From Cobalt, Leica M vs world.
p.1 #1From Cobalt, Leica M vs world.
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 very strong interpretation of colour. In some ways we are almost in "Kodachrome" territory: skies drifting towards teal, very orange lights, saturation and separation deliberately given a strong character.
Yes, the colour of NON-modified M9s, meaning the colour of the original JPG, is very obviously something unique that Leica never really reproduced again with the same strength in the generations that followed....We are basically in this philosophy:
"We give you the JPG. With the RAW, you’re on your own."
And in fact the RAW is a fairly bare DNG, with embedded profiles that are certainly nothing miraculous and, on the Adobe side, essentially matrix-based profiles.
With the M240, the prototype with video and a million options that even today makes me miss ONLY one thing, the viewfinder lines turning RED... they could have kept that going, CHEAP BASTARDS! [MrGreen] (https://www.juzaphoto.com/shared_files/layout/icon_mrgreen.gif), and then with the later M262, the JPG gets toned down a little but still stays on the strong side.
To give you an idea, we are more in the territory of Fuji’s Provia: punchy, but still reasonably balanced.
From the M10 onwards, in my opinion the philosophy itself changes.
We move much more into Fuji Pro Neg Std territory: softer, cleaner, less characterised, much more generic.
But let’s get to the RAW files, because that is where things get much more interesting.
The data is here:
https://www.cobalt-image.com/spectre-laboratory-measurements/
We have already looked at the separation of the M9, CCD and all the mythology that comes with it.
Just to give a reference, the sacred monsters we have measured, almost all Nikon, sit roughly between 85 and 89.
Then, apart from a few interesting exceptions I will talk about, we often find Canon above 70, followed by Sony, Fuji etc.
And Leica?
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
And here, in my opinion, something VERY interesting comes out.
Some people will remember that, having first measured the M10-R and then various SL and Q models, we had started wondering whether Leica simply did not care that much about hardware colour separation...
These numbers tell a different story, because the M240 and M10 are EXCELLENT.
80.7 and 80.8 mean that Leica, after the mess of the M8 and the whole IR story, had not only fixed the problem, but had clearly invested quite heavily in the spectral discrimination of the sensor + CFA system as well.
The real drop happens with the M10-R.
And it is a huge one.
We go from 80.8 on the M10 to 57.5 on the M10-R.
At the same time they gain almost twice the megapixels, greater dynamic range and a much better recovery capability.
Here though we need to be precise; Leica does NOT officially state, at least not in the specifications I have found, that the M10-R is simply "the M10 Monochrom sensor with a Bayer filter added on top".
The M10 Monochrom and M10-R clearly belong to the same roughly 40 MP generation, but from that we cannot automatically conclude that the silicon is literally identical, even if their shared branching strongly suggests there is more to it...
What we can say for certain is that the M10-R uses a 40.89 MP CMOS sensor without a low-pass filter.
And it is exactly from this generation that separation collapses.
The most plausible explanation, looking at the SSFs, is that the CFA trade-off changed heavily.
More selective filters separate R, G and B better, but throw away more light; broader, more transmissive filters let more photons reach the sensor, but they also make the three spectral responses overlap more.
So you can gain efficiency, noise performance, DR etc., while losing the ability to distinguish between two different spectra that produce similar RGB values.
And careful: this is an explanation compatible with the data, it does NOT automatically mean that "more megapixels = worse colour".
Quite the opposite.
The M11 climbs slightly back to 61.60, but still remains miles away from the M240 and M10.
And it is a 60.3 MP sensor with RGB filter, UV/IR filtering and no low-pass filter. Leica states that clearly in the specifications.
Now go and look at the other monsters, A7R6, GFX100, Hasselblad 100... all low.
So it becomes very easy to think:
"Yeah well, sensors are insanely dense now, pixels are small, they need more open CFAs to keep quantum efficiency and DR up, so separation goes down."
Nice.
Except then the 45 MP Nikons come along and ruin the easy explanation, because some of them sit between 85 and 89.
So high resolution and strong spectral separation are NOT necessarily incompatible.
And this, in my opinion, is the most interesting point of all.
The M10-R tells us that, moving to the 40 MP generation, Leica clearly accepted a very different compromise compared with the M10.
The M11 recovers something, but does not even come close to the levels of the 24 MP M bodies.
Meanwhile Nikon manages to put 45 MP on full frame and stay right up there with some of the best values we have ever measured.
So the question remains:
WHY THE HELL IS NIKON STILL UP THERE with 45 megapixels?
Because clearly this is not just a megapixel issue.....and that is the next interesting thing to understand.
PS. translated from my topic on another italian forum.
https://www.juzaphoto.com/topic2.php?l=it&t=5356732&show=1
Edited on Sep 05, 2026 at 05:17 AM · View previous versions
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 very strong interpretation of colour. In some ways we are almost in "Kodachrome" territory: skies drifting towards teal, very orange lights, saturation and separation deliberately given a strong character.
Yes, the colour of NON-modified M9s, meaning the colour of the original JPG, is very obviously something unique that Leica never really reproduced again with the same strength in the generations that followed....We are basically in this philosophy:
"We give you the JPG. With the RAW, you’re on your own."
And in fact the RAW is a fairly bare DNG, with embedded profiles that are certainly nothing miraculous and, on the Adobe side, essentially matrix-based profiles.
With the M240, the prototype with video and a million options that even today makes me miss ONLY one thing, the viewfinder lines turning RED... they could have kept that going, CHEAP BASTARDS! [MrGreen] (https://www.juzaphoto.com/shared_files/layout/icon_mrgreen.gif), and then with the later M262, the JPG gets toned down a little but still stays on the strong side.
To give you an idea, we are more in the territory of Fuji’s Provia: punchy, but still reasonably balanced.
From the M10 onwards, in my opinion the philosophy itself changes.
We move much more into Fuji Pro Neg Std territory: softer, cleaner, less characterised, much more generic.
But let’s get to the RAW files, because that is where things get much more interesting.
