Interesting Breakdown Of Arri Colour Science

[BTM_Pix]

I thought the title was slightly clickbaity but I’m glad I clicked on this one.

There are a lot of “Arri look on your xyz camera” videos but I think this is a far more interesting take on it.

The question is are you doomed to never match it because it is baked in (not least the IR cut filter) and it can’t be reverse engineered in post?

Can anyone who has bought the Arri license for their supported Panasonic camera chime in with their thoughts?

 

[kye]

Very interesting.

I've been looking into this for a long time and hadn't heard some of these insights.  Very useful.

One thing he got wrong, at least for v3 of their colour science is that there isn't anything luma-specific done to the image inside the camera, it's all done in the LUT.  The evidence for this is when people do under/over tests, where the camera is deliberately under and over exposed, when you correct the image in post to the correct exposure the colours are all the same.  If they were warming the highlights or cooling the shadows in-camera then when you overexposed and brought the exposure down in post then that warmth would be baked-in and your mid-tones would be warm (or they'd be cooler if you underexposed) but that's not what we see.

It's also no secret that they compress the skin-tone hue range and also tend to skew yellow, especially compared to Canon which skews magenta/red.

The IR-cut filter letting in a bit of far-red so the skin-tones get that scattering is interesting.  To a certain extent it might be 'recoverable' in post (ie, perhaps we can guess what might have been there based on what info we do have).  Perhaps the key aspect of any such attempts would be to blur this new channel once it's been simulated, as this is the information coming from deeper in the skin and is scattered a bit.
It's been hinted at that part of the 'Cooke Look' was that they used materials that slightly blurred light at a range of frequencies within skin tones, so the lenses sort-of worked like a skin-hue-only diffusion filter.

Potentially anyone with a full-spectrum camera (OG BMPCC BMMC anyone?) could seek out IR-cut filters that are designed to let in more far-red.  Potentially even people with IR-cut filters on their sensors could get strong ND filters to boost the relative proportion of IR coming into the camera.  I'm guessing that the right amount of the right ND might do it - have enough ND that the IR is boosted but not so much that the blacks become polluted.  It's funny he mentioned that the iPhone has a strong IR-cut filter - I was testing my iPhone 17 Pro the other day with lots of ND (because it doesn't have an iris!) and I got a ton of IR pollution, including getting a non-trivial amount of it when using the same vND I use on my GH7 which I've never seen any IR pollution on.  So the iPhone must have less IR filtration than the GH7.

It's worth adding that a lot of these things are also accomplished by film emulation.

[ND64]

22 hours ago, kye said:

It's been hinted at that part of the 'Cooke Look' was that they used materials that slightly blurred light at a range of frequencies within skin tones, so the lenses sort-of worked like a skin-hue-only diffusion filter.

Some of Nikon's portrait F mount lenses do the same. They deliberately let the red light doesn't hit the same plane as blue/green. As like not-much-corrected glasses of the old days. So red is a bit out of focus. Very very tiny amount. Tho that amount changes with focus distance. They usually optimize for portrait distances. They abandoned that approach with Z mount lenses. 

[kye]

11 hours ago, ND64 said:

Some of Nikon's portrait F mount lenses do the same. They deliberately let the red light doesn't hit the same plane as blue/green. As like not-much-corrected glasses of the old days. So red is a bit out of focus. Very very tiny amount. Tho that amount changes with focus distance. They usually optimize for portrait distances. They abandoned that approach with Z mount lenses. 

How interesting.

I guess something like that would be very difficult to determine for cinematographers etc, but if you're a lens manufacturer with all the test equipment then buying all your competitors lenses and then testing them and taking them apart would be pretty standard operating procedure.

[Aussie Ash]

Looking at the bigger picture the current lead that ARRI has in this field can be eroded over the next few years by other camera manufacturers.As an example

Black Magic is already up to Generation 6 color science and each generation is a step better than the last

[eatstoomuchjam]

48 minutes ago, Aussie Ash said:

Black Magic is already up to Generation 6 color science and each generation is a step better than the last

Not quite.  It's announced, but not available yet, last I checked - unless the latest Resolve 21 beta included it.

Though if there's really something to inclusion of additional near-infrared wavelengths making the subtle difference for skin tones is true and if BMD's ir cut filter removes that wavelength, there would always be at least some difference.

That said, a lot of the way that guy describes things remind me of someone explaining why they think their $4,000 toslink cable is giving richer midtones than someone else's $30 toslink cable.  There seems to be a mix of some genuine understanding of things mixed in with pseudo-scientific bullshit.

Regardless, even if were all 100% true and based in pure science, a decent colorist is a big equalizer for a bunch of differences in the SOOC image.  😅

[Simon Young]

On 5/26/2026 at 11:25 AM, BTM_Pix said:

I thought the title was slightly clickbaity but I’m glad I clicked on this one.

