Dragon vs Alexa color accuracy?

[Hrvoje Simic]

I think I get what you mean Hrvoje, but the truth is also, that without transforming the image to a display-ready format (REC709 monitors don't care if it's linear to light, or tone-mapped for our sake) we can't make a subjective evaluation, which is what we're doing when looking at a side-by-side comparison. Therefore the most neutral subjective comparison is to transform from ACES using the most modern display transform maths available from SMPTE for REC709 monitors. This way we will see the color difference for a given scene between the cameras, given that ACES is a calibrated colourspace for both cameras.

The scopes were all very small and no-where near clipping. I'm planning to repeat the test with the DSCLabs chart I got at NAB this year to go further and do this test in more depth with some varied scenes.

Every time you test a Raw camera through its final color image you are testing camera + its transform(s), needed to get to that image.

 

[Christoffer Glans]

Every time you test a Raw camera through its final color image you are testing camera + its transform(s), needed to get to that image.

At this time I would wait for the new Red color science with a proper LUT transform. This way we can shoot something using RWG and LOG3G10 with a regular LUT to Rec709, then have another LUT for Alexa to transform into the same color space output. Only then, after processing should colors be judged. There are too many parameters that can screw up a true sensor color comparison and with the new color science it's time to check all of this again.

 

[Brett Harrison]

Every time you test a Raw camera through its final color image you are testing camera + its transform(s), needed to get to that image.

I didn't use the manufacturers' transforms - crucially, I used an independent transform and originated each camera image from the one colour space: ACES Primaries. I hope I'm expressing this clearly.

It will be interesting to know what benefits come with this new gamut, over the ACES primaries, which is also a wide gamut colour space. I'm guessing with the new colour space will come an improved colorimetry.

I personally like aspects of each, and I like Graeme's attitude that accuracy and beauty aren't mutually exclusive. I feel that an ARRI image already has a lot of non-linear fu performed on it whereas RED is more truthful and a richer starting point. There's always more dimension and potential in a RED image, IMO (and not just because of the greater resolution).

Looking at the images in the same colour space under tungsten, it seems that there's an orange cast to the ARRI footage.

At the same time I also agree that accuracy and beauty aren't always pals. Even the colour science literature often delineates between accuracy and what is preferable to the viewer.

 

[abreu-canedo]

Great conversation!

So, a few thinGs I think should be added to the conversation.

LUTs and transforms are different things. I always favor transforms since LUTs can sometimes map values outside of a given color space. LUts were generally an on-set tool, and then some coloriss found it useful as a way to finish for different color spaces. But generally speaking, top colorists are not using LUTs to cor correct.

Also, a lot of what the color matrix in each camera has to do in terms of subtracting and adding different colors via signal processing (colorimetry) has to do with how pure the dyes are in the color filter array. No dyes are completely pure and there is always some spectral overlap. For example, the blue dyes still let in some red and green wavelengths. So the signal processing chain has to subtract some of those reds and greens from the blue channel. But the more pure the dyes are, the wider the gamut of each camera is. In addition, the cleaner the signal is because it is not having to do as much signal amplification to counteract spectral overlap, thus reducing how much noise is generated in the process.

Now. For transparency, I will go ahead and state the in my personal and subjective experience I have found the Sony F65 to have the widest range of colors and this is due in part to the purity of the dyes used in its color filter array, the cleanliness of the signal to noise ratio within he sensor itself because instead of using aluminum conductors within the photosites (which Arri and red use), the Sony F65 uses copper...which is a better conductor. Then these things are compounded by the 16bit raw image. Even though Red processes images at 16bit, my analyses of the final image output is 14bit. I cannot be 100% certain about this since Red has not stated this officially, but there are ways of calculating this.

OK, disclaimer out of the way. You don't have to agree with my opinion, and I'm not brand loyal and the F65 has its downsides too. Nominally, weight. But it's still lighter than the comparable Alexa...by two pounds. Jeez, here I go on another tangent again. Back to the topic at hand. Sorry, I get a little self consious about my disposition toward the F65 since it's not a popular camera. Sorry about that!

