Color Management/custom profiles at capture - slides

Hello,

I am digitizing many thousands of professional slides from a photographer’s archive. How do people implement color management? (ex: X-rite Colorchecker, LaserSoft IT8 transparency/reflective targets which sounds extra useful for me, etc).

I would like to make color profiles for the camera + lens + light source + film stock but not sure where to begin.

I am struggling to get a strong common baseline that does not take a TON of work to only barely come close to match the original, which most professionals shooting slides would want/prefer. So I think color profiles for consistency is the next best step. Mostly Kodachrome, Ektachrome and Provia, more rarely Provia 400 or a few others. Drum or flatbed scanning is not realistic for time purposes, and the selects will likely be rescanned much more critically later.

My gear and set up:
Canon R5, EF 100 2.8L and Sigma 70mm ART, custom built Kodak projector/scanner (for bulk speed) and vertical copy stand set up (for more precision), both with 96-98ish CRI lighting. Epson V550 that I… dislike.

Extra thought - Is there any advantage to this with color negatives? What would the process be there? I’ve been digging through the forum but all this stuff gets glanced over in places. Happy to be pointed to a whole thread!

New to participating in NLP forum but long-time lurker. Cheers!

1 Like

The issue you are raising is twofold. On the one hand it is easy to set up a DLSR scanning pipeline which would give reasonably good results. On the other it is exceedingly difficult to achieve color (and tonal) reproduction of a high quality and in repeatable manner. The point in the range between “reasonably good” and “high quality”, where one finds satisfaction is a completely subjective matter, and is a very non-linear function of the effort, research, and expense.

Given the mention of “many thousands” of “professional” slides, one major pitfall in such an endeavour would be going too far too fast without enough preparation. The risk is having to redo the mechanical part of the work if something was done wrong early in the pipeline. It would be very beneficial to get a solid understanding of color management first, although this might be impractical. The subject is very deep, difficult, and not even complete. One book able to give a solid grasp on the topic is Hunt’s Reproduction of Color.

More to the point of your questions from the practical perspective, a couple of probably the best resources on the camera profile creation procedures are Basic workflow for making an ICC profile using a test target and How to Make a Better Custom Camera Input Profile. Either is not specific to DSLR scanning but can easily be extrapolated. Also the surrounding pages are very worthy companion reading. It is advisable to use IT8 transparency targets, preferably made on film similar to what is going to be scanned. In fact, without such calibration the whole quality affair is a moot point. The targets are typically sourced from http://www.targets.coloraid.de/.

The essence of the scanning procedure then becomes quite simple.

  • Create an ICC profile using exact same scanning and conversion procedure as to be used for the slides.
  • Scan a slide;
  • Apply the ICC profile;
  • Convert to the preferred working space;
  • Apply a look profile / correction;
  • Make individual corrections on per image basis.

There are inherent difficulties on each step, some of which are generic, while others are specific to DSLR scanning, particularly, of slides. Here are several important factors which may make gauging the expectations more grounded in reality.

  • Scanning slides involves extracting a higher dynamic range. While modern cameras have much wider sensor dynamic range, the issue comes from the optical path. Due to the high image contrast, issues such as internal flare, chromatic, and other aberrations, affect the result significantly. Flare can be alleviated by adjusting the blackpoint, but the aberrations can only be reduced somewhat. For example, the purple/green fringing of chromatic aberration can be reduced by the lens profile, but only at the cost of softening the local contrast in the affected areas. Because of this, consumer-level DSLR scanning cannot truly compete with drum scanning, although does present a very meaningful compromise.

  • The extracted resolution is a subjective question. A ~45Mp sensor is a reasonable match to a regular 35mm film frame. In some cases film can record much more, and due to the Bayer mosaicing used by most sensors, the effective resolution of the sensor is less, especially in relation to color.

  • While it is possible (and desirable) to calibrate the scanning flow for a particular film stock, factors such as irregular development, storage conditions, and film aging make such calibration less useful and potentially harmful for non-matching images. Batch variation within the same stock can also be a factor.

  • Existing software solutions, including NLP, can only approximate the look of an image based on some knowable / measurable data points. The results are often initially pleasing, but not in comparison to a calibrated conversion. Also, the results may not be repeatable for the same image on different versions of the software.

  • Exact color reproduction of a scanned negative film, as of now (2022), is an unsolved problem. Calibrated approach still makes sense, but the difficulty is in the inversion. Here is an experimental and difficult to reproduce method which represents the cutting edge research in this area: GitHub - arufahc/negicc: ICC Profile for Color Negative Film. The document referenced there can also serve as a measure of sorts of one’s color management mastery. If one can understand all the reasoning in this text (together with its involved background), then one’s grasp of the color management could be considered adequate. The other mentioned factors are equally at play with the color negative DSLR scanning.