The data is here:
https://www.cobalt-image.com/spectre-laboratory-measurements/
We have already looked at the separation of the M9, CCD and all the mythology that comes with it.
Just to give a reference, the sacred monsters we have measured, almost all Nikon, sit roughly between 85 and 89.
Then, apart from a few interesting exceptions I will talk about, we often find Canon above 70, followed by Sony, Fuji etc.
And Leica?
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
And here, in my opinion, something VERY interesting comes out.
Some people will remember that, having first measured the M10-R and then various SL and Q models, we had started wondering whether Leica simply did not care that much about hardware colour separation...
These numbers tell a different story, because the M240 and M10 are EXCELLENT.
80.7 and 80.8 mean that Leica, after the mess of the M8 and the whole IR story, had not only fixed the problem, but had clearly invested quite heavily in the spectral discrimination of the sensor + CFA system as well.
The real drop happens with the M10-R.
And it is a huge one.
We go from 80.8 on the M10 to 57.5 on the M10-R.
At the same time they gain almost twice the megapixels, greater dynamic range and a much better recovery capability.
Here though we need to be precise; Leica does NOT officially state, at least not in the specifications I have found, that the M10-R is simply "the M10 Monochrom sensor with a Bayer filter added on top".
The M10 Monochrom and M10-R clearly belong to the same roughly 40 MP generation, but from that we cannot automatically conclude that the silicon is literally identical, even if their shared branching strongly suggests there is more to it...
What we can say for certain is that the M10-R uses a 40.89 MP CMOS sensor without a low-pass filter.
And it is exactly from this generation that separation collapses.
The most plausible explanation, looking at the SSFs, is that the CFA trade-off changed heavily.
More selective filters separate R, G and B better, but throw away more light; broader, more transmissive filters let more photons reach the sensor, but they also make the three spectral responses overlap more.
So you can gain efficiency, noise performance, DR etc., while losing the ability to distinguish between two different spectra that produce similar RGB values.
And careful: this is an explanation compatible with the data, it does NOT automatically mean that "more megapixels = worse colour".
Quite the opposite.
The M11 climbs slightly back to 61.60, but still remains miles away from the M240 and M10.
And it is a 60.3 MP sensor with RGB filter, UV/IR filtering and no low-pass filter. Leica states that clearly in the specifications.
Now go and look at the other monsters, A7R6, GFX100, Hasselblad 100... all low.
So it becomes very easy to think:
"Yeah well, sensors are insanely dense now, pixels are small, they need more open CFAs to keep quantum efficiency and DR up, so separation goes down."
Nice.
Except then the 45 MP Nikons come along and ruin the easy explanation, because some of them sit between 85 and 89.
So high resolution and strong spectral separation are NOT necessarily incompatible.
And this, in my opinion, is the most interesting point of all.
The M10-R tells us that, moving to the 40 MP generation, Leica clearly accepted a very different compromise compared with the M10.
The M11 recovers something, but does not even come close to the levels of the 24 MP M bodies.
Meanwhile Nikon manages to put 45 MP on full frame and stay right up there with some of the best values we have ever measured.
So the question remains:
WHY THE HELL IS NIKON STILL UP THERE with 45 megapixels?
Because clearly this is not just a megapixel issue.....and that is the next interesting thing to understand.
PS. translated from my topic on another italian forum.
https://www.juzaphoto.com/topic2.php?l=it&t=5356732&show=1
Edited on Sep 05, 2026 at 05:17 AM · View previous versions
Sep 04, 2026 at 11:36 AM
p.1 #2From Cobalt, Leica M vs world.
I realize this is in reference to M ... but, I'd be curious to see the SL journey as well (heck Q also).
Then, to see if the new SL3-P might be a candidate to what might be possible in an M ... potentially an improved M10-R (if you will), badged as an M12 or ? ... looking toward an optimized quality vs. quantity of MP's, kind of thing. (sheer speculation of course).
Sep 04, 2026 at 02:04 PM
p.1 #3From Cobalt, Leica M vs world.
Ulysseita wrote:
And Leica?
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
Here is a question about Spectre: once you apply a Cobalt Spectre profile on any of these cameras, does it even matter anymore? After all, the profile corrects for the weaker spectral separation of the original sensor.
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?
And Leica?
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
Here is a question about Spectre: once you apply a Cobalt Spectre profile on any of these cameras, does it even matter anymore? After all, the profile corrects for the weaker spectral separation of the original sensor.
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?
Sep 04, 2026 at 02:49 PM
p.1 #4From Cobalt, Leica M vs world.
Fred Miranda wrote:
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60Here is a question about Spectre: once you apply a Cobalt Spectre profile on any of these cameras, does it even matter anymore? After all, the profile corrects for the weaker spectral separation of the original sensor.
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?
https://www.cobalt-image.com/tutorials/spectre-benchmarks-explained/
The example is already explained, and the Nikon could basically be used there as if it were an M10.
The key point is that while profiled Fadgi separation can be improved, and by a lot, it is still based on a reference built around a broad and carefully selected set of representative surface colours, essentially summarised by something like the ColorChecker SG. Hardware separation, on the other hand, cannot be improved through profiling.
In the “normal” world, the M10-R and M10 can do a very similar job. But in the example from the link, with autumn trees, things already change quite a lot because there is a much broader range of greens than what can really be represented by an SG target, and that is where the difference in hardware colour separation starts to become visible.
Anecdote: I spent a weekend in Wales with a friend. I had the GFX and A1 II, while he had an X2D.
He was wearing a kind of dark green jacket, almost undergrowth green, but under certain semi-gloss lighting conditions it shifted slightly towards brown. The GFX generic and with Spectre saw it as brown. The X2D with Spectre did too, and so did Hasselblad’s own Phocus software. The Sony A1 II did not reproduce exactly the same colour either, but it got closer.
I suspect a D700, which is right at the top of our hardware-separation ranking, would probably have come even closer again.
Why can cameras costing thousands, fully calibrated and even using proprietary colour software with a strong reputation like Hasselblad’s, still miss a colour like that?
