From Cobalt, Leica M vs world.
p.2 #1From Cobalt, Leica M vs world.
Well if all M shooter have to add this variable into each shot taken. Then picture would pass them by... 😆
Sep 05, 2026 at 02:23 PM
p.2 #2From Cobalt, Leica M vs world.
maybe i don't have enough wrinkles in my brain to understand this ranking, but it absolutely makes no sense to me. for one, SL3/Q3/Q343 share the exact same sensor, how come they are ranked differently?
that M10-R "overall score" seems pulled out of thin air. how did you arrive at this low score? Yes i read this but it's just a bunch of word salad and still doesn't explain the huge disparity of score.
also: is this an ad?
Edited on Sep 05, 2026 at 03:29 PM · View previous versions
that M10-R "overall score" seems pulled out of thin air. how did you arrive at this low score? Yes i read this but it's just a bunch of word salad and still doesn't explain the huge disparity of score.
also: is this an ad?
Edited on Sep 05, 2026 at 03:29 PM · View previous versions
Sep 05, 2026 at 03:10 PM
p.2 #3From Cobalt, Leica M vs world.
@Ulysseita - I am trying to understand the ranking on your web site
https://www.cobalt-image.com/spectre-laboratory-measurements/
Measured color accuracy
Leica M11- overall hardware 62, with Spectre profile 95
Nikon Z8 - overall hardware 89, with Spectre profile 93.99
So - with Spectre M11 > Z8? I believe you said profiles can only go so far earlier in this thread, how can the M11 be best in class and beat Z8 with Spectre?
https://www.cobalt-image.com/spectre-laboratory-measurements/
Measured color accuracy
Leica M11- overall hardware 62, with Spectre profile 95
Nikon Z8 - overall hardware 89, with Spectre profile 93.99
So - with Spectre M11 > Z8? I believe you said profiles can only go so far earlier in this thread, how can the M11 be best in class and beat Z8 with Spectre?
Sep 05, 2026 at 03:23 PM
p.2 #4From Cobalt, Leica M vs world.
This is such an area no one wants to enter. It is not unimportant but there is no absolute answer. It is not any manufacture doesn’t care but there is no clear right or wrong answer to it. And even human eye have different LMS response and personal perceived color have different build in preference. Considering all lens coating and even within the same manufacture, color cast can be different.
It is a complicated topic. I am sure every manufacture pay lots of attention on CFA design and their signal pipeline to optimize this but few publicly put claim that ours is better because there is no universal standard or variables are just too much to define the score.
Even adobe profile was trying to match all camera with roughly the same baseline by understand individual camera CFA response to match CIE XYZ for a given illumination, a standard try to mathematically define human perceived color for a given condition. This provide a baseline for further editing.
I assume the score is try to see how far away camera as is from CIE XYZ standard (which itself can be deviate to individual person? ) but do you use any profile before the comparison? Or there is raw reading out of camera? Not an expert here and not sure how the score works.
Given tone curve, white balance, lens color, human perception difference, it is almost impossible to get a universal standard.
In my view, To put it simple, if you are happy with certain profile product most of time, under most of shooting conditions, feel free to use it.
If you are the one always want to do some editing on the images to fit your vision in term of color, contrast, tone, then I am not sure the profile is needed. For example, using FUJI camera in adobe LR, the color, tone difference are so huge for each profile chosen make color a moot point.
My 2C.
It is a complicated topic. I am sure every manufacture pay lots of attention on CFA design and their signal pipeline to optimize this but few publicly put claim that ours is better because there is no universal standard or variables are just too much to define the score.
Even adobe profile was trying to match all camera with roughly the same baseline by understand individual camera CFA response to match CIE XYZ for a given illumination, a standard try to mathematically define human perceived color for a given condition. This provide a baseline for further editing.
I assume the score is try to see how far away camera as is from CIE XYZ standard (which itself can be deviate to individual person? ) but do you use any profile before the comparison? Or there is raw reading out of camera? Not an expert here and not sure how the score works.
Given tone curve, white balance, lens color, human perception difference, it is almost impossible to get a universal standard.
In my view, To put it simple, if you are happy with certain profile product most of time, under most of shooting conditions, feel free to use it.
If you are the one always want to do some editing on the images to fit your vision in term of color, contrast, tone, then I am not sure the profile is needed. For example, using FUJI camera in adobe LR, the color, tone difference are so huge for each profile chosen make color a moot point.
My 2C.
Sep 05, 2026 at 07:37 PM
p.2 #5From Cobalt, Leica M vs world.
Ne314satel wrote:
Guys, I can't smoke that much..
I understand, it's Friday.
It can get heady in here
Guys, I can't smoke that much..
I understand, it's Friday.
It can get heady in here
Sep 05, 2026 at 08:36 PM
p.2 #6From Cobalt, Leica M vs world.
Using the spectre profile completely transformed my experience with the M10-R. Many people shoot in Auto WB, bring the files into Lightroom, and try to fix things by switching the profile to Adobe Standard with a "Daylight" WB setting...and honestly, it just doesn't look great.
However, if you use that same Daylight WB in Lightroom and pair it with the Spectre profile instead, everything looks correct and accurate, and the improvement is not subtle. I was really hoping Cobalt would develop this for the M240 series, especially for the M-D since it's locked to Auto WB. I'm so glad this option is available now.
However, if you use that same Daylight WB in Lightroom and pair it with the Spectre profile instead, everything looks correct and accurate, and the improvement is not subtle. I was really hoping Cobalt would develop this for the M240 series, especially for the M-D since it's locked to Auto WB. I'm so glad this option is available now.
Sep 05, 2026 at 09:10 PM
p.2 #7From Cobalt, Leica M vs world.