In my experience, the red dragon sensor, even when processed with weapon boards, don't reproduce colors with the subtlety of Alexa, Varicam, or F65. Don't get me wrong, it's still a great image and it has its upsides. For example, the reds have much better resolution with their new 8k sensors, and dynamic range is beyond the old F65 and potentially even Alexa...although I find it tricky to test for that given different gamma curves. Also, the color reproduction differs depending on which OLPF is used on the Reds. Some OLPFs reproduce cleaner blues and reds, while compromising speed. I also found the the dark greens in the Red cameras to have some trouble with saturation. I found, in my limited experience, that the darker greens tend toward grey.

Sony F65 does have 8k horizontal photosites, but only about 17.6 active megapixels. The confusion about whether the F65 is actually an 8k sensor or not arises because the remaining 3k megapixels in the sensor are what Sony refers to inactive photosites. These are not photo sensitive photosites but instead are inactive to help minimize moire and softness that happens with certain types of CFA bayer patterns. So while these inactive "pixels" are not gathering light, they still go tribute to the quality of the image. So there you have it. The myth of Sony's 8k F65 explained. Now, I don't personally refer to the F65 as an 8k camera, since it is meant to output a 4k image, but it's 17.6 inactive megapixels do equate with around a 7k sensor if you don't consider inactive pixels to count. This may seem off topic, but photosite distribution relative to color filter arrays are still part of how a camera handles color.

I will conclude my subjective assessment by stating that while I will shoot the crap out of any red camera any day, and while I find all red products to be great at a what they do, I still find the colors a bit, for lack of better words, less subtle/complex than the Alexa, F65, or Varicam. Varicam has really true yeallows (very hard to do, even for Alexa), rich yet subtle cyans (also really hard to do), however, the Varicams do that Canon thing where they compress warm tones into a smaller range to reduce skintone blemishes. This can be flattering to the lay person, but I want complexity and subtlety...not compression. Alexas have cooler reds, and they level off saturation above 50% chrominance to make it behave like certain Kodak film stocks. F65 used to have problematic yellows and orangish reds with baked in codecs using Slog2, but in RAW, they deliver very accurate and subtle colors, a wide range of blues from turquoise, through cyan to deep indigos (very hard to do since blue wavelengths are much smaller and do not penetrate photosites with as much facility as larger and deeper wavelengths in warmer tones). Red still delivers beautiful work, and can be better suited for some types of works above the Alexa. But I my humble opinion, while the Red cameras are now the resolution kings with their 8k sensor, the Alexa still will render more complex tonality.

Either way, it's a great time to be a cinematographer. And I'll definitely be curious to see what Red does with the Helium sensor and what light-iron can squeeze out of the 8k VV red sensor!

 

[Mark Wuerthner]

Looking at the images in the same colour space under tungsten, it seems that there's an orange cast to the ARRI footage.

Arri states on their website that when using ACES the image will be more contrasty and will take on a warmer color cast. So I would question the concept of using ACES as a kind of "neutral" way to compare cameras as cameras respond differently to the ACES transform, though I definitely agree that it is a good starting point to try and match cameras.

https://www.arri.com/camera/alexa/learn/working_with_aces/

 

[abreu-canedo]

Looking at the images in the same colour space under tungsten, it seems that there's an orange cast to the ARRI footage.

Arri states on their website that when using ACES the image will be more contrasty and will take on a warmer color cast. So I would question the concept of using ACES as a kind of "neutral" way to compare cameras as cameras respond differently to the ACES transform, though I definitely agree that it is a good starting point to try and match cameras.

https://www.arri.com/camera/alexa/learn/working_with_aces/

I agree. I find ACES to be a bit stylized. But pleasing non the least. I suppose it is meant to be a starting point, not a final look.