Because the standard observer is enormous in scope, while what we see with our eyes has to be compressed, sampled and reconstructed by electronics with a limited number of channels. Most of the time that approximation works extraordinarily well and reproduces a huge percentage of real-world colour.
But every now and then, a synthetic colour like that jacket falls between the cracks of that approximation, and almost every photographic system misses it to some degree.
Spectre reduces the profiling gap. It cannot remove the sensor gap.
Maybe that’s why I still keep a D700 in a drawer. If needed, I’ll just pull it out and use it as a photographic colorimeter.
Ulysseita wrote
And Leica?M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?
https://www.cobalt-image.com/tutorials/spectre-benchmarks-explained/
The example is already explained, and the Nikon could basically be used there as if it were an M10.
The key point is that while profiled Fadgi separation can be improved, and by a lot, it is still based on a reference built around a broad and carefully selected set of representative surface colours, essentially summarised by something like the ColorChecker SG. Hardware separation, on the other hand, cannot be improved through profiling.
In the “normal” world, the M10-R and M10 can do a very similar job. But in the example from the link, with autumn trees, things already change quite a lot because there is a much broader range of greens than what can really be represented by an SG target, and that is where the difference in hardware colour separation starts to become visible.
Anecdote: I spent a weekend in Wales with a friend. I had the GFX and A1 II, while he had an X2D.
He was wearing a kind of dark green jacket, almost undergrowth green, but under certain semi-gloss lighting conditions it shifted slightly towards brown. The GFX generic and with Spectre saw it as brown. The X2D with Spectre did too, and so did Hasselblad’s own Phocus software. The Sony A1 II did not reproduce exactly the same colour either, but it got closer.
I suspect a D700, which is right at the top of our hardware-separation ranking, would probably have come even closer again.
Why can cameras costing thousands, fully calibrated and even using proprietary colour software with a strong reputation like Hasselblad’s, still miss a colour like that?
Because the standard observer is enormous in scope, while what we see with our eyes has to be compressed, sampled and reconstructed by electronics with a limited number of channels. Most of the time that approximation works extraordinarily well and reproduces a huge percentage of real-world colour.
But every now and then, a synthetic colour like that jacket falls between the cracks of that approximation, and almost every photographic system misses it to some degree.
Spectre reduces the profiling gap. It cannot remove the sensor gap.
Maybe that’s why I still keep a D700 in a drawer. If needed, I’ll just pull it out and use it as a photographic colorimeter.
Sep 04, 2026 at 03:57 PM
p.1 #5From Cobalt, Leica M vs world.
Guys, I can't smoke that much..
I understand, it's Friday.
I understand, it's Friday.
Sep 04, 2026 at 04:25 PM
p.1 #6From Cobalt, Leica M vs world.
Ulysseita wrote:
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60Here is a question about Spectre: once you apply a Cobalt Spectre profile on any of these cameras, does it even matter anymore? After all, the profile corrects for the weaker spectral separation of the original sensor.
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?https://www.cobalt-image.com/tutorials/spectre-benchmarks-explained/
The example is already explained, and the Nikon could basically be used there as if it were an M10.
The key point is that while profiled Fadgi separation can be improved, and by a lot, it is still based on a reference built around a broad and carefully selected set of representative surface colours, essentially summarised by something like the ColorChecker SG. Hardware separation, on the other hand, cannot be improved through profiling.
In the “normal” world, the M10-R and M10 can do a very similar job. But in the example from the link, with autumn trees, things already change quite a...Show more →
I understand there is a sensor gap, but once you apply the Spectre profile, the numbers seem to even out. The M240 still has an edge over the M10-R, but they end up looking almost identical.
Am I right, or am I missing something here?
Fred Miranda wrote
Ulysseita wrote
And Leica?M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?
The example is already explained, and the Nikon could basically be used there as if it were an M10.
The key point is that while profiled Fadgi separation can be improved, and by a lot, it is still based on a reference built around a broad and carefully selected set of representative surface colours, essentially summarised by something like the ColorChecker SG. Hardware separation, on the other hand, cannot be improved through profiling.
In the “normal” world, the M10-R and M10 can do a very similar job. But in the example from the link, with autumn trees, things already change quite a...Show more →
I understand there is a sensor gap, but once you apply the Spectre profile, the numbers seem to even out. The M240 still has an edge over the M10-R, but they end up looking almost identical.
Am I right, or am I missing something here?
Sep 04, 2026 at 06:34 PM
p.1 #7From Cobalt, Leica M vs world.
Well that explains why I feel like color discrimination has been de-evolving. My favorite color output of all time came from the Nikon Df and the original Q (and SL 601 that shares the same sensor). SL2-S is a close second place.
Sep 04, 2026 at 07:01 PM
p.1 #8From Cobalt, Leica M vs world.
Fred Miranda wrote:
M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60Here is a question about Spectre: once you apply a Cobalt Spectre profile on any of these cameras, does it even matter anymore? After all, the profile corrects for the weaker spectral separation of the original sensor.
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?https://www.cobalt-image.com/tutorials/spectre-benchmarks-explained/
The example is already explained, and the Nikon could basically be used there as if it were an M10.
The key point is that while profiled Fadgi separation can be improved, and by a lot, it is still based on a reference built around a broad and carefully selected set of representative surface colours, essentially summarised by something like the ColorChecker SG. Hardware separation, on the other hand, cannot be improved through profiling.
In the “normal” world, the M10-R and M10 can do a very similar job. But in the example from the link, with autumn trees, things already change quite a lot because there is a much broader range of greens than what can really be represented by an SG target, and that is where the difference in hardware colour separation starts to become visible.
Anecdote: I spent a weekend in Wales with a friend. I had the GFX and A1 II, while he had an X2D.
He was wearing a kind of dark green jacket, almost undergrowth green, but under certain semi-gloss lighting conditions it shifted slightly towards brown. The GFX generic and with Spectre saw it as brown. The X2D with Spectre did too, and so did Hasselblad’s own Phocus software. The Sony A1 II did not reproduce exactly the same colour either, but it got closer.