@Ulysseita: I'm a happy customer who uses Spectre for both A7RV and Nikon Z6III. I'm very happy that your product exists.
One thing I've been curious about. What would be the difference between Hasselblad HNCS and Spectre. They sound similar in many ways?
One thing I've been curious about. What would be the difference between Hasselblad HNCS and Spectre. They sound similar in many ways?
Sep 06, 2026 at 01:47 AM
p.2 #8From Cobalt, Leica M vs world.
gregfountain wrote:
I understand, it's Friday.It can get heady in here
I honestly felt like blazing a sacred monster should be a requirement before wandering through this Pynchon-esque labyrinth.
Ne314satel wrote
Guys, I can't smoke that much..I understand, it's Friday.
I honestly felt like blazing a sacred monster should be a requirement before wandering through this Pynchon-esque labyrinth.
Sep 06, 2026 at 05:31 AM
p.2 #9From Cobalt, Leica M vs world.
Ulysseita wrote:
Having finally tried, and tested with a monochromator and all the related stuff, pretty much every Leica M from the M9 to the EV1, sorry, I mean the M11, because the EV1 for me is NOT an M...
Have you already tested Leica SL2-S? What is its rating?
Having finally tried, and tested with a monochromator and all the related stuff, pretty much every Leica M from the M9 to the EV1, sorry, I mean the M11, because the EV1 for me is NOT an M...
Have you already tested Leica SL2-S? What is its rating?
Sep 06, 2026 at 07:05 AM
p.2 #10From Cobalt, Leica M vs world.
Ulysseita wrote:
Having finally tried, and tested with a monochromator and all the related stuff, pretty much every Leica M from the M9 to the EV1, sorry, I mean the M11, because the EV1 for me is NOT an M...
We now have a pretty clear set of values describing both the colour of Leica’s JPGs and what is really happening upstream in the RAW.
If we talk about JPGs.
The M9 is the queen and, as has been said many times by people who have worked with Leica forever, Overgaard and company, over in Wetzlar they had clearly decided to create their own...Show more →
I would assume they’re the same, but have you done any comparison of the M10 and M10-P?
Having finally tried, and tested with a monochromator and all the related stuff, pretty much every Leica M from the M9 to the EV1, sorry, I mean the M11, because the EV1 for me is NOT an M...
We now have a pretty clear set of values describing both the colour of Leica’s JPGs and what is really happening upstream in the RAW.
If we talk about JPGs.
The M9 is the queen and, as has been said many times by people who have worked with Leica forever, Overgaard and company, over in Wetzlar they had clearly decided to create their own...Show more →
I would assume they’re the same, but have you done any comparison of the M10 and M10-P?
Sep 06, 2026 at 08:24 AM
|
p.2 #11From Cobalt, Leica M vs world.
@DrSnuggles
Hi,
I’ll answer this one since I’m the creator of the Spectre profiles. It is actually a very interesting question.
There are a few preliminary distinctions that are useful to make, because there is a difference in the architecture of the software pipeline, a difference in the architecture of the profiles themselves, and a difference in measurement methodology, data modelling and fitting.
So, to make an honest comparison between Spectre operating inside the Adobe pipeline and Hasselblad HNCS operating inside Phocus, we first need to keep those distinctions in mind.
Logically, Spectre works like this.
Every sensor/lens system is measured in a controlled laboratory environment using a monochromator, in 10 nm steps across the visible spectrum. From those measurements we reconstruct the spectral sensitivity functions of the complete imaging system.
From that point onward, the modelling is essentially mathematical.
The “probe” we use is a virtual target containing roughly 3,000 colour samples. These samples are selected from a much larger dataset and distributed so as to form a perceptually uniform sampling grid relative to the standard observer.
The same set of samples is then processed in parallel by our profiling software using both the CIE 2006 2° Standard Observer and the SSF curves measured from the actual camera system in the laboratory.
The association between those device-dependent responses and the corresponding scene-referred colorimetric data becomes the ground truth used by the fitter.
For Cobalt Standard (S), we calculate the model for three different illuminants: D75, D55 and LED-B1. In other words, the profile has explicit information about how that particular camera system behaves under those three precise spectral conditions.
Inside the Adobe pipeline this also involves deriving what are called the ColorMatrix values in the DNG specification.
Among other things, these matrices are involved in determining the camera neutral and therefore the linear RGB multipliers that ultimately constitute the effective white balance transformation.
This is why exactly the same WB coordinate — “Daylight”, for example — can produce different results with Adobe Standard and Cobalt Standard (S).
That is also essentially what Fred is seeing with the Leica: the WB preset may be nominally the same, but the two profiles contain a different interpretation of the camera response, and therefore both the resulting white balance and the colour transformation can differ.
Is Spectre accurate? Yes, very much so. But it lives inside the Adobe imaging pipeline, and that pipeline imposes its own architecture, advantages and limitations. That subject alone could become a very long discussion.
HNCS is somewhat different.
We know part of how Hasselblad's system works, but obviously not the entire story because HNCS is proprietary.
Architecturally, its core is closer to an ICC-style colour transform, complemented by additional colour correction tables operating across different illuminants. This is probably where the apparent similarity with Spectre comes from.
But I would describe this primarily as a similarity in the *goal and structure of the colour-management systems*, rather than saying that the profiles themselves are equivalent.
Spectre, as the name suggests, originates from spectral measurements of the actual camera system.
In the case of Hasselblad, we do not know exactly how the source data for HNCS are acquired. They may use physical colour targets — presumably extremely well characterised ones, under controlled illumination — or they may employ spectral measurements or additional proprietary methods. Hasselblad does not publicly disclose enough information for us to say.
And, more importantly, we also do not know the complete design philosophy behind their correction tables.
Are they optimised exclusively for minimum colorimetric error?