 

[Hrvoje Simic]

LUTs and transforms are different things. I always favor transforms since LUTs can sometimes map values outside of a given color space.

Unless designed for a specific response, specific color input and target and used in a specific way.

LUts were generally an on-set tool, and then some coloriss found it useful as a way to finish for different color spaces.

LUT is a data set.
It can carry a lot of stuff.

F65 used to have problematic yellows and orangish reds with baked in codecs using Slog2,

That was a property of many Sony cameras. Slog2 is a gamma. Color mapping in S-Gamut was the culprit. Because it was the first attempt in that route and garbage in terms of color accuracy. S-Gamut3cine performance is greatly improved. Although still far from perfect.



Nice post, thanks for sharing your experience.

 

[abreu-canedo]

Unless designed for a specific response, specific color space as an input base and target and used in a specific way.



LUT is a data set.
It can carry a lot of stuff.


Nice post, thanks for sharing your experience.

Ha! For sure, LUTs are just look up tables. You're absolutely right. And yes, LUTs design for specific inputs and outputs will behave much more consistently than generic LUTs. Sorry, English is my second language, so I sometimes rush things. Plus, I'm not a color or cinematography master by any means, so I still have a lot to learn myself.

Thanks for the kind words, and the addition.

 

[Satsuki Murashige]

Great conversation!

So, a few thinGs I think should be added to the conversation.

LUTs and transforms are different things. I always favor transforms since LUTs can sometimes map values outside of a given color space. LUts were generally an on-set tool, and then some coloriss found it useful as a way to finish for different color spaces. But generally speaking, top colorists are not using LUTs to cor correct.

Also, a lot of what the color matrix in each camera has to do in terms of subtracting and adding different colors via signal processing (colorimetry) has to do with how pure the dyes are in the color filter array. No dyes are completely pure and there is always some spectral overlap. For example, the blue dyes still let in some red and green wavelengths. So the signal processing chain has to subtract some of those reds and greens from the blue channel. But the more pure the dyes are, the wider the gamut of each camera is. In addition, the cleaner the signal is because it is not having to do as much signal amplification to counteract spectral overlap, thus reducing how much noise is generated in the process.

Now. For transparency, I will go ahead and state the in my personal and subjective experience I have found the Sony F65 to have the widest range of colors and this is due in part to the purity of the dyes used in its color filter array, the cleanliness of the signal to noise ratio within he sensor itself because instead of using aluminum conductors within the photosites (which Arri and red use), the Sony F65 uses copper...which is a better conductor. Then these things are compounded by the 16bit raw image. Even though Red processes images at 16bit, my analyses of the final image output is 14bit. I cannot be 100% certain about this since Red has not stated this officially, but there are ways of calculating this.

OK, disclaimer out of the way. You don't have to agree with my opinion, and I'm not brand loyal and the F65 has its downsides too. Nominally, weight. But it's still lighter than the comparable Alexa...by two pounds. Jeez, here I go on another tangent again. Back to the topic at hand. Sorry, I get a little self consious about my disposition toward the F65 since it's not a popular camera. Sorry about that!

In my experience, the red dragon sensor, even when processed with weapon boards, don't reproduce colors with the subtlety of Alexa, Varicam, or F65. Don't get me wrong, it's still a great image and it has its upsides. For example, the reds have much better resolution with their new 8k sensors, and dynamic range is beyond the old F65 and potentially even Alexa...although I find it tricky to test for that given different gamma curves. Also, the color reproduction differs depending on which OLPF is used on the Reds. Some OLPFs reproduce cleaner blues and reds, while compromising speed. I also found the the dark greens in the Red cameras to have some trouble with saturation. I found, in my limited experience, that the darker greens tend toward grey.

