. Whenever I've graded Alexa footage with mixed lighting I've never really had that much of a problem, but it's always a struggle with Dragon.

I've had those issues in post with many cameras, and yes even Alexa. That is normal. Different light sources cover the visible spectum differently and when mixed and reflected off the skin cause unnatural color response which very sensitive high precision cameras can capture and brain in real life ignores/"irons" interpretatively.

Alexa "chokes" colors in a specific way, which can make some of those factors be less visible but doesn't remove them and still does have its trainwreck color situations with mixed light scenarios. I've had to correct them.

If, theoretically, you could just emulate Alexa response from the start you'd condition and limit the performance of this sensor. If a different path is used it can still look excellent and natural/organic.

Here's two others. The original look is how Dragon looks through the RWG / Log3G10 with 709 Lut (it's a work in progress from Graeme) together with a node that blocks saturation at the shadow/highlight extremes. The tweaked is the same but with changed balance values for reds and yellows together with some experimentation with luma vs sat.

RWG / Log3G10 - 709LUT

RWG / Log3G10 - Tweak - 709LUT

First one seems more technically accurate, second one closer to your liking. Camera focused color scientist cannot chase the second one because that would necessitate reduction of image accuracy and more push towards a creative look.

 

[Dermot Shane]

Resolve 12.5.3
ACEScct
IDT = n/a
ODT = 709
node 1 = OpenFX
OFX = Resolve / color space transform
input gamma = project
input gamut = project
output gamut = Alexa
output gamma = linear

yea it's that simple to get to a decent starting point when matching the camera's, (Baselight offers similar tools) no LUT damage needed or wanted.. just clean maths

it will be warmer to start from, but easly corrected to match Alexa

 

[Anton Esteban]

Experimenting with Helium footage I've definitely seen that you can get better color separation than dragon footage. Out of the box, especially without the new color science, Helium doesn't look that different. But once you start pushing and pulling colors you start to see the difference. One way you can quickly see how much 'sauce' RED is adding to their sensor data is to switch the color science from RWG to CameraRGB. When you shoot the same scene with a Dragon and a Helium and then switch the color science from RWG to CameraRGB you start seeing how much the color science is pushing the color data into a specific direction. The more you see the change in color, that means its trying to fix a lot of problems under the hood which is not a good thing because its basically adding artifacts and noise. When you switch Helium to CameraRGB there isn't as much of an extreme change which means, at least to me, that its mapping color fairly accurately right out of the box. Or at least is doing a better job than the Dragon sensor.

 

[Sabovic Adis]

So, the color space drop-down list can distinguish between sensors?

 

[Anton Esteban]

From what I understand from Graeme on the RWG thread, he mentioned that RWG is close to CameraRGB but RWG is not sensor specific. Somehow RWG is supposed to be a universal RED color space that brings all RED cameras closer to the same starting point regardless of color space i.e. rec 709, rec 2020, etc.

 

[Sabovic Adis]

Maybe the third drop-down list is needed, indicating the sensor used. Ha! :biggrin:

 

[Aaron Lochert]

A good analogy for me to explain the difference between Alexa and RED cameras is similar to the difference between a good shotgun mic and an amazing shotgun mic. A good mic picks up all sound frequencies with clarity but an amazing mic will pick up all those sound frequencies but suppress the frequencies it thinks you don't want to hear

I've always held the opinion this is the key difference between RED and Arri's philosophies.

Arri has been in the motion picture business for almost a hundred years. They have a good understanding of what they think looks "good", which doesn't always line up with looking "accurate". They're providing you with the right canvas and the best paint. It's still your job to paint the painting, but they've started you off in the right direction.

RED is pushing for absolute technical advancement. They want to put every tool you could possibly need (and even some that you don't) at your fingertips. They're throwing you into the art supply store and you've gotta figure out what to buy and how to use it to make something artistic in the end. There's a higher risk to messing up because they've taken off the training wheels.

The same thing can be found in so many different industries. Flat, accurate speakers vs pleasing, often bass heavy ones, digital amplification vs tube amplification, etc. etc.

 

[Marc Wielage]

I was thinking about how to match or create some kind of Red to Alexa LUT or a node structure in Resolve to replicate how Alexa handles saturation and colors. I was wondering if there's someone who's been trying this out? What could be done to Red material in order to bring it closer to how Alexa looks as a starting point before grading.

If the end result you're trying to create is a Red camera that puts out an Alexa-like image, I don't think it's possible for the simple reason that the color science is much more than just LUTs. There are thousands of unpredictable, non-linear changes that happen at different exposure levels, with different hues and different amounts of saturation, that result in different pictures. I'll put it to you this way: you can't take two $1000 guitars of completely different manufacture and make them have the same tone and timbre. And you can't EQ two different $1000 speakers and make them sound exactly the same. They're different, period.

To me, I say just celebrate the differences, learn to exploit the positive attributes of the camera you're using at the moment, and don't worry about the camera you don't have. Red cameras can make great pictures, Alexas can make great pictures, and in truth both cameras can look horrible in the wrong hands. It is possible to match them to a point within a scene, but not 100% perfectly. I don't think there's a simple plug-in solution for lighting, exposure, lenses, and color correction, because each of those is a creative choice.