Are some regions intentionally reshaped according to Hasselblad's preferred rendering?
How much of what we call the “HNCS look” belongs to calibration, and how much belongs to rendering intent?
Those distinctions matter a lot.
HNCS is an extremely good and sophisticated colour system. But comparing it directly with Spectre is difficult because you are really comparing two entire imaging ecosystems, not simply two sets of matrices or LUTs.
Hasselblad controls the sensor, camera firmware, metadata, RAW interpretation, profiling and the Phocus processing pipeline. This gives them the possibility of designing all of those components as a single closed system.
Spectre has a rather different problem to solve.
We start from the measured physical behaviour of a camera and then have to describe that behaviour as accurately as possible within a pipeline — Adobe's — that we do not control.
That difference is fundamental.
There is also a practical consequence.
HNCS only needs to make a Hasselblad behave like Hasselblad intends a Hasselblad to behave.
Spectre is trying to create a common colorimetric framework across cameras whose native spectral responses may be radically different: Sony, Nikon, Canon, Leica, Fujifilm, Hasselblad and so on.
So one of the goals is not simply “make this camera produce nice colour”, but rather: *given different spectral sensitivities, how closely can we make these cameras describe the same scene within the constraints of the same processing environment?*
That is a slightly different problem.
If you compare a Spectre-profiled Hasselblad file in Lightroom with the same RAW processed through HNCS in Phocus, I would therefore not necessarily expect pixel-for-pixel identical colour, nor would I consider a difference between them proof that one of the two is wrong.
They have different source models, different transforms, different rendering pipelines and potentially different rendering intents.
Under some conditions they may converge remarkably closely. Under others they may diverge, particularly with colours whose spectra interact strongly with differences in sensor sensitivity, or under illuminants that depart significantly from the calibration conditions.
So I think the most honest answer is:
Spectre and HNCS do have conceptual similarities. Both are sophisticated attempts to characterise a camera and produce controlled colour rather than simply accepting its native RGB response.
But once you look at the systems as a whole, that similarity becomes relatively small compared with the architectural differences between them.
HNCS is Hasselblad's colour science operating inside an ecosystem Hasselblad controls from beginning to end.
Spectre is a spectral characterisation of the camera system translated into a model capable of operating inside Adobe's ecosystem — and, importantly, designed so that cameras from completely different manufacturers can share the same colorimetric reference.
Neither description by itself tells you which image you will aesthetically prefer.
And I would actually be suspicious of anyone reducing that comparison to a simple “HNCS is more accurate” or “Spectre is more accurate” statement without first defining *accurate relative to what, under which illuminant, for which spectra, and at which point in the processing pipeline*.
Hi,
I’ll answer this one since I’m the creator of the Spectre profiles. It is actually a very interesting question.
There are a few preliminary distinctions that are useful to make, because there is a difference in the architecture of the software pipeline, a difference in the architecture of the profiles themselves, and a difference in measurement methodology, data modelling and fitting.
So, to make an honest comparison between Spectre operating inside the Adobe pipeline and Hasselblad HNCS operating inside Phocus, we first need to keep those distinctions in mind.
Logically, Spectre works like this.
Every sensor/lens system is measured in a controlled laboratory environment using a monochromator, in 10 nm steps across the visible spectrum. From those measurements we reconstruct the spectral sensitivity functions of the complete imaging system.
From that point onward, the modelling is essentially mathematical.
The “probe” we use is a virtual target containing roughly 3,000 colour samples. These samples are selected from a much larger dataset and distributed so as to form a perceptually uniform sampling grid relative to the standard observer.
The same set of samples is then processed in parallel by our profiling software using both the CIE 2006 2° Standard Observer and the SSF curves measured from the actual camera system in the laboratory.
The association between those device-dependent responses and the corresponding scene-referred colorimetric data becomes the ground truth used by the fitter.
For Cobalt Standard (S), we calculate the model for three different illuminants: D75, D55 and LED-B1. In other words, the profile has explicit information about how that particular camera system behaves under those three precise spectral conditions.
Inside the Adobe pipeline this also involves deriving what are called the ColorMatrix values in the DNG specification.
Among other things, these matrices are involved in determining the camera neutral and therefore the linear RGB multipliers that ultimately constitute the effective white balance transformation.
This is why exactly the same WB coordinate — “Daylight”, for example — can produce different results with Adobe Standard and Cobalt Standard (S).
That is also essentially what Fred is seeing with the Leica: the WB preset may be nominally the same, but the two profiles contain a different interpretation of the camera response, and therefore both the resulting white balance and the colour transformation can differ.
Is Spectre accurate? Yes, very much so. But it lives inside the Adobe imaging pipeline, and that pipeline imposes its own architecture, advantages and limitations. That subject alone could become a very long discussion.
HNCS is somewhat different.
We know part of how Hasselblad's system works, but obviously not the entire story because HNCS is proprietary.
Architecturally, its core is closer to an ICC-style colour transform, complemented by additional colour correction tables operating across different illuminants. This is probably where the apparent similarity with Spectre comes from.
But I would describe this primarily as a similarity in the *goal and structure of the colour-management systems*, rather than saying that the profiles themselves are equivalent.
Spectre, as the name suggests, originates from spectral measurements of the actual camera system.
In the case of Hasselblad, we do not know exactly how the source data for HNCS are acquired. They may use physical colour targets — presumably extremely well characterised ones, under controlled illumination — or they may employ spectral measurements or additional proprietary methods. Hasselblad does not publicly disclose enough information for us to say.
And, more importantly, we also do not know the complete design philosophy behind their correction tables.
Are they optimised exclusively for minimum colorimetric error?
Are some regions intentionally reshaped according to Hasselblad's preferred rendering?
How much of what we call the “HNCS look” belongs to calibration, and how much belongs to rendering intent?