Sony F65 does have 8k horizontal photosites, but only about 17.6 active megapixels. The confusion about whether the F65 is actually an 8k sensor or not arises because the remaining 3k megapixels in the sensor are what Sony refers to inactive photosites. These are not photo sensitive photosites but instead are inactive to help minimize moire and softness that happens with certain types of CFA bayer patterns. So while these inactive "pixels" are not gathering light, they still go tribute to the quality of the image. So there you have it. The myth of Sony's 8k F65 explained. Now, I don't personally refer to the F65 as an 8k camera, since it is meant to output a 4k image, but it's 17.6 inactive megapixels do equate with around a 7k sensor if you don't consider inactive pixels to count. This may seem off topic, but photosite distribution relative to color filter arrays are still part of how a camera handles color.

I will conclude my subjective assessment by stating that while I will shoot the crap out of any red camera any day, and while I find all red products to be great at a what they do, I still find the colors a bit, for lack of better words, less subtle/complex than the Alexa, F65, or Varicam. Varicam has really true yeallows (very hard to do, even for Alexa), rich yet subtle cyans (also really hard to do), however, the Varicams do that Canon thing where they compress warm tones into a smaller range to reduce skintone blemishes. This can be flattering to the lay person, but I want complexity and subtlety...not compression. Alexas have cooler reds, and they level off saturation above 50% chrominance to make it behave like certain Kodak film stocks. F65 used to have problematic yellows and orangish reds with baked in codecs using Slog2, but in RAW, they deliver very accurate and subtle colors, a wide range of blues from turquoise, through cyan to deep indigos (very hard to do since blue wavelengths are much smaller and do not penetrate photosites with as much facility as larger and deeper wavelengths in warmer tones). Red still delivers beautiful work, and can be better suited for some types of works above the Alexa. But I my humble opinion, while the Red cameras are now the resolution kings with their 8k sensor, the Alexa still will render more complex tonality.

Either way, it's a great time to be a cinematographer. And I'll definitely be curious to see what Red does with the Helium sensor and what light-iron can squeeze out of the 8k VV red sensor!

A very insightful and well-written post, thanks for sharing your thoughts! I've had many of the same aesthetic observations about these cameras and you've helped me understand some of the technical reasons for those. Much appreciated!

 

[Bob Gundu]

Very insightful post. Thank you :smile:

Great conversation!

So, a few thinGs I think should be added to the conversation.

LUTs and transforms are different things. I always favor transforms since LUTs can sometimes map values outside of a given color space. LUts were generally an on-set tool, and then some coloriss found it useful as a way to finish for different color spaces. But generally speaking, top colorists are not using LUTs to cor correct.

Also, a lot of what the color matrix in each camera has to do in terms of subtracting and adding different colors via signal processing (colorimetry) has to do with how pure the dyes are in the color filter array. No dyes are completely pure and there is always some spectral overlap. For example, the blue dyes still let in some red and green wavelengths. So the signal processing chain has to subtract some of those reds and greens from the blue channel. But the more pure the dyes are, the wider the gamut of each camera is. In addition, the cleaner the signal is because it is not having to do as much signal amplification to counteract spectral overlap, thus reducing how much noise is generated in the process.

Now. For transparency, I will go ahead and state the in my personal and subjective experience I have found the Sony F65 to have the widest range of colors and this is due in part to the purity of the dyes used in its color filter array, the cleanliness of the signal to noise ratio within he sensor itself because instead of using aluminum conductors within the photosites (which Arri and red use), the Sony F65 uses copper...which is a better conductor. Then these things are compounded by the 16bit raw image. Even though Red processes images at 16bit, my analyses of the final image output is 14bit. I cannot be 100% certain about this since Red has not stated this officially, but there are ways of calculating this.

OK, disclaimer out of the way. You don't have to agree with my opinion, and I'm not brand loyal and the F65 has its downsides too. Nominally, weight. But it's still lighter than the comparable Alexa...by two pounds. Jeez, here I go on another tangent again. Back to the topic at hand. Sorry, I get a little self consious about my disposition toward the F65 since it's not a popular camera. Sorry about that!