With all that said, the approach of quickly scanning everything for preview, ensuring some baseline quality, and subsequent rescanning of selects, that approach is sound. IT8 transparency targets, the referenced guides, and some open source tools like dcRaw and Argyll can make it all very achievable.

3 Likes

The “beauty” of reproducing slides is, that you have the original slides that you can check against the reproduction.

Slides can be dense and dynamic range as well as noise could be the one thing to watch carefully.

My usual recommendation is to play around (test) with a bunch of representative slides with varying exposure, contrast and density, and try things like multishot stitched images and/or stacked images for reduced noise and higher dynamic range, all of which has nothing to do with Negative Lab Pro, which is not needed for reproductions of positives, unless you try this:

1 Like

This is exactly the sort of detailed response I hoped was possible. Thank you so much! Sometimes researching online on your own leads to a lot of old or outdated info, knowledge or software that doesn’t exist or is not suited to camera scanning or today’s technology, etc.

I am not looking for perfect reproduction of 25-40 year old slides (they were stored very nicely though!), just a better way to do it than simply eye-balling it every single time since there are so many! This is all very helpful.

As my grandfather used to say, “you’re a gentleman and a scholar and I don’t care what your sister says about you!” He was a funny man.

Thanks again.

1 Like

Thank you for the tips, much appreciated. The dense slides are a challenge, as are the not dense/very thin highlights in those particular slides! Doubly so if they are a high contrast scene. So much to learn about slides that has not been a challenge, per se, with negatives.

Stacking them is something I had not considered, like stacking astro images. I will look into it. In that case is it expose for the highlights? Or perhaps shoot a bracket and combine, HDR style?

The answer depends on your images and your gear. I’d give both lenses a try and would expect the Sigma to render more evenly. As for exposure/bracketing etc., your source material will teach you.

Take a few shots of different slides that exhibit extremes as well as not and post them here…if you want hints that relate to something other than to thin air.

1 Like

Well, you have thousands of slides and they are old slides, presumably from a variety of different films, the colours may have faded and there probably won’t be an IT8 target for some/many of the films you are trying to copy, you may not even know what films they were shot on. Although there are several easily accessible ways to make a camera profile for a given lighting setup using standard software (Lightroom, Capture One etc.) the only ones I know use reflected targets such as the Macbeth Color Checker or X-Rite Passport, or even Passport SG.

It is generally only dedicated scanner software such as Silverfast or Vuescan that can make use of the IT8, so-called “IT 8.7 or ISO 12641 compatible software”. I do have a friend who has succeeded in making his own profiles using bespoke colour profiling software but it certainly wasn’t straightforward though it was effective, and you’ll also find that the software required can be pretty expensive. In fact the only mention of such a profile used for camera scanning that I’ve seen is from the high end DT Heritage where they include a profile for their light sources in their dedicated software, but that’s a very expensive system of course. As you’ve found out it doesn’t seem to get discussed in digital camera scanning forums. Now that might be because we just want to get the best from our slides and not be tied down to the colour produced on film, and that’s an equally legitimate perspective, we’re not trying to reproduce the Mona Lisa.

I’d definitely take Digitizer’s advice and seek out a selection of potentially ‘difficult’ slides, dense, thin, faded etc. and then see how, if at all, they fall short of your expectations. I’d also settle on one lens and maybe use Vlads Test Target to choose between them, it will be fine for this, and may also show up any potential problems with flare etc. Of course you’ll be shooting on RAW so provided the technical aspects are correct the colour can be sorted at any time.

1 Like

I’ll just add one more point whilst I think of it. For absolute perfection you would need to remove the slides from their mounts and photograph them held completely flat, under Museum glass perhaps. Do you really want to be doing this? Personally I think it’s kind of a shame to do this with a family archive, particularly with those lovely Kodachrome card mounts. That’s where stacking comes in as an alternative I suppose but you don’t want to make the whole process so onerous that you never complete the project.

1 Like

Just a point re the dense areas of transparencies or negs - these are far better reproduced with a camera based digitising setup than with a scanner. With a scanner the amount of light that can be squeezed through the dense area is fixed by the scanner beam speed and brightness and this leads to noisy results or lost data. With a camera based system this can be increased without practical limit. If necessary a separate exposure for these areas can easily be HDR combined with exposures for the other areas.

I have recently digitised thousands of slides with my camera set-up and am constantly amazed at the amount of information that can be extracted from the dense areas, particularly Kodachrome. I used to produce slide presentations and project them with four Carousel SAV2000s combined with two Electrosonic dissolve units on twin screens. I was pretty impressed with the results.

Now with a digital projector and the visual options available for presentation with an application such as “Keynote” I am “blown-away” with the image quality and sharpness that has been previously unseen in those transparencies. They compete incredibly well with current photos shot on a digital camera!

As you seem not to be an easily fend off type, and also because of your funny grandpa, let’s elaborate some bits.