Those distinctions matter a lot.
HNCS is an extremely good and sophisticated colour system. But comparing it directly with Spectre is difficult because you are really comparing two entire imaging ecosystems, not simply two sets of matrices or LUTs.
Hasselblad controls the sensor, camera firmware, metadata, RAW interpretation, profiling and the Phocus processing pipeline. This gives them the possibility of designing all of those components as a single closed system.
Spectre has a rather different problem to solve.
We start from the measured physical behaviour of a camera and then have to describe that behaviour as accurately as possible within a pipeline — Adobe's — that we do not control.
That difference is fundamental.
There is also a practical consequence.
HNCS only needs to make a Hasselblad behave like Hasselblad intends a Hasselblad to behave.
Spectre is trying to create a common colorimetric framework across cameras whose native spectral responses may be radically different: Sony, Nikon, Canon, Leica, Fujifilm, Hasselblad and so on.
So one of the goals is not simply “make this camera produce nice colour”, but rather: *given different spectral sensitivities, how closely can we make these cameras describe the same scene within the constraints of the same processing environment?*
That is a slightly different problem.
If you compare a Spectre-profiled Hasselblad file in Lightroom with the same RAW processed through HNCS in Phocus, I would therefore not necessarily expect pixel-for-pixel identical colour, nor would I consider a difference between them proof that one of the two is wrong.
They have different source models, different transforms, different rendering pipelines and potentially different rendering intents.
Under some conditions they may converge remarkably closely. Under others they may diverge, particularly with colours whose spectra interact strongly with differences in sensor sensitivity, or under illuminants that depart significantly from the calibration conditions.
So I think the most honest answer is:
Spectre and HNCS do have conceptual similarities. Both are sophisticated attempts to characterise a camera and produce controlled colour rather than simply accepting its native RGB response.
But once you look at the systems as a whole, that similarity becomes relatively small compared with the architectural differences between them.
HNCS is Hasselblad's colour science operating inside an ecosystem Hasselblad controls from beginning to end.
Spectre is a spectral characterisation of the camera system translated into a model capable of operating inside Adobe's ecosystem — and, importantly, designed so that cameras from completely different manufacturers can share the same colorimetric reference.
Neither description by itself tells you which image you will aesthetically prefer.
And I would actually be suspicious of anyone reducing that comparison to a simple “HNCS is more accurate” or “Spectre is more accurate” statement without first defining *accurate relative to what, under which illuminant, for which spectra, and at which point in the processing pipeline*.
Sep 06, 2026 at 09:22 AM
p.2 #12From Cobalt, Leica M vs world.
I applaud you guys for coming on here and explaining yourself/opening up to questions-
I enjoy your products and have had nothing but great interaction with you when I reach out. Good stuff here!
I enjoy your products and have had nothing but great interaction with you when I reach out. Good stuff here!
Sep 06, 2026 at 10:04 AM
p.2 #13From Cobalt, Leica M vs world.
Out of interest if the people who create these profiles are here, for repro work, would your regular Repro or Spectre provide an advantage in C1 over our current system which uses custom ICC profiles created using the exact camera/ISO/lens/lighting combination being used for the images themselves? We achieve Fadgi 3* with a Colorchecker SG chart, tests done using OpenDice, but we're not able to achieve 4*. Perhaps due to the ancient Gretag ProfileMaker software we're still using? Current cameras are a mix of Sony A7r (III&IV) bodies, though Fuji GFX is on the radar.
Sep 06, 2026 at 10:43 AM
p.2 #14From Cobalt, Leica M vs world.
Raamiel wrote:
@DrSnuggles@
Hi,
I’ll answer this one since I’m the creator of the Spectre profiles. It is actually a very interesting question.
There are a few preliminary distinctions that are useful to make, because there is a difference in the architecture of the software pipeline, a difference in the architecture of the profiles themselves, and a difference in measurement methodology, data modelling and fitting.
So, to make an honest comparison between Spectre operating inside the Adobe pipeline and Hasselblad HNCS operating inside Phocus, we first need to keep those distinctions in mind.
Logically, Spectre works like this.
Every sensor/lens system is measured in a controlled laboratory environment...Show more →
Thanks for the detail reply. This is technically sounding.
If the goal is product photography or art reproduction, I think this product seems a nice investment.
I have a few questions:
Does Cobalt offer any trial run for a few days or a week before buy?
How the score was given? Any independent source can confirm the results?
@DrSnuggles@
Hi,
I’ll answer this one since I’m the creator of the Spectre profiles. It is actually a very interesting question.
There are a few preliminary distinctions that are useful to make, because there is a difference in the architecture of the software pipeline, a difference in the architecture of the profiles themselves, and a difference in measurement methodology, data modelling and fitting.
So, to make an honest comparison between Spectre operating inside the Adobe pipeline and Hasselblad HNCS operating inside Phocus, we first need to keep those distinctions in mind.
Logically, Spectre works like this.
Every sensor/lens system is measured in a controlled laboratory environment...Show more →
Thanks for the detail reply. This is technically sounding.
If the goal is product photography or art reproduction, I think this product seems a nice investment.
I have a few questions:
Does Cobalt offer any trial run for a few days or a week before buy?
How the score was given? Any independent source can confirm the results?
Sep 06, 2026 at 11:23 AM
p.2 #15From Cobalt, Leica M vs world.
@RustyRus
Thank you! It is genuinely a pleasure for me to talk about this stuff, because colour science is as much a passion as it is my job.
And I really appreciate the kind words.
@bemei
Hi,
yes, potentially.
The Spectre Repro profile is specifically designed for reproduction work.
In the Adobe version it is built as a dual-illuminant D50–D55 profile. We deliberately gave it that small amount of flexibility because, in real reproduction environments, the actual spectral distribution of the lighting is not always perfectly identical to the theoretical D50 reference.