In my experience, the red dragon sensor, even when processed with weapon boards, don't reproduce colors with the subtlety of Alexa, Varicam, or F65. Don't get me wrong, it's still a great image and it has its upsides. For example, the reds have much better resolution with their new 8k sensors, and dynamic range is beyond the old F65 and potentially even Alexa...although I find it tricky to test for that given different gamma curves. Also, the color reproduction differs depending on which OLPF is used on the Reds. Some OLPFs reproduce cleaner blues and reds, while compromising speed. I also found the the dark greens in the Red cameras to have some trouble with saturation. I found, in my limited experience, that the darker greens tend toward grey.

Sony F65 does have 8k horizontal photosites, but only about 17.6 active megapixels. The confusion about whether the F65 is actually an 8k sensor or not arises because the remaining 3k megapixels in the sensor are what Sony refers to inactive photosites. These are not photo sensitive photosites but instead are inactive to help minimize moire and softness that happens with certain types of CFA bayer patterns. So while these inactive "pixels" are not gathering light, they still go tribute to the quality of the image. So there you have it. The myth of Sony's 8k F65 explained. Now, I don't personally refer to the F65 as an 8k camera, since it is meant to output a 4k image, but it's 17.6 inactive megapixels do equate with around a 7k sensor if you don't consider inactive pixels to count. This may seem off topic, but photosite distribution relative to color filter arrays are still part of how a camera handles color.

I will conclude my subjective assessment by stating that while I will shoot the crap out of any red camera any day, and while I find all red products to be great at a what they do, I still find the colors a bit, for lack of better words, less subtle/complex than the Alexa, F65, or Varicam. Varicam has really true yeallows (very hard to do, even for Alexa), rich yet subtle cyans (also really hard to do), however, the Varicams do that Canon thing where they compress warm tones into a smaller range to reduce skintone blemishes. This can be flattering to the lay person, but I want complexity and subtlety...not compression. Alexas have cooler reds, and they level off saturation above 50% chrominance to make it behave like certain Kodak film stocks. F65 used to have problematic yellows and orangish reds with baked in codecs using Slog2, but in RAW, they deliver very accurate and subtle colors, a wide range of blues from turquoise, through cyan to deep indigos (very hard to do since blue wavelengths are much smaller and do not penetrate photosites with as much facility as larger and deeper wavelengths in warmer tones). Red still delivers beautiful work, and can be better suited for some types of works above the Alexa. But I my humble opinion, while the Red cameras are now the resolution kings with their 8k sensor, the Alexa still will render more complex tonality.

Either way, it's a great time to be a cinematographer. And I'll definitely be curious to see what Red does with the Helium sensor and what light-iron can squeeze out of the 8k VV red sensor!

 

[Curtis boggs]

scarlet x color

scarlet x color

Hello,

I don't want to throw this thread off track, I'm learning a bunch.
BUT, I just got a Scarlet X and waiting on a Scarlet W.

In Redcine-x, what color space and gamma setting should I be using under various circumstances? Redcolor2, .. etc.
If someone could point me to some good threads I'd be great, .. I'm alittle confused with that I found in my search.

Thank you,

Curtis

 

[Brett Harrison]

Looking at the images in the same colour space under tungsten, it seems that there's an orange cast to the ARRI footage.

Arri states on their website that when using ACES the image will be more contrasty and will take on a warmer color cast. So I would question the concept of using ACES as a kind of "neutral" way to compare cameras as cameras respond differently to the ACES transform, though I definitely agree that it is a good starting point to try and match cameras.

https://www.arri.com/camera/alexa/learn/working_with_aces/

That link didn't work for me but I found this:

https://www.arri.com/camera/alexa_mini/learn/working_with_aces/

There's no mention of an orange colour cast or any modification of the image. I can confirm this by going from ACES to LogC AWG and originating with LogC - they're the same. This is unsurprising since ACES Primaries is just a colour space with a linear transform - if you go into it then out of it, there's no modification that takes place. Perhaps what you read was referring to ACEScc but I doubt that, ACES wouldn't be taken seriously if it were imparting a colour cast.