I suspect a D700, which is right at the top of our hardware-separation ranking, would probably have come even closer again.
Why can cameras costing thousands, fully calibrated and even using proprietary colour software with a strong reputation like Hasselblad’s, still miss a colour like that?
Because the standard observer is enormous in scope, while what we see with our eyes has to be compressed, sampled and reconstructed by electronics with a limited number of channels. Most of the time that approximation works extraordinarily well and reproduces a huge percentage of real-world colour.
But every now and then, a synthetic colour like that jacket falls between the cracks of that approximation, and almost every photographic system misses it to some degree.
Spectre reduces the profiling gap. It cannot remove the sensor gap.
Maybe that’s why I still keep a D700 in a drawer. If needed, I’ll just pull it out and use it as a photographic colorimeter.I understand there is a sensor gap, but once you apply the Spectre profile, the numbers seem to even out. The M240 still has an edge over the M10-R, but they end up looking almost identical.
Am I right, or am I missing something here?
these are the numbers:
"Overall Hardware
A combined score derived from our core hardware measurements. It includes Separation Performance under both D50 and StdA reference illuminants, together with other hardware-specific indicators such as the Vora and Luther–Ives indices. The result provides a concise overview of the sensor’s overall quality, spectral behaviour, and profiling potential.
Separation Performance
A measure of the sensor’s ability to separate colour across the full chromaticity range of the standard observer under the D50 reference illuminant. Unlike patch-based profile tests, it provides a broader view of the sensor’s overall colour discrimination.
Generic Profile
The SMI score of the generic Adobe profile. It represents the baseline profile accuracy of the camera, evaluated on a restricted set of ColorChecker SG patches.
Spectre Profile
The SMI score of the Spectre profile. It represents the improved level of colour accuracy achieved through our spectral profiling method, evaluated on the same restricted ColorChecker SG patch set.
FADGI 2023 (Generic)
The FADGI 2023 star rating of the generic Adobe profile. Based on a restricted set of ColorChecker SG patches, it offers a practical classification of profile quality.
FADGI 2023 (Spectre)
The FADGI 2023 star rating of the Spectre profile. Based on the same restricted patch set, it highlights the practical improvement in profile quality delivered by Spectre."
Fred, imagine photographing ten slightly different green leaves.
A profile can make the greens more accurate. But if the camera records two of those leaves as the same colour, the profile cannot know that they should be different. It can correct their colour, but it cannot recover the distinction.
Getting a colour right and telling two colours apart are different things.
The scores you posted measure how accurately the profiled cameras reproduce the reference chart. Both can do extremely well there, without being equally capable of distinguishing all the subtle variations in those leaves.
That is why I linked the foliage comparison in the Step 5 on the article. The lower images already have Spectre applied to both cameras, yet differences remain within the greens. The similar chart scores and the different foliage rendering are entirely compatible.
https://www.cobalt-image.com/tutorials/spectre-benchmarks-explained/
Ulysseita wrote
Fred Miranda wrote
Ulysseita wrote
And Leica?M9: 72.2
M240: 80.70
M10: 80.80
M10-R: 57.5
M11: 61.60
Does that mean a camera like the M10-R would actually match the standard M10 in this regard when both are using the Spectre profile?
The example is already explained, and the Nikon could basically be used there as if it were an M10.
The key point is that while profiled Fadgi separation can be improved, and by a lot, it is still based on a reference built around a broad and carefully selected set of representative surface colours, essentially summarised by something like the ColorChecker SG. Hardware separation, on the other hand, cannot be improved through profiling.
In the “normal” world, the M10-R and M10 can do a very similar job. But in the example from the link, with autumn trees, things already change quite a lot because there is a much broader range of greens than what can really be represented by an SG target, and that is where the difference in hardware colour separation starts to become visible.
Anecdote: I spent a weekend in Wales with a friend. I had the GFX and A1 II, while he had an X2D.
He was wearing a kind of dark green jacket, almost undergrowth green, but under certain semi-gloss lighting conditions it shifted slightly towards brown. The GFX generic and with Spectre saw it as brown. The X2D with Spectre did too, and so did Hasselblad’s own Phocus software. The Sony A1 II did not reproduce exactly the same colour either, but it got closer.
I suspect a D700, which is right at the top of our hardware-separation ranking, would probably have come even closer again.
Why can cameras costing thousands, fully calibrated and even using proprietary colour software with a strong reputation like Hasselblad’s, still miss a colour like that?
Because the standard observer is enormous in scope, while what we see with our eyes has to be compressed, sampled and reconstructed by electronics with a limited number of channels. Most of the time that approximation works extraordinarily well and reproduces a huge percentage of real-world colour.
But every now and then, a synthetic colour like that jacket falls between the cracks of that approximation, and almost every photographic system misses it to some degree.
Spectre reduces the profiling gap. It cannot remove the sensor gap.
Maybe that’s why I still keep a D700 in a drawer. If needed, I’ll just pull it out and use it as a photographic colorimeter.
Am I right, or am I missing something here?
these are the numbers:
"Overall Hardware
A combined score derived from our core hardware measurements. It includes Separation Performance under both D50 and StdA reference illuminants, together with other hardware-specific indicators such as the Vora and Luther–Ives indices. The result provides a concise overview of the sensor’s overall quality, spectral behaviour, and profiling potential.
Separation Performance
A measure of the sensor’s ability to separate colour across the full chromaticity range of the standard observer under the D50 reference illuminant. Unlike patch-based profile tests, it provides a broader view of the sensor’s overall colour discrimination.
Generic Profile
The SMI score of the generic Adobe profile. It represents the baseline profile accuracy of the camera, evaluated on a restricted set of ColorChecker SG patches.
Spectre Profile
The SMI score of the Spectre profile. It represents the improved level of colour accuracy achieved through our spectral profiling method, evaluated on the same restricted ColorChecker SG patch set.