In Capture One, because of the architecture of the ICC pipeline, the Repro profile is necessarily built as a single-illuminant D50 profile.
Our published reproduction tests are performed under D50 illumination using the Adobe version of the profiles. Performance in Capture One is comparable, although there are two variables that no generic profile can completely predict: the actual spectral characteristics of the light used in your studio, and the optical transmission of the lens if it differs significantly from the one used during our laboratory measurements.
In practice, the second point is usually less problematic than it sounds. If you are using modern native lenses from the camera manufacturer, especially lenses contemporary with the camera generation, coating and spectral transmission differences are generally within a relatively narrow range.
Your current approach is actually a very rigorous one: building a custom ICC profile for the exact camera / ISO / lens / lighting combination removes several variables that a general-purpose profile has to accommodate.
So I would not automatically assume that Spectre Repro must outperform such a system.
However, if you are already consistently reaching FADGI 3-star and are being limited by the profiling stage rather than by illumination uniformity, target measurement, noise or capture geometry, then yes, I think Spectre Repro would be very interesting to test. Reaching 4-star may be possible.
The fact that you are still using Gretag ProfileMaker could indeed be part of the limitation. It was an excellent tool, but both profiling methods and the mathematical tools available for fitting camera responses have moved considerably since then.
With the Sony A7R III/IV in particular, the sensor itself certainly gives you enough information to work with. A GFX system would also be a very interesting platform for this type of application.
@zhangyue
For Spectre we deliberately followed a fairly simple design philosophy:
**accurate before pleasing.**
The objective is to extract as much useful colour information as possible from the sensor and translate that response into a representation of the scene that is as physically and colorimetrically meaningful as we can make it within the constraints of the host RAW pipeline.
Spectre is therefore intended primarily as a starting point for post-production, rather than as a finished “look”.
Interestingly, though, improving the underlying colorimetric behaviour often means that relatively little correction is required afterwards. One of the limitations we see in profiles such as Adobe Standard is that parts of the available chromatic response are compressed or reshaped quite aggressively. Removing some of those compromises can make the initial rendering already look more natural, even before creative grading.
Regarding a trial version, unfortunately I have to disappoint you.
These profiles are essentially data files. There is no executable component in which we can implement a licence key, activation server or time limitation. Once the files have been downloaded, they can operate indefinitely, so technically we have no practical way of providing a genuine 7-day trial. For that reason, the download itself has to represent the final acceptance of the purchase.
As for the scores, this deserves a more nuanced answer.
The scores published in our benchmarks are generated using our own metrics and our own standardized testing methodology. They are based on technically measurable properties, but camera colour science is one of those fields where defining the *metric* is almost as important as performing the measurement itself.
A good example is SMI, derived from ISO 17321-1.
We implemented the underlying logic of SMI, but developed our own derivative metric rather than using the original formulation unchanged. Over time we found the original implementation increasingly less discriminating for the specific differences we wanted to quantify between modern camera systems.
So, can an independent laboratory confirm our scores?
**Yes — provided that it reproduces the same measurement methodology, normalization and calculation algorithm.**
I want to be very explicit about that, because I don't want this to sound like an elegant way of avoiding independent verification.
A laboratory equipped to perform comparable spectral measurements could reproduce our measurements and verify our indices if we supplied them with the exact methodology and calculation algorithm.
If they used a completely different metric, however, the numerical score would obviously not be expected to be identical — because they would no longer be measuring exactly the same thing in exactly the same way.
That said, some underlying hardware characteristics would remain visible regardless of the scoring system.
For example, properties related to spectral channel separation, overlap between the sensor sensitivities, metameric behaviour and the amount of independent colour information available from the CFA are physical characteristics of the imaging system. Different well-designed metrics may assign them different numerical values, but they should still reveal broadly the same underlying pattern.
So I would distinguish between two questions:
“Can somebody independently reproduce the exact Cobalt score?”
Yes, if they reproduce our protocol.
“Can somebody independently investigate whether the physical behaviour behind that score is real?”
Absolutely — and they do not necessarily need to use our metric to do that.
Thank you! It is genuinely a pleasure for me to talk about this stuff, because colour science is as much a passion as it is my job.
And I really appreciate the kind words.
@bemei
Hi,
yes, potentially.
The Spectre Repro profile is specifically designed for reproduction work.
In the Adobe version it is built as a dual-illuminant D50–D55 profile. We deliberately gave it that small amount of flexibility because, in real reproduction environments, the actual spectral distribution of the lighting is not always perfectly identical to the theoretical D50 reference.
In Capture One, because of the architecture of the ICC pipeline, the Repro profile is necessarily built as a single-illuminant D50 profile.
Our published reproduction tests are performed under D50 illumination using the Adobe version of the profiles. Performance in Capture One is comparable, although there are two variables that no generic profile can completely predict: the actual spectral characteristics of the light used in your studio, and the optical transmission of the lens if it differs significantly from the one used during our laboratory measurements.
In practice, the second point is usually less problematic than it sounds. If you are using modern native lenses from the camera manufacturer, especially lenses contemporary with the camera generation, coating and spectral transmission differences are generally within a relatively narrow range.
Your current approach is actually a very rigorous one: building a custom ICC profile for the exact camera / ISO / lens / lighting combination removes several variables that a general-purpose profile has to accommodate.
So I would not automatically assume that Spectre Repro must outperform such a system.
However, if you are already consistently reaching FADGI 3-star and are being limited by the profiling stage rather than by illumination uniformity, target measurement, noise or capture geometry, then yes, I think Spectre Repro would be very interesting to test. Reaching 4-star may be possible.