 

[Hal Long]

So many great points in this thread. But getting to the OP, if it's purely a question of color accuracy, the Varicam is tough to beat. The form is smaller than F65 and it's pretty damned accurate. Color separation and skin tones look great. The Alexa is very accurate but the Alexa look seems to be the selling point, not strictly accuracy. But I'm still very happy when shooting on Dragon.

 

[Mark Wuerthner]

That link didn't work for me but I found this:

https://www.arri.com/camera/alexa_mini/learn/working_with_aces/

There's no mention of an orange colour cast or any modification of the image. I can confirm this by going from ACES to LogC AWG and originating with LogC - they're the same. This is unsurprising since ACES Primaries is just a colour space with a linear transform - if you go into it then out of it, there's no modification that takes place. Perhaps what you read was referring to ACEScc but I doubt that, ACES wouldn't be taken seriously if it were imparting a colour cast.

Sorry, I didn't do the quote thing properly, the "orange color cast" line in my previous post is a quote from your posting. Your link to Arri ACES is good and if you go to the FAQ section Arri states:
Q: Does ACES have different tonescale response than ARRI cameras’ default processing?

A: Yes. It’s a bit more contrasty, which can bring out more detail.

Arri also says:
Q: Are ACES fleshtones better or worse than ARRI’s fleshtones?

A: They are different. The ACES skin tone rendering, when applied to ARRI camera output, tends to be "tawny", that is to say, for ARRI camera output ACES will render skin a bit more yellow. Whether this is a welcome or unwelcome change will depend on the subject and the context.

 

[Hrvoje Simic]

Sorry, I didn't do the quote thing properly, the "orange color cast" line in my previous post is a quote from your posting. Your link to Arri ACES is good and if you go to the FAQ section Arri states:
Q: Does ACES have different tonescale response than ARRI cameras’ default processing?

A: Yes. It’s a bit more contrasty, which can bring out more detail.

Arri also says:
Q: Are ACES fleshtones better or worse than ARRI’s fleshtones?

A: They are different. The ACES skin tone rendering, when applied to ARRI camera output, tends to be "tawny", that is to say, for ARRI camera output ACES will render skin a bit more yellow. Whether this is a welcome or unwelcome change will depend on the subject and the context.

And you can also check on that page what ACES route does with saturated highlights along with that "a bit more yellow".

 

[David Mullen ASC]

Interesting because I find the Alexa Rec.709 output on the camera to be on the yellow side.

 

[Brett Harrison]

Thanks for pointing out the FAQ part Mark. I noticed when reading it it was referring to the display rendering transform and the characteristic curve, additional contrast and the tawny warm/cool bias.

I could see from the flow chart that both images start from the camera negative image (linear gamma 1.0) and I knew from my workflow that I was transforming to ACES Primaries (linear gamma 1.0) and then transforming from that to go to the Mistika implementation of BT1886 transform - this means that it's a 'clear window' between camera negative (the starting point for ACES images out of .ARI files, as stated in the FAQ) and the BT1886 OETF rendering.

The BT.1886 transform is basically an updated version of Rec.709 but with better maths for modern flat panel displays (as well as some other clever colour science worked in there)

https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.1886-0-201103-I!!PDF-E.pdf

EDIT: here's a better reference for OETF https://www.smpte.org/sites/default/files/2014-05-06-EOTF-Miller-1-2-handout.pdf

REDCineX has an option for BT.1886 if you want to compare known footage at home against RG4 (another 709 type transform).

Essentially what the FAQ is referring to when it talks about ACES rendering is the display transform, i.e. ACES-particular way of taking a linear image, or a log image, and rendering it to a flat panel or CRT display in a gamma 2.4 contrast curve (i.e. not linear) i.e. Rec.709/BT.1886/RGx

I jumped into Mistika to confirm this visually and made a 4-way comparison. For good measure I started from Arri Wide Gamut (AWG), as well as ACES, when debayering from the .ARI file, both of which were then transformed to BT.1886.