FADGI 2023 (Generic)
The FADGI 2023 star rating of the generic Adobe profile. Based on a restricted set of ColorChecker SG patches, it offers a practical classification of profile quality.
FADGI 2023 (Spectre)
The FADGI 2023 star rating of the Spectre profile. Based on the same restricted patch set, it highlights the practical improvement in profile quality delivered by Spectre."
Fred, imagine photographing ten slightly different green leaves.
A profile can make the greens more accurate. But if the camera records two of those leaves as the same colour, the profile cannot know that they should be different. It can correct their colour, but it cannot recover the distinction.
Getting a colour right and telling two colours apart are different things.
The scores you posted measure how accurately the profiled cameras reproduce the reference chart. Both can do extremely well there, without being equally capable of distinguishing all the subtle variations in those leaves.
That is why I linked the foliage comparison in the Step 5 on the article. The lower images already have Spectre applied to both cameras, yet differences remain within the greens. The similar chart scores and the different foliage rendering are entirely compatible.
https://www.cobalt-image.com/tutorials/spectre-benchmarks-explained/
Sep 04, 2026 at 11:31 PM
p.1 #9From Cobalt, Leica M vs world.
For a limited time, I'll be accepting trades of an M10-R for my MP240.
Sep 05, 2026 at 07:42 AM
p.1 #10From Cobalt, Leica M vs world.
Seabassius wrote:
For a limited time, I'll be accepting trades of an M10-R for my MP240.
Fair trade based on the evidence!
For a limited time, I'll be accepting trades of an M10-R for my MP240.
Fair trade based on the evidence!
Sep 05, 2026 at 10:34 AM
|
p.1 #11From Cobalt, Leica M vs world.
I greatly appreciate this work. Much more useful than many other aspects of 'feature testing'.
Two thoughts:
1 - I am wondering how this relates to similar SMI measure e.g. done by DXO labs - they show cameras much more closely together. Example: https://www.dxomark.com/Cameras/Leica/M11---Measurements
M11 is shown (Color Response tab) with (ISO 17321-based) SMI 84, Nikon Z8 gets SMI 83 (less accurate) https://www.dxomark.com/Cameras/Nikon/Z8---Measurements
2 - Regarding the 'real life' foliage shots, it is hard for me to tell if this could also be related to white balance, Leica warmer (more yellow), Nikon coolor (more blue)? I have found ISO settings in raw converters to deliver different results for the same image and same number (5500 is not 5500), or different camera same scene different WB setting to achieve similar appearance (or calibrating of the same gray card).
Testing my hypothesis, I took a screenshot and pasted into Graphic Converter 12, and clicked on Filter -> Xe8472 Automatic. This is the Leica image which now looks much closer to the Nikon image shared above?
Edited on Sep 05, 2026 at 11:24 AM · View previous versions
Two thoughts:
1 - I am wondering how this relates to similar SMI measure e.g. done by DXO labs - they show cameras much more closely together. Example: https://www.dxomark.com/Cameras/Leica/M11---Measurements
M11 is shown (Color Response tab) with (ISO 17321-based) SMI 84, Nikon Z8 gets SMI 83 (less accurate) https://www.dxomark.com/Cameras/Nikon/Z8---Measurements
2 - Regarding the 'real life' foliage shots, it is hard for me to tell if this could also be related to white balance, Leica warmer (more yellow), Nikon coolor (more blue)? I have found ISO settings in raw converters to deliver different results for the same image and same number (5500 is not 5500), or different camera same scene different WB setting to achieve similar appearance (or calibrating of the same gray card).
Testing my hypothesis, I took a screenshot and pasted into Graphic Converter 12, and clicked on Filter -> Xe8472 Automatic. This is the Leica image which now looks much closer to the Nikon image shared above?
Edited on Sep 05, 2026 at 11:24 AM · View previous versions
Sep 05, 2026 at 10:59 AM
p.1 #12From Cobalt, Leica M vs world.
mzbe wrote:
I greatly appreciate this work. Much more useful than many other aspects of 'feature testing'.
Two thoughts:
1 - I am wondering how this relates to similar SMI measure e.g. done by DXO labs - they show cameras much more closely together. Example: https://www.dxomark.com/Cameras/Leica/M11---Measurements
M11 is shown (Color Response tab) with (ISO 17321-based) SMI 84, Nikon Z8 gets SMI 83 (less accurate) https://www.dxomark.com/Cameras/Nikon/Z8---Measurements
2 - Regarding the 'real life' foliage shots, it is hard for me to tell if this could also be related to white balance, Leica warmer (more yellow), Nikon coolor (more blue)? I have found ISO settings in raw converters to...Show more →
1-DXOMARK is measuring something related, but not directly equivalent to what we are showing.
Their SMI figure is based on ISO 17321. In their published protocol, DXO uses the classic GretagMacbeth ColorChecker with 18 chromatic patches, measured from RAW under controlled illuminants, and derives the ISO 17321 colour matrix from that data. Their own measurement pages explicitly show the resulting 3×3 colour matrix alongside the SMI value.
https://www.dxomark.com/dxomark-camera-sensor-testing-protocol-and-scores/
That is why their M11 and Z8 can end up extremely close, 84 versus 83 under D50, even though our measurements reveal a much larger difference in colour separation.
Our test is deliberately more demanding.
Instead of evaluating the camera through an 18-colour ColorChecker and a simple 3×3 matrix, we use the much larger ColorChecker SG set and evaluate the camera after applying the final profiled transform that we actually use. In our case, the relevant SG set contains 84 chromatic colours, so we are probing a substantially larger portion of colour space and a much wider variety of spectral relationships.
There is another important difference: ISO-style SMI is historically tied to the CIELAB ΔE76 metric, whereas for our profiling comparison we use CIEDE2000. CIEDE2000 is considerably more perceptually refined, particularly in regions where plain Euclidean distance in Lab does not correspond well to what the eye actually sees.
So I would not expect the numbers to rank cameras identically.