The fact that you are still using Gretag ProfileMaker could indeed be part of the limitation. It was an excellent tool, but both profiling methods and the mathematical tools available for fitting camera responses have moved considerably since then.
With the Sony A7R III/IV in particular, the sensor itself certainly gives you enough information to work with. A GFX system would also be a very interesting platform for this type of application.
@zhangyue
For Spectre we deliberately followed a fairly simple design philosophy:
**accurate before pleasing.**
The objective is to extract as much useful colour information as possible from the sensor and translate that response into a representation of the scene that is as physically and colorimetrically meaningful as we can make it within the constraints of the host RAW pipeline.
Spectre is therefore intended primarily as a starting point for post-production, rather than as a finished “look”.
Interestingly, though, improving the underlying colorimetric behaviour often means that relatively little correction is required afterwards. One of the limitations we see in profiles such as Adobe Standard is that parts of the available chromatic response are compressed or reshaped quite aggressively. Removing some of those compromises can make the initial rendering already look more natural, even before creative grading.
Regarding a trial version, unfortunately I have to disappoint you.
These profiles are essentially data files. There is no executable component in which we can implement a licence key, activation server or time limitation. Once the files have been downloaded, they can operate indefinitely, so technically we have no practical way of providing a genuine 7-day trial. For that reason, the download itself has to represent the final acceptance of the purchase.
As for the scores, this deserves a more nuanced answer.
The scores published in our benchmarks are generated using our own metrics and our own standardized testing methodology. They are based on technically measurable properties, but camera colour science is one of those fields where defining the *metric* is almost as important as performing the measurement itself.
A good example is SMI, derived from ISO 17321-1.
We implemented the underlying logic of SMI, but developed our own derivative metric rather than using the original formulation unchanged. Over time we found the original implementation increasingly less discriminating for the specific differences we wanted to quantify between modern camera systems.
So, can an independent laboratory confirm our scores?
**Yes — provided that it reproduces the same measurement methodology, normalization and calculation algorithm.**
I want to be very explicit about that, because I don't want this to sound like an elegant way of avoiding independent verification.
A laboratory equipped to perform comparable spectral measurements could reproduce our measurements and verify our indices if we supplied them with the exact methodology and calculation algorithm.
If they used a completely different metric, however, the numerical score would obviously not be expected to be identical — because they would no longer be measuring exactly the same thing in exactly the same way.
That said, some underlying hardware characteristics would remain visible regardless of the scoring system.
For example, properties related to spectral channel separation, overlap between the sensor sensitivities, metameric behaviour and the amount of independent colour information available from the CFA are physical characteristics of the imaging system. Different well-designed metrics may assign them different numerical values, but they should still reveal broadly the same underlying pattern.
So I would distinguish between two questions:
“Can somebody independently reproduce the exact Cobalt score?”
Yes, if they reproduce our protocol.
“Can somebody independently investigate whether the physical behaviour behind that score is real?”
Absolutely — and they do not necessarily need to use our metric to do that.
Sep 06, 2026 at 02:48 PM
p.2 #16From Cobalt, Leica M vs world.
Raamiel wrote:
In the Adobe version it is built as a dual-illuminant D50–D55 profile. We deliberately gave it that small amount of flexibility because, in real reproduction environments, the actual spectral distribution of the lighting is not always perfectly identical to the theoretical D50 reference.
In Capture One, because of the architecture of the ICC pipeline, the Repro profile is necessarily built as a single-illuminant D50 profile.
So ... your point about accuracy ... one thing that caught my attention about HNCS was that 3000K = 3000K, 5100K = 5100K, 6500K =6500K 10000K=10000K kind of thing throughout the spectrum as presented by a table posted by Overgaard.
Being one who is prone toward 6500K = 6500K, etc. (not the same as aesthetic preferences, per se ... no more than the profile diff's of Kodachrome vs. Fujichrome vs. Ektachrome) ... I've had an appreciation for the accuracy of the HNCS, in this regard.
Simply asked ... How closely can Spectre achieve that level of accuracy, when applied to the non-Hassy bodies / profiles (recognizing Spectre doesn't have the end-end process of the Hassy controls).
https://www.overgaard.dk/Leica-Cameras-in-Color-Compendium-Their-Actual-Kelvin-Values-by-Thorsten-Overgaard-Leica-M11-Leica-M10-Leica-M9-Leica-SL3-Leica-Q3-Leica-M240.html
In the Adobe version it is built as a dual-illuminant D50–D55 profile. We deliberately gave it that small amount of flexibility because, in real reproduction environments, the actual spectral distribution of the lighting is not always perfectly identical to the theoretical D50 reference.
In Capture One, because of the architecture of the ICC pipeline, the Repro profile is necessarily built as a single-illuminant D50 profile.
So ... your point about accuracy ... one thing that caught my attention about HNCS was that 3000K = 3000K, 5100K = 5100K, 6500K =6500K 10000K=10000K kind of thing throughout the spectrum as presented by a table posted by Overgaard.
Being one who is prone toward 6500K = 6500K, etc. (not the same as aesthetic preferences, per se ... no more than the profile diff's of Kodachrome vs. Fujichrome vs. Ektachrome) ... I've had an appreciation for the accuracy of the HNCS, in this regard.
Simply asked ... How closely can Spectre achieve that level of accuracy, when applied to the non-Hassy bodies / profiles (recognizing Spectre doesn't have the end-end process of the Hassy controls).
https://www.overgaard.dk/Leica-Cameras-in-Color-Compendium-Their-Actual-Kelvin-Values-by-Thorsten-Overgaard-Leica-M11-Leica-M10-Leica-M9-Leica-SL3-Leica-Q3-Leica-M240.html
Sep 06, 2026 at 03:09 PM
p.2 #17From Cobalt, Leica M vs world.