TOP right and
TOP left are BT.1886 originating from ACES and AWG - can you tell which is which?

BOTTOM left is the ACES RRT, starting from ACES debayer option
BOTTOM right is the ARRI Rec709 k1s1 (default in camera rendering) LUT applied to LogC

Note the effects referred to in the FAQ.

Note how similar the the ARRI LUT and the Mistika BT.1886 images are.

EDIT:

So just as a follow up to the question of ACES being a questionable foundation for comparing ALEXA to Dragon, the above test shows that in this context it imparts no modifications, and additionally it demonstrates that Mistika's display transform (used as a neutral point of comparison) is actually very similar to the ARRI proprietary display transform.

It also demonstrates that using ACES linear doesn't impart the tawny, contrasty effects, which are only a result of the RRT, not of linear ACES images (i.e. ACES Primaries, LogC, AWG)

Going back to the original question of the thread, which is how ARRI and RED differ under tungsten light, we can see - under the same image transform from RAW camera info - the colour differences and the subtle rendering differences clearly in the original test image I posted.

And we know now from this test that the Mistika transform from ACES to BT.1886 is actually a good 'neighbourhood' colorspace within which to hold such a comparison (due to the similarity).

Whether or not ACES or BT.1886 looks different slightly from the usual manufacturers' transforms is neither here nor there because when they're going through the same processing pipeline, you get to view the degree of difference between the RAW info, which is always the starting point no matter your final workflow.

 

[Mike Nagel]

Great conversation!

So, a few thinGs I think should be added to the conversation.

LUTs and transforms are different things. I always favor transforms since LUTs can sometimes map values outside of a given color space. LUts were generally an on-set tool, and then some coloriss found it useful as a way to finish for different color spaces. But generally speaking, top colorists are not using LUTs to cor correct.

Also, a lot of what the color matrix in each camera has to do in terms of subtracting and adding different colors via signal processing (colorimetry) has to do with how pure the dyes are in the color filter array. No dyes are completely pure and there is always some spectral overlap. For example, the blue dyes still let in some red and green wavelengths. So the signal processing chain has to subtract some of those reds and greens from the blue channel. But the more pure the dyes are, the wider the gamut of each camera is. In addition, the cleaner the signal is because it is not having to do as much signal amplification to counteract spectral overlap, thus reducing how much noise is generated in the process.

Now. For transparency, I will go ahead and state the in my personal and subjective experience I have found the Sony F65 to have the widest range of colors and this is due in part to the purity of the dyes used in its color filter array, the cleanliness of the signal to noise ratio within he sensor itself because instead of using aluminum conductors within the photosites (which Arri and red use), the Sony F65 uses copper...which is a better conductor. Then these things are compounded by the 16bit raw image. Even though Red processes images at 16bit, my analyses of the final image output is 14bit. I cannot be 100% certain about this since Red has not stated this officially, but there are ways of calculating this.

OK, disclaimer out of the way. You don't have to agree with my opinion, and I'm not brand loyal and the F65 has its downsides too. Nominally, weight. But it's still lighter than the comparable Alexa...by two pounds. Jeez, here I go on another tangent again. Back to the topic at hand. Sorry, I get a little self consious about my disposition toward the F65 since it's not a popular camera. Sorry about that!

In my experience, the red dragon sensor, even when processed with weapon boards, don't reproduce colors with the subtlety of Alexa, Varicam, or F65. Don't get me wrong, it's still a great image and it has its upsides. For example, the reds have much better resolution with their new 8k sensors, and dynamic range is beyond the old F65 and potentially even Alexa...although I find it tricky to test for that given different gamma curves. Also, the color reproduction differs depending on which OLPF is used on the Reds. Some OLPFs reproduce cleaner blues and reds, while compromising speed. I also found the the dark greens in the Red cameras to have some trouble with saturation. I found, in my limited experience, that the darker greens tend toward grey.