Bracketing could be useful and yes, the exposure should be for highlights. Depending on your camera’s ISO performance it may not be necessary to actually shoot several exposures. Would be sufficient to just convert the raw file with different exposure compensations and merge those. I’m not familiar with R5 but as a recent camera it should easily handle +2 stops compensation without making the noise any more apparent. Naturally, use the minimal native ISO when shooting.

Unfortunately, the issues of a high contrast input do not disappear with this approach. Diffraction doesn’t go anywhere with any kind of area sensor. Relatively even areas where the density differences are small should come out fine, but areas of high local contrast should be watched for flare spillover. The only truly efficient alternative is dot scanning, which drum scanners use. One thing to try would be to use a higher-than-target scanning resolution with a lens having higher than 1:1 ratio and then resize the image down.

This is where the rigs based on scanner lenses come into play. These have way better performance than consumer macro lenses, but often can’t capture the whole film frame at once, meaning stitching - another dimension of complexity.

Evenness does not need to be a factor. Use flat field correction both when preparing the profile and when shooting the slides. In addition to the lens vignetting this will also take care of any unevenness of the light source.

The Canon lens is fine, it should give around 4200-4600 ppi on the R5 sensor, which is likely sufficient. The Sigma one is likely sharper, but I wasn’t able to find any measurements.

While it is true that IT8 transparency targets are typically used in traditional scanning, there is nothing that prevents one from using them with DSLR as we will shortly demonstrate. For making an ICC profile the target doesn’t need to be a particular one. ColorChecker, IT8, reflection, transparency, anything. One can take a sheet of paper, put some oil and acrylic paints on it in squares, and call it a target. As long as the patches were measured with a spectro, and the target layout description file exists, it will work fine. A profile can be created and will be very much more useful for images of the same kind than no profile at all.

The procedure itself is not difficult, but one does need to do the homework. The software is open source, free, modern, high quality, and comes with a decent amount of community support. One can get it here https://www.argyllcms.com, although this is not the only choice available.

Just to avoid further confusion, reproduction is a completely different problem domain with dissimilar constraints. What we all are mostly going for is a faithful interpretation of the look, which is already a very tall order if we consider the technicalities.

Without calibration it really can’t.

Flatness is a valid concern and it does need to be addressed. There are various approaches, e.g. ANR glass backing, all kinds of scanning frames, and the suggested focus stacking. Given that you (OP) already got a scanning rig, this issue should be familiar. Museum glass, as well as e.g. polarizing filters, are not advisable as these modify the spectral composition of light and, additionally, may interfere with the spectrum of the lightsource.

Besides, while 96-98 CRI lighting sounds reassuring, its worth it to measure the light composition with a spectro to be aware of the peaks/dips, and, more importantly, know which colors are the outliers. CRI rating calculation has important deficiencies which are rather relevant for scanning.

That mostly depends on the scanner and the software. The usual film scanner choice, the Epson family of scanners culminating with V850 and all the way back to 4990 in combination with VueScan allows modifying the (real, analogue) scanning exposure and offers multiscan (i.e. HDR). The deficiency of flatbed scanning is really not the exposure control, and not even the single-pass dynamic range. It is the low resolution (~2200 ppi) and the wait. That’s the primary reason DSLR scanning beats flatbeds. A ~45Mpx camera with a decent 1:1 macro lens achieves around 4600 ppi and is much faster, that’s all.


Before we look at an example of DSLR calibration using a transparency target, let’s stop for a moment and consider all the steps one needs to perform in this seemingly fast way of scanning.

Preparatory steps

  • Take a flat field capture (at the beginning of each scanning session)
  • Prepare the ICC profile (once for a DSLR / lens / target / lightsource combination)

Scanning steps, for each frame

  • Capture a normal single exposure
  • Do focus bracketing (unless the alignment is good)
  • Do exposure bracketing (unless after-the-capture compensation gives good results)

Processing steps, for each capture

  • Interpret the RAW
  • Apply the flat field correction
  • Apply the calibration profile
  • Convert to a working space

Processing steps, for each frame

  • Merge the exposures
  • Merge the focus brackets
  • Save the master scan

Clearly, this all doesn’t look like something that should be done manually, especially when we are talking about “many thousands” of images. A meaningful automation seems to be a must, and that means scripting. That, in turn, means command line tools. This is another area where open source tooling beats commercial software pretty much every time. However, it also means more homework. This brings us back to the question of personal choice. Oh, well.


Now for the promised example. Here is a capture of a transparency IT8 target on a 35mm slide, this is how ACR sees it. Note the black background and some details at the sides. The background is a piece of black cardboard with a 35mm hole in it, where the target is inserted. The details at the sides are where the frame area turned out to be bigger than the piece of cardboard, should have been avoided, but not critical. By the way, this was shot through a regular