If I understand the OP’s original concern and commentary, the Spectre profiles do a good job at consistency of color transformation for a given illuminant because they characterized the spectral response of each camera. However, this transformation only achieves overall color accuracy and does not necessarily improve color separation, which remains an inherent hardware characteristic. I infer from this that the Spectre monochromator measurements were not done on a per-pixel level but rather over several pixels, maybe even diffused, to get cleaner results. The hardware color separation, however, will be the result of a combined spectral response of each pixel and the spatial convolution of nearby pixels. I would expect denser arrays with smaller pixel pitch to yield more spectral mixing and lower color separation. (This can also explain why tests using color checkers don’t show this as the squares span multiple pixels and the colors difference between squares is relatively large.)
Why do Nikon’s result break this correlation? Perhaps Nikon has a better spectral unmixing algorithm built into its processing pipeline? What I mean by spectral unmixing is a multispectral spatial deconvolution.
Why do Nikon’s result break this correlation? Perhaps Nikon has a better spectral unmixing algorithm built into its processing pipeline? What I mean by spectral unmixing is a multispectral spatial deconvolution.
Sep 06, 2026 at 05:27 PM
p.2 #18From Cobalt, Leica M vs world.
Erich6_ wrote:
If I understand the OP’s original concern and commentary, the Spectre profiles do a good job at consistency of color transformation for a given illuminant because they characterized the spectral response of each camera. However, this transformation only achieves overall color accuracy and does not necessarily improve color separation, which remains an inherent hardware characteristic. I infer from this that the Spectre monochromator measurements were not done on a per-pixel level but rather over several pixels, maybe even diffused, to get cleaner results. The hardware color separation, however, will be the result of a combined spectral response of each pixel...Show more →
Overcoming the effect you describe is the point of e.g. Hasselblad's multi-shot approach to increase color accuracy: https://www.hasselblad.com/h-system/h6d-400c-multi-shot/
With other cameras, using their multi-shot capabilities and downsampling to native sensor resolution is likely to achieve a similar effect by averaging cross-pixel color information (and partially backfilling color interpolation losses caused by the Bayer array).
If I understand the OP’s original concern and commentary, the Spectre profiles do a good job at consistency of color transformation for a given illuminant because they characterized the spectral response of each camera. However, this transformation only achieves overall color accuracy and does not necessarily improve color separation, which remains an inherent hardware characteristic. I infer from this that the Spectre monochromator measurements were not done on a per-pixel level but rather over several pixels, maybe even diffused, to get cleaner results. The hardware color separation, however, will be the result of a combined spectral response of each pixel...Show more →
Overcoming the effect you describe is the point of e.g. Hasselblad's multi-shot approach to increase color accuracy: https://www.hasselblad.com/h-system/h6d-400c-multi-shot/
With other cameras, using their multi-shot capabilities and downsampling to native sensor resolution is likely to achieve a similar effect by averaging cross-pixel color information (and partially backfilling color interpolation losses caused by the Bayer array).
Sep 06, 2026 at 05:44 PM
p.2 #19From Cobalt, Leica M vs world.
Interested to see, above, how highly @mzbe, @Fred Miranda and @zhangyue consider the Spectre profiles. I have the Base Packs for Adobe for the:
Ricoh GR Digital 4 (love the CCD sensor),
Ricoh GR IIIx, and
Leica M10
...and am now considering buying the Spectre profile for my M10. It's interesting to see how highly the (now old) M10 scores on the Spectre metrics. When I got the M10 in 2017, like many other people, I initially found it difficult to get good highlight roll-off with this camera. But that problem was solved by exposing for the highlight and raising the shadows — or, more simply, by just making sure not to overexpose. Eventually, I learned that the problem was how Leica spread the large dynamic range more to the shadows than to the highlights, apparently to more easily get a contrast look similar to that of the M9. Subsequently, I learned from people who had good contacts at Leica, that this contrast curve could not be changed through firmware updates because it was baked into into the hardware.
In any case, I suppose that the Spectre M10 profile will give a much better starting point for further post-processing. Please let me have your further thoughts on the Spectre profile for the M10.
For color, I often use the Cobalt Elite Portra 800 simulation, whose look I like a lot. For B&W — I generally like high-contrast — I could post-process further manually, use one of Cobalt Elite film simulations, or perhaps try the lasts Nik Collection v9 packages. Looking at the Nik Collection instructional videos, however, I find the Nik user interface much too complicated in terms of the plethora of choices that need to be made, which seems more suitable for a professional user than for occasional use.
Ricoh GR Digital 4 (love the CCD sensor),
Ricoh GR IIIx, and
Leica M10
...and am now considering buying the Spectre profile for my M10. It's interesting to see how highly the (now old) M10 scores on the Spectre metrics. When I got the M10 in 2017, like many other people, I initially found it difficult to get good highlight roll-off with this camera. But that problem was solved by exposing for the highlight and raising the shadows — or, more simply, by just making sure not to overexpose. Eventually, I learned that the problem was how Leica spread the large dynamic range more to the shadows than to the highlights, apparently to more easily get a contrast look similar to that of the M9. Subsequently, I learned from people who had good contacts at Leica, that this contrast curve could not be changed through firmware updates because it was baked into into the hardware.
In any case, I suppose that the Spectre M10 profile will give a much better starting point for further post-processing. Please let me have your further thoughts on the Spectre profile for the M10.
For color, I often use the Cobalt Elite Portra 800 simulation, whose look I like a lot. For B&W — I generally like high-contrast — I could post-process further manually, use one of Cobalt Elite film simulations, or perhaps try the lasts Nik Collection v9 packages. Looking at the Nik Collection instructional videos, however, I find the Nik user interface much too complicated in terms of the plethora of choices that need to be made, which seems more suitable for a professional user than for occasional use.