Sony F65 does have 8k horizontal photosites, but only about 17.6 active megapixels. The confusion about whether the F65 is actually an 8k sensor or not arises because the remaining 3k megapixels in the sensor are what Sony refers to inactive photosites. These are not photo sensitive photosites but instead are inactive to help minimize moire and softness that happens with certain types of CFA bayer patterns. So while these inactive "pixels" are not gathering light, they still go tribute to the quality of the image. So there you have it. The myth of Sony's 8k F65 explained. Now, I don't personally refer to the F65 as an 8k camera, since it is meant to output a 4k image, but it's 17.6 inactive megapixels do equate with around a 7k sensor if you don't consider inactive pixels to count. This may seem off topic, but photosite distribution relative to color filter arrays are still part of how a camera handles color.

I will conclude my subjective assessment by stating that while I will shoot the crap out of any red camera any day, and while I find all red products to be great at a what they do, I still find the colors a bit, for lack of better words, less subtle/complex than the Alexa, F65, or Varicam. Varicam has really true yeallows (very hard to do, even for Alexa), rich yet subtle cyans (also really hard to do), however, the Varicams do that Canon thing where they compress warm tones into a smaller range to reduce skintone blemishes. This can be flattering to the lay person, but I want complexity and subtlety...not compression. Alexas have cooler reds, and they level off saturation above 50% chrominance to make it behave like certain Kodak film stocks. F65 used to have problematic yellows and orangish reds with baked in codecs using Slog2, but in RAW, they deliver very accurate and subtle colors, a wide range of blues from turquoise, through cyan to deep indigos (very hard to do since blue wavelengths are much smaller and do not penetrate photosites with as much facility as larger and deeper wavelengths in warmer tones). Red still delivers beautiful work, and can be better suited for some types of works above the Alexa. But I my humble opinion, while the Red cameras are now the resolution kings with their 8k sensor, the Alexa still will render more complex tonality.

Either way, it's a great time to be a cinematographer. And I'll definitely be curious to see what Red does with the Helium sensor and what light-iron can squeeze out of the 8k VV red sensor!

LUTs are color transforms in 1D/3D space but are limited to resolution (--> cube size) and obviously to the interpolation ability of the host application, which nowadays pretty much all apps do well. The vector (transforms) vs. bitmap (LUTs) analogy comes to mind. For color grading, (if needed) I prefer pure transforms but for technical application 3D LUTs are needed, as no color transform can do certain jobs.

I agree wit the color assessment and also find F65 colors in raw to be very very good, but Arri gets much more usage so we don't get to see F65's beauty that often...

Re F65 8K "resolution": In my understanding F65 square pixel raster is rotated 45° off kilter, so in a traditional horizontal assessment u come to something like 5.7K horizontally. Not 8K, not 7K.

U come to "8K" if u count two staggered rows (45° angled) together ;-)

 

[Christoffer Glans]

I agree wit the color assessment and also find F65 colors in raw to be very very good, but Arri gets much more usage so we don't get to see F65's beauty that often...

Re F65 8K "resolution": In my understanding F65 square pixel raster is rotated 45° off kilter, so in a traditional horizontal assessment u come to something like 5.7K horizontally. Not 8K, not 7K.

U come to "8K" if u count two staggered rows (45° angled) together ;-)

Both Oblivion and Lucy are two examples that look very very good. Super clean without any unnatural colors, especially skin tones.

Even though measured resolution differs, I think the debayer and scaledown makes things a bit different on the F65 with it's rotated pixel raster; a bit more organic.

 

[Mike Nagel]

Both Oblivion and Lucy are two examples that look very very good. Super clean without any unnatural colors, especially skin tones.

Even though measured resolution differs, I think the debayer and scaledown makes things a bit different on the F65 with it's rotated pixel raster; a bit more organic.

Lucy... meh...

Oblivion looked great all around.

AFTER EARTH (horrible movie) looked great as well... look at the forest and the foliage... color rendition is superb