DXO's SMI is essentially asking: “How well can this sensor be mapped through an ISO 17321-style linear colour transform over this relatively small reference set?”
Our profiled SG measurement is asking something closer to: “After giving the camera the best practical colour characterisation we can build, how well does it still distinguish and reproduce a much broader set of real surface colours?”
And that distinction is exactly why the M11 and Z8 can look almost identical in DXO's SMI while separating much more strongly in our measurements.
In fact, DXO themselves note that SMI for DSLRs typically sits in a fairly narrow range and is “not very discriminating”, which is one reason they do not include it in the overall DxOMark score.
https://www.dxomark.com/glossary/color-depth/
2-
With Spectre profiles, that variable is essentially removed from the comparison.
Spectre lets me work deterministically with the known illuminant, independently of Adobe’s own WB calibration system, which is often not perfectly aligned between different camera models. So the Leica and Nikon files are matched to the same illuminant rather than relying on each camera’s As Shot value or Lightroom’s interpretation of a nominal number such as 5500K.
The comparison is therefore controlled for the relevant variables as far as possible. The only meaningful variable left is the lens, and in my experience, in this specific case, that is not enough to explain the difference we are seeing.
Regarding the Graphic Converter test, I would be careful drawing conclusions from it. An automatic correction algorithm is free to alter white balance, contrast and colour relationships globally until the image looks more “neutral” or closer to another rendering. The fact that it makes the Leica image look more like the Nikon does not necessarily mean the original difference was caused by WB.
In this case the important point is that the original RAW files were already matched using the same known illuminant through Spectre, so the remaining difference is not simply one camera being warmer and the other cooler.
I greatly appreciate this work. Much more useful than many other aspects of 'feature testing'.
Two thoughts:
1 - I am wondering how this relates to similar SMI measure e.g. done by DXO labs - they show cameras much more closely together. Example: https://www.dxomark.com/Cameras/Leica/M11---Measurements
M11 is shown (Color Response tab) with (ISO 17321-based) SMI 84, Nikon Z8 gets SMI 83 (less accurate) https://www.dxomark.com/Cameras/Nikon/Z8---Measurements
2 - Regarding the 'real life' foliage shots, it is hard for me to tell if this could also be related to white balance, Leica warmer (more yellow), Nikon coolor (more blue)? I have found ISO settings in raw converters to...Show more →
1-DXOMARK is measuring something related, but not directly equivalent to what we are showing.
Their SMI figure is based on ISO 17321. In their published protocol, DXO uses the classic GretagMacbeth ColorChecker with 18 chromatic patches, measured from RAW under controlled illuminants, and derives the ISO 17321 colour matrix from that data. Their own measurement pages explicitly show the resulting 3×3 colour matrix alongside the SMI value.
https://www.dxomark.com/dxomark-camera-sensor-testing-protocol-and-scores/
That is why their M11 and Z8 can end up extremely close, 84 versus 83 under D50, even though our measurements reveal a much larger difference in colour separation.
Our test is deliberately more demanding.
Instead of evaluating the camera through an 18-colour ColorChecker and a simple 3×3 matrix, we use the much larger ColorChecker SG set and evaluate the camera after applying the final profiled transform that we actually use. In our case, the relevant SG set contains 84 chromatic colours, so we are probing a substantially larger portion of colour space and a much wider variety of spectral relationships.
There is another important difference: ISO-style SMI is historically tied to the CIELAB ΔE76 metric, whereas for our profiling comparison we use CIEDE2000. CIEDE2000 is considerably more perceptually refined, particularly in regions where plain Euclidean distance in Lab does not correspond well to what the eye actually sees.
So I would not expect the numbers to rank cameras identically.
DXO's SMI is essentially asking: “How well can this sensor be mapped through an ISO 17321-style linear colour transform over this relatively small reference set?”
Our profiled SG measurement is asking something closer to: “After giving the camera the best practical colour characterisation we can build, how well does it still distinguish and reproduce a much broader set of real surface colours?”
And that distinction is exactly why the M11 and Z8 can look almost identical in DXO's SMI while separating much more strongly in our measurements.
In fact, DXO themselves note that SMI for DSLRs typically sits in a fairly narrow range and is “not very discriminating”, which is one reason they do not include it in the overall DxOMark score.
https://www.dxomark.com/glossary/color-depth/
2-
With Spectre profiles, that variable is essentially removed from the comparison.
Spectre lets me work deterministically with the known illuminant, independently of Adobe’s own WB calibration system, which is often not perfectly aligned between different camera models. So the Leica and Nikon files are matched to the same illuminant rather than relying on each camera’s As Shot value or Lightroom’s interpretation of a nominal number such as 5500K.
The comparison is therefore controlled for the relevant variables as far as possible. The only meaningful variable left is the lens, and in my experience, in this specific case, that is not enough to explain the difference we are seeing.
Regarding the Graphic Converter test, I would be careful drawing conclusions from it. An automatic correction algorithm is free to alter white balance, contrast and colour relationships globally until the image looks more “neutral” or closer to another rendering. The fact that it makes the Leica image look more like the Nikon does not necessarily mean the original difference was caused by WB.
In this case the important point is that the original RAW files were already matched using the same known illuminant through Spectre, so the remaining difference is not simply one camera being warmer and the other cooler.
Sep 05, 2026 at 11:14 AM
p.1 #13From Cobalt, Leica M vs world.
Seabassius wrote:
For a limited time, I'll be accepting trades of an M10-R for my MP240.
And yes, the M240 is beautiful, but I had completely forgotten about those streaks in the shadows once you get into what are now basically medium-high ISOs.
At least the M9 just gives me big random blobs of noise instead.
For a limited time, I'll be accepting trades of an M10-R for my MP240.
And yes, the M240 is beautiful, but I had completely forgotten about those streaks in the shadows once you get into what are now basically medium-high ISOs.
At least the M9 just gives me big random blobs of noise instead.
Sep 05, 2026 at 11:25 AM
p.1 #14From Cobalt, Leica M vs world.