Sep 06, 2026 at 09:57 PM
p.2 #20From Cobalt, Leica M vs world.
Curious ... have you tested under D65?
If the goal is accuracy, using partial spectrum illuminants automatically skew the inherent reflected spectral response. So, to score the sensors ability to be accurate using a skewed illuminant seems a bit antithetical.
That said, I get the prospect of understanding its response in different lighting, and D50 being chosen as a reference to "Daylight" balance, and A for reference to tungsten. But, imo ... if I want to know if the sensor is accurately recording spectral response, I'd want to begin with full spectrum D65. THEN, moving about through partial spectrum illuminants is a variant matter, and its ability to provide spectral separation / accuracy under partial spectrum illumination.
Seeing the Hassy have a low score under D50 or A ... ummm, is that a condition of the Hassy being inaccurate, or a condition of the test illuminant skewing spectral response through the use of a partial spectrum illuminant.
Imo, the matter of "human vision" emulation (and relative to partial spectrum lighting conditions) and the matter of technical accuracy are two different things. It seems that using an illuminant other than D65 source automatically skews the spectral response, as it pertains to accuracy (vs. perception under skewed illuminants).
Since, both D50 and A are warmer than D65, what does the use of D50 and A tell us about when we encounter cooler lighting conditions, relative to accuracy. Imo, this presents a gap in the utility to discern technical accuracy of the spectral response. That's not to say that I don't appreciate the work that has been done to this point, but given that D50 represents a narrow band of time that natural lighting exists at that illuminant response, it isn't a complete representation of how accurate the response in full spectrum, nor in context of ever-varying natural lighting conditions. Simply to say, I think you MUST start with D65, full spectrum, if we are to have a discussion about the ability to be accurate and/or the ability to differentiate spectral response.
Using partial spectrum illumination for the test method ... well, leaves me wondering if the test method has left something behind ... from a purely technical understanding of spectral response accuracy. I mean if I shine a red light on a white paper, the spectral response will be red, not white. Whereas if I shine white light on white paper, I expect a white response. Using D50 and Std A have their place, but I'd start with D65, and go forward (both directions) from there. Using warm (D50) and warmer (A), partial spectrum illuminants for assessing accuracy means you're starting from a skewed base (accuracy and separation) to begin with, so it seems to me. I'm aware that lighting industry uses those reference points, but I can't see how the use of partial spectrum illuminants can yield an accurate picture of picture accuracy. Set the camera to 6500K and use D65 illuminant ... start there, imo.
Imo, the table that Overgaard provided indicates that the Hassy is very accurate vs. the different Leica models. While we don't have the full technical details of the Hassy ... his indication of how good the Hassy is vs. your testing suggest TWO VERY different stories. They can't BOTH be right, as it pertains to technical accuracy. I suspect that if you tested the accuracy of the Hassy under D65 illumination as the full spectrum standard, the Hassy (et al) would rank differently in your scoring.
Again, appreciate the work ...
If the goal is accuracy, using partial spectrum illuminants automatically skew the inherent reflected spectral response. So, to score the sensors ability to be accurate using a skewed illuminant seems a bit antithetical.
That said, I get the prospect of understanding its response in different lighting, and D50 being chosen as a reference to "Daylight" balance, and A for reference to tungsten. But, imo ... if I want to know if the sensor is accurately recording spectral response, I'd want to begin with full spectrum D65. THEN, moving about through partial spectrum illuminants is a variant matter, and its ability to provide spectral separation / accuracy under partial spectrum illumination.
Seeing the Hassy have a low score under D50 or A ... ummm, is that a condition of the Hassy being inaccurate, or a condition of the test illuminant skewing spectral response through the use of a partial spectrum illuminant.
Imo, the matter of "human vision" emulation (and relative to partial spectrum lighting conditions) and the matter of technical accuracy are two different things. It seems that using an illuminant other than D65 source automatically skews the spectral response, as it pertains to accuracy (vs. perception under skewed illuminants).
Since, both D50 and A are warmer than D65, what does the use of D50 and A tell us about when we encounter cooler lighting conditions, relative to accuracy. Imo, this presents a gap in the utility to discern technical accuracy of the spectral response. That's not to say that I don't appreciate the work that has been done to this point, but given that D50 represents a narrow band of time that natural lighting exists at that illuminant response, it isn't a complete representation of how accurate the response in full spectrum, nor in context of ever-varying natural lighting conditions. Simply to say, I think you MUST start with D65, full spectrum, if we are to have a discussion about the ability to be accurate and/or the ability to differentiate spectral response.
Using partial spectrum illumination for the test method ... well, leaves me wondering if the test method has left something behind ... from a purely technical understanding of spectral response accuracy. I mean if I shine a red light on a white paper, the spectral response will be red, not white. Whereas if I shine white light on white paper, I expect a white response. Using D50 and Std A have their place, but I'd start with D65, and go forward (both directions) from there. Using warm (D50) and warmer (A), partial spectrum illuminants for assessing accuracy means you're starting from a skewed base (accuracy and separation) to begin with, so it seems to me. I'm aware that lighting industry uses those reference points, but I can't see how the use of partial spectrum illuminants can yield an accurate picture of picture accuracy. Set the camera to 6500K and use D65 illuminant ... start there, imo.
Imo, the table that Overgaard provided indicates that the Hassy is very accurate vs. the different Leica models. While we don't have the full technical details of the Hassy ... his indication of how good the Hassy is vs. your testing suggest TWO VERY different stories. They can't BOTH be right, as it pertains to technical accuracy. I suspect that if you tested the accuracy of the Hassy under D65 illumination as the full spectrum standard, the Hassy (et al) would rank differently in your scoring.
Again, appreciate the work ...
Sep 07, 2026 at 03:21 AM