There are a lot of “Arri look on your xyz camera” videos but I think this is a far more interesting take on it.

The question is are you doomed to never match it because it is baked in (not least the IR cut filter) and it can’t be reverse engineered in post?

Can anyone who has bought the Arri license for their supported Panasonic camera chime in with their thoughts?

 

It’s not even close.

 

 

[kye]

Yeah, there are differences for sure.  Some situations show larger differences than others.

One theme is that the Panny versions are much more saturated in skin tones than the Alexa.  Coming from the kings of colour science, this has got to be deliberate, so even ARRI aren't trying to match the Alexa.  

I think that instead of asking "are they identical", the better question would be "does this impart some magic".

I mean, everyone agrees that ARRI has the magic but their cameras don't even match each other between sensors, so it makes no sense to have a higher bar for the Panny version than we'd apply for ARRI themselves.

My impression was that the ARRILogC profile and the ARRI709 LUT is the most like the Alexa, with the V-Log -> CST -> ARRI709 LUT not being as good.  This makes sense as whatever colour / gamma secrets ARRI applies in-camera can be put into that profile as it's customised to the Panny sensors and also safely locked inside the camera away from prying eyes (unlike anything they put in their LUT).

However, the Alexa also responds differently in the spatial dimension, with the far-red response including a more spatially distributed response from skin tones.  We also know that Alexas process the image spatially in-camera due to their texture options and processing.  Who knows what is going on with texture in there.  IMHO the texture of Alexa images is right up there in importance as their colour response.

None of this includes temporal aspects either.  While I struggle to think of what processing might be occurring in-camera between frames, there might be some (we can't tell), and that's beyond the possibility that the hardware itself has some sort of secret properties that contribute to the image.

FDTimes did an entire episode on the Alexa 35, with interviews of over a dozen people and 100+ pages:
https://www.fdtimes.com/pdfs/free/115FDTimes-June2022-2.04-150.pdf

Here is the image pipeline in the Alexa 35 (page 59):

To give some idea about how stunningly out of our depth basically everyone on the internet is who talks about this stuff, starting on page 116, Dr. Tamara Seybold talks about Textures..

"For example, the debayering already needed to obtain the full color image doesn’t only generate RGB values but also influences the perceived sharpness and grain rendering. And many more steps influence the clarity and grain that are important aspects of the texture of an image. So we, in the image science team, pushed hard to obtain the best results by really optimizing each and every step in the image processing pipeline, not only for the best color rendition but also for the best texture, as we call it. We did that in a holistic way, optimizing steps in the beginning of the pipeline together with later steps so that the overall result would be best. At some point, this came down to having more than 30 parameters that we had to optimize together—a huge amount. We specifically had to build a small “texture grading machine” to be able to optimize all these parameters together."
(emphasis added)

I don't know about anyone else here, but I would struggle to even list 30 parameters, let alone identify all the parameters, isolate the 30 that matter, then find the sweet-spot (or sweet spots) in a 30-dimensional space.

This is regarding the Alexa 35, but I remember reading in there somewhere that the innovation of the Textures feature is that you can choose different profiles on the new camera, whereas on the old ones you only had the one, and that on the previous models they had chosen a texture configuration that was their best attempt at a one-size-fits-all.  So the inference was that the previous cameras were also doing this kind of processing.

By implementing their colour science inside the camera, they could be doing all sorts of stuff.  They could have things that analyse the image and then apply different treatments depending on the scene the camera was capturing.  They certainly have a team capable enough and a camera with enough processing power to have a dozen, or a hundred, or a thousand, LUTs or algorithms inside it and be changing these things based on context or WB setting or sensor temperature or whatever the hell else they found was useful.

The sheer depth of knowledge that has gone into their image science is incredible.  In 2009, Glenn Kennel joined ARRI as their CTO, which was a new position at that time, and in 2010 he was promoted to President and CEO.  Glenn had previously worked for Kodak from 1980, and worked on various things that involved the gradual digitisation of the pipeline, including things like telecines and film scanners etc.  My understanding is that his contributions at ARRI were pivotal for the development of the Alexa, which was the first digital camera to gain wide acceptance within the industry and did so due its film-like response.

A bit of searching revealed some interesting discussions we already had during lockdowns.. 

 

 

[ND64]

10 hours ago, kye said:

Here is the image pipeline in the Alexa 35 (page 59):

Interesting that WB is placed before De-bayer. Photographers call it "white balance scaling" and prefer it not to be done by manufacturer, since its digitally boosting channels, which is irreversible process. After all its not like all things they're doing is science. Some decisions are just an engineer thinking its the best compromise.