 

[Hrvoje Simic]

yea it's that simple to get to a decent starting point when matching the camera's, (Baselight offers similar tools) no LUT damage needed or wanted.. just clean maths

Your comment may be interpreted as if all LUTs are generally damaging. It depends how they are designed and used. In this context "clean math" is in many cases insufficient to target the specifics of the signal. Math is more limited in particulars, LUTs in flexibility.

 

[Michael Tiemann]

If the end result you're trying to create is a Red camera that puts out an Alexa-like image, I don't think it's possible for the simple reason that the color science is much more than just LUTs. There are thousands of unpredictable, non-linear changes that happen at different exposure levels, with different hues and different amounts of saturation, that result in different pictures. I'll put it to you this way: you can't take two $1000 guitars of completely different manufacture and make them have the same tone and timbre. And you can't EQ two different $1000 speakers and make them sound exactly the same. They're different, period.

To me, I say just celebrate the differences, learn to exploit the positive attributes of the camera you're using at the moment, and don't worry about the camera you don't have. Red cameras can make great pictures, Alexas can make great pictures, and in truth both cameras can look horrible in the wrong hands. It is possible to match them to a point within a scene, but not 100% perfectly. I don't think there's a simple plug-in solution for lighting, exposure, lenses, and color correction, because each of those is a creative choice.

Coming from a primarily audio background, the above analogy speaks to me! But better (because they are more in the same price range as ARRI vs. RED) might be Steinway vs. Yamaha concert grand pianos. Lang Lang gets his sound (and his endorsement money) from Steinway. Chick Corea gets his sound (and his endorsement money) from Yamaha. The classical repertoire does seem better represented on a Steinway, whereas the Yamaha is generally preferred for jazz. But...with a lot of knowledge of acoustics and mic technique, one can produce much better solutions than throwing only EQ (which is like a LUT) at the problem.

10 years ago the debate was all about film vs. digital, and many argued that whatever the convenience and economics of digital might be, if digital were not the literal equal of film in terms of output, then it was inferior and must be dismissed. Over time, many have come to embrace the fact that digital can be both more than film (in terms of image quality) and less (in terms of cost, turn-around time, size, weight, etc). The artists who have focused on making the best of this new medium have done better than those who have focused on merely emulating the past. I similarly believe that embracing whatever tools excite you most and do the best job in your hands is the best way to create new standards for the industry and the art. This will happen by embracing the limitations of those tools, and innovating, rather than imitating the competition.

 

[Sabovic Adis]

^
That's why today we have all the kick-mics, snare-mics, tom-mics, gutar-mics, guitar-caninet-mics, male-mics, female-mics...wonder if we ever get right-cheek cameras, right-cheek cameras, eyes-cameras, knees...:rofl:

And movies itself don't get any better. :biggrin:

 

[Michael Tiemann]

^
That's why today we have all the kick-mics, snare-mics, tom-mics, gutar-mics, guitar-caninet-mics, male-mics, female-mics...wonder if we ever get right-cheek cameras, right-cheek cameras, eyes-cameras, knees...:rofl:

And movies itself don't get any better. :biggrin:

Nah...that's why we have so many kinds of lights...and ways to use them.

 

[Sabovic Adis]

Nah...that's why we have so many kinds of lights...and ways to use them.

And the end credits get longer and longer...:rofl:

:thumbsup:

 

[Adrian Jebef]

If the end result you're trying to create is a Red camera that puts out an Alexa-like image, I don't think it's possible for the simple reason that the color science is much more than just LUTs. There are thousands of unpredictable, non-linear changes that happen at different exposure levels, with different hues and different amounts of saturation, that result in different pictures.

Marc's right. The cameras are different. They have different sensors with different dyes that capture different voltages of light that end up producing different images. "Color Science" is a tricky term. It implies that there are absolutes. Like mathematics. As if there is a real-world "color" equation where every color in front of the sensor must equal its own representation within the recorded raw file. As if the transform from raw to log or to ACES or to 709 or to whatever is done correctly then that color will appear because "color science". And if it doesn't then you need a different LUT or transform or the "new color science". It doesn't work this way. Cameras are both superior and inferior to the human eye in a multitude of ways. Different cameras produce different results because they are different. Though you can get different cameras to match each other fairly well it is usually done with a creative approach rather than with a scientific one.

As was posted earlier in this thread it is much simpler to decide what you favor about one camera over another (highlight rolloff, color saturation, etc) and then build a creative LUT that will apply these preferences than to try to build a mathematical transform that will work with two completely different sets of equations.

 

[jake blackstone]

Your comment may be interpreted as if all LUTs are generally damaging. It depends how they are designed and used. In this context "clean math" is in many cases insufficient to target the specifics of the signal. Math is more limited in particulars, LUTs in flexibility.

LUTs can be more damaging, than transforms, as they don't concatenate well and can introduce rounding errors, resulting in noise. That is one of the reasons transforms are the foundation for the modern color science moving forward, like ACES.