How exactly you test this metric score? First of all, I feel M11 and M11p has one of most neutral base Leica color out of all Leica M camera I used.
I also find for the same scene, use manual white balance, by switch lens, I got very different output. Is there a way you can directly output sensor CFA info without lens attached?
I also find for the same scene, use manual white balance, by switch lens, I got very different output. Is there a way you can directly output sensor CFA info without lens attached?
Sep 05, 2026 at 11:34 AM
p.1 #15From Cobalt, Leica M vs world.
zhangyue wrote:
How exactly you test this metric score? First of all, I feel M11 and M11p has one of most neutral base Leica color out of all Leica M camera I used.
I also find for the same scene, use manual white balance, by switch lens, I got very different output. Is there a way you can directly output sensor CFA info without lens attached?
I agree that lenses (coatings) matter - Leica vs. Zeiss (on M) vs. Sigma (on SL) -> going in a similar direction as the example above "Leica vs. Nikon"
How exactly you test this metric score? First of all, I feel M11 and M11p has one of most neutral base Leica color out of all Leica M camera I used.
I also find for the same scene, use manual white balance, by switch lens, I got very different output. Is there a way you can directly output sensor CFA info without lens attached?
I agree that lenses (coatings) matter - Leica vs. Zeiss (on M) vs. Sigma (on SL) -> going in a similar direction as the example above "Leica vs. Nikon"
Sep 05, 2026 at 11:42 AM
p.1 #16From Cobalt, Leica M vs world.
This nerdy deep dive is activating my old DIT neurons. I love it.
Sep 05, 2026 at 11:58 AM
p.1 #17From Cobalt, Leica M vs world.
Ulysseita wrote:
Spectre lets me work deterministically with the known illuminant, independently of Adobe’s own WB calibration system, which is often not perfectly aligned between different camera models. So the Leica and Nikon files are matched to the same illuminant rather than relying on each camera’s As Shot value or Lightroom’s interpretation of a nominal number such as 5500K.
To be fair, I had watched that video and it left me with as many questions as before.
- The 'known illuminant' concept may work for artificial light, but the natural light spectrum continuously changes throughout the day - so "cloudy" etc. is just a rough preset AND is hard coded in ACR (in your example) based on Adobe profiles (e.g. Capture One has different conventions for similarly named illuminant profiles -> both circumvent a technical term like D50).
- In the YT video above you keep clicking through illuminant profiles to see what sticks and then fiddling with sliders. We all do that (nothing wrong) -> but even after the video I don't understand how Spectre makes that guessing game easier?
Lastly, as I have told you I am a big fan of Spectre profiles and want them for all my cameras - your method in producing them results in superior perceived (by me) color quality for my images. This seems to confirm that your approach is effective and sound.
Spectre lets me work deterministically with the known illuminant, independently of Adobe’s own WB calibration system, which is often not perfectly aligned between different camera models. So the Leica and Nikon files are matched to the same illuminant rather than relying on each camera’s As Shot value or Lightroom’s interpretation of a nominal number such as 5500K.
To be fair, I had watched that video and it left me with as many questions as before.
- The 'known illuminant' concept may work for artificial light, but the natural light spectrum continuously changes throughout the day - so "cloudy" etc. is just a rough preset AND is hard coded in ACR (in your example) based on Adobe profiles (e.g. Capture One has different conventions for similarly named illuminant profiles -> both circumvent a technical term like D50).
- In the YT video above you keep clicking through illuminant profiles to see what sticks and then fiddling with sliders. We all do that (nothing wrong) -> but even after the video I don't understand how Spectre makes that guessing game easier?
Lastly, as I have told you I am a big fan of Spectre profiles and want them for all my cameras - your method in producing them results in superior perceived (by me) color quality for my images. This seems to confirm that your approach is effective and sound.
Sep 05, 2026 at 11:59 AM
p.1 #18From Cobalt, Leica M vs world.
Ulysseita wrote:
And yes, the M240 is beautiful, but I had completely forgotten about those streaks in the shadows once you get into what are now basically medium-high ISOs.
At least the M9 just gives me big random blobs of noise instead.
Perhaps Leica mitigated the banding issues in the 262 series when removing Live View, video processing, and EVF support.
The M-D, in particular, lacks a rear screen entirely, meaning the camera's internal circuitry runs significantly cooler. This maintains a cleaner noise floor, which I think naturally keeps that banding suppressed during regular shooting.
I remember shooting an M240 many years ago, and the pushed shadow areas not only had banding but also a stubborn green color shift.
Of course, at very high ISOs like 6400 (for this sensor tech), files from both the M240 and M262 (including the M-D) start to really fall apart. Fortunately, Lightroom's Denoise minimizes this a lot today and this algorithm keeps getting better and better. I've even seen it remove some banding as well now.
Seabassius wrote
For a limited time, I'll be accepting trades of an M10-R for my MP240.At least the M9 just gives me big random blobs of noise instead.
Perhaps Leica mitigated the banding issues in the 262 series when removing Live View, video processing, and EVF support.
The M-D, in particular, lacks a rear screen entirely, meaning the camera's internal circuitry runs significantly cooler. This maintains a cleaner noise floor, which I think naturally keeps that banding suppressed during regular shooting.
I remember shooting an M240 many years ago, and the pushed shadow areas not only had banding but also a stubborn green color shift.
Of course, at very high ISOs like 6400 (for this sensor tech), files from both the M240 and M262 (including the M-D) start to really fall apart. Fortunately, Lightroom's Denoise minimizes this a lot today and this algorithm keeps getting better and better. I've even seen it remove some banding as well now.
Sep 05, 2026 at 12:26 PM
p.1 #19From Cobalt, Leica M vs world.
mzbe wrote:
To be fair, I had watched that video and it left me with as many questions as before.
- The 'known illuminant' concept may work for artificial light, but the natural light spectrum continuously changes throughout the day - so "cloudy" etc. is just a rough preset AND is hard coded in ACR (in your example) based on Adobe profiles (e.g. Capture One has different conventions for similarly named illuminant profiles -> both circumvent a technical term like D50).