 

[Hrvoje Simic]

LUTs can be more damaging, than transforms, as they don't concatenate well and can introduce rounding errors, resulting in noise.

Can but don't have to.

As I said:

Mathematical transform (base value > function):
Advantage - flexibility
Limitation - particulars

LUT transformation (base value > result value):
Advantage - particulars
Limitation - flexibility

 

[Christoffer Glans]

If the end result you're trying to create is a Red camera that puts out an Alexa-like image, I don't think it's possible for the simple reason that the color science is much more than just LUTs. There are thousands of unpredictable, non-linear changes that happen at different exposure levels, with different hues and different amounts of saturation, that result in different pictures. I'll put it to you this way: you can't take two $1000 guitars of completely different manufacture and make them have the same tone and timbre. And you can't EQ two different $1000 speakers and make them sound exactly the same. They're different, period.

To me, I say just celebrate the differences, learn to exploit the positive attributes of the camera you're using at the moment, and don't worry about the camera you don't have. Red cameras can make great pictures, Alexas can make great pictures, and in truth both cameras can look horrible in the wrong hands. It is possible to match them to a point within a scene, but not 100% perfectly. I don't think there's a simple plug-in solution for lighting, exposure, lenses, and color correction, because each of those is a creative choice.

I don't think either that there's a simple solution, or else there would have been. However I'm not the kind of person who quit trying to push the boundaries and I know we can't get a perfect match between the two, but if a LUT or transform could come close that would use the best of both worlds, Red's resolution, image format, R3D files structure and in-camera proxies together with Alexas nicer color handling. Maybe we need a few LUT/transforms for this purpose; daylight, tungsten, mixed, low-key and so on, I dunno, but so far I've been able to grade it close to Alexa using only the curve transform tools in Resolve, but never really been able to get a match that works on many different lighting situations.

In pure daylight, sunny weather I've been able to come close, but where Red handles things the worst is duller low-key when there's not many highlights. In those cases everything in the image just melts together and get blurry with a lot of Red-noise creeping into everything when pushing the transforms.

 

[Björn Benckert]

Shoot some good starts with luminance and color patterns with both cameras same lens etc. Then in Log-c and Redlogfilm use something like match grade to create a CDL or 3Dlut to bring one camera to the other in log. Still this comes with a bit of difficulty as the cameras behave differently you would need to shoot them about the same in terms of where your clipping point is. But if thats done right then you got some what closer.

https://vimeo.com/94369569

 

[Hrvoje Simic]

I don't think either that there's a simple solution, or else there would have been. However I'm not the kind of person who quit trying to push the boundaries and I know we can't get a perfect match between the two, but if a LUT or transform could come close that would use the best of both worlds, Red's resolution, image format, R3D files structure and in-camera proxies together with Alexas nicer color handling. Maybe we need a few LUT/transforms for this purpose; daylight, tungsten, mixed, low-key and so on, I dunno, but so far I've been able to grade it close to Alexa using only the curve transform tools in Resolve, but never really been able to get a match that works on many different lighting situations.

Matching cameras per scene is one thing, matching them to behave the same throughout different circumstances is another. Expecting the latter is not realistic. Bacause their behavior is determined by a bunch of factors, nicely summed up by Adrian.

A fixed transform or LUT type of image transformation cannot lock variable targets. Period.

What can be done, however, is optimization of various aspects of image response, specifically per sensor, color space and gamma, with no intention of matching cameras but using scene values as a reference. And this is what I am doing and have been doing for the last year and a half with various cameras. Some parts of the method sharing same principles used in creation of IDTs and some I've had to develop myself.

 

[Björn Benckert]

Matching cameras per scene is one thing, matching them to behave the same throughout different circumstances is another. Expecting the latter is not realistic. Bacause their behavior is determined by a bunch of factors, nicely summed up by Adrian.

A fixed transform or LUT type of image transformation cannot lock variable targets. Period.

What can be done, however, is optimization of various aspects of image response, specifically per sensor, color space and gamma, with no intention of matching cameras but using scene values as a reference. And this is what I am doing and have been doing for the last year and a half with various cameras. Some parts of the method sharing same principles used in creation of IDTs and some I've had to develop myself.

So what you are saying is that if you got two cameras, Camera A with a wider DR and Camera B with a smaller DR. Then it´s not possible to create a calibration lut that forces camera A to look like camera B? Explain why it´s not possible. To me that would be the same as calibrating a screen or anything else, no? As long as the A camera has a bigger palette it should be possible, no?

To me weapon and even more so helium captures more than the alexa sensor, should not be too difficult to create a lut that if applies to a REDWIDEGAMUTRGB file makes it look very much like a ARRI LOG C file, and from there they could be treated the same.

But again it´s quite different in terms where the sensor clip so the helium cam would need atleast ND 0.3 to match in that regard, possibly even more. I think alexa color science is one thing, then when comparing them people tend to clip the RED as even if the DR is wider it does not really cover the DR of ARRI which clips later.