- In the YT video above you keep clicking through illuminant profiles to see what sticks and then fiddling with sliders. We all...Show more →
I think there are two separate things being mixed together here.
First, “deterministic WB” does not mean that Spectre magically identifies the exact illuminant for you.
You are absolutely right that natural daylight changes continuously throughout the day, and that a label such as “Cloudy” can only ever represent a reference condition, not every possible daylight spectrum.
What deterministic means in this context is that, once the illuminant is known or selected, the colour transform has a defined physical meaning and behaves consistently across cameras.
With the normal Adobe workflow, 5500K on camera A is not necessarily equivalent to 5500K on camera B. The relationship between temperature, tint and the underlying camera calibration can be different, which is exactly what Fred was seeing with the M10-R and M11.
Spectre removes much of that ambiguity because the illuminant behaviour is calculated from the measured spectral response of each camera and anchored to known reference illuminants.
If I know I am under daylight, I don't need the camera to guess the WB for me. If I don't know the illuminant, or I am dealing with mixed light, then of course some judgement is still required. No profiling system can recover the exact spectral power distribution of an unknown light source from a single RGB photograph.
In the video, when I move between illuminant settings, I am showing how those different physically defined starting points behave. It isn't meant to suggest “keep clicking until something looks nice”. If the illuminant is known, there is no guessing game. If it is unknown, Spectre gives you much better-defined reference points from which to work.
Regarding the M11 appearing very neutral: that is perfectly possible and does not contradict the hardware-separation score.
A camera can have an excellent neutral rendering after profiling while still having weaker intrinsic spectral separation. Profiling can move the colours it captured into very accurate positions. What it cannot do is recreate two spectral stimuli that the CFA has already collapsed into nearly the same RGB response.
That is the distinction between rendering accuracy and hardware colour discrimination.
And yes, lenses can alter colour because their transmission and coatings are not spectrally neutral. But that is a different variable again. A hardware CFA/separation measurement should be based on controlled spectral sensitivity data, not on a JPEG, Adobe rendering or an arbitrary lens/scene combination.
So changing lenses can absolutely change the colour of a photograph, but it does not change the intrinsic spectral separation capability of the sensor itself.
At the end I completely agree with you that Spectre does not remove the need to identify an unknown illuminant. What it removes is the additional uncertainty of having the same supposed illuminant interpreted differently by different cameras and profiles.
Ulysseita wrote
Spectre lets me work deterministically with the known illuminant, independently of Adobe’s own WB calibration system, which is often not perfectly aligned between different camera models. So the Leica and Nikon files are matched to the same illuminant rather than relying on each camera’s As Shot value or Lightroom’s interpretation of a nominal number such as 5500K.- The 'known illuminant' concept may work for artificial light, but the natural light spectrum continuously changes throughout the day - so "cloudy" etc. is just a rough preset AND is hard coded in ACR (in your example) based on Adobe profiles (e.g. Capture One has different conventions for similarly named illuminant profiles -> both circumvent a technical term like D50).
- In the YT video above you keep clicking through illuminant profiles to see what sticks and then fiddling with sliders. We all...Show more →
I think there are two separate things being mixed together here.
First, “deterministic WB” does not mean that Spectre magically identifies the exact illuminant for you.
You are absolutely right that natural daylight changes continuously throughout the day, and that a label such as “Cloudy” can only ever represent a reference condition, not every possible daylight spectrum.
What deterministic means in this context is that, once the illuminant is known or selected, the colour transform has a defined physical meaning and behaves consistently across cameras.
With the normal Adobe workflow, 5500K on camera A is not necessarily equivalent to 5500K on camera B. The relationship between temperature, tint and the underlying camera calibration can be different, which is exactly what Fred was seeing with the M10-R and M11.
Spectre removes much of that ambiguity because the illuminant behaviour is calculated from the measured spectral response of each camera and anchored to known reference illuminants.
If I know I am under daylight, I don't need the camera to guess the WB for me. If I don't know the illuminant, or I am dealing with mixed light, then of course some judgement is still required. No profiling system can recover the exact spectral power distribution of an unknown light source from a single RGB photograph.
In the video, when I move between illuminant settings, I am showing how those different physically defined starting points behave. It isn't meant to suggest “keep clicking until something looks nice”. If the illuminant is known, there is no guessing game. If it is unknown, Spectre gives you much better-defined reference points from which to work.
Regarding the M11 appearing very neutral: that is perfectly possible and does not contradict the hardware-separation score.
A camera can have an excellent neutral rendering after profiling while still having weaker intrinsic spectral separation. Profiling can move the colours it captured into very accurate positions. What it cannot do is recreate two spectral stimuli that the CFA has already collapsed into nearly the same RGB response.
That is the distinction between rendering accuracy and hardware colour discrimination.
And yes, lenses can alter colour because their transmission and coatings are not spectrally neutral. But that is a different variable again. A hardware CFA/separation measurement should be based on controlled spectral sensitivity data, not on a JPEG, Adobe rendering or an arbitrary lens/scene combination.
So changing lenses can absolutely change the colour of a photograph, but it does not change the intrinsic spectral separation capability of the sensor itself.
At the end I completely agree with you that Spectre does not remove the need to identify an unknown illuminant. What it removes is the additional uncertainty of having the same supposed illuminant interpreted differently by different cameras and profiles.
Sep 05, 2026 at 12:55 PM
p.1 #20From Cobalt, Leica M vs world.
Ulysseita wrote:
80.7 and 80.8 mean that Leica, after the mess of the M8 and the whole IR story, had not only fixed the problem,...
Except they didn't.
I've had the M8 which had the IR issue, then the M9 (which didn't or practically negligible) then the M240 which annoyed the hell out of me because IR contamination was still an issue, although nowhere near as bad as the M8, but enough to ruin shots when you least expect it.
Sep 05, 2026 at 02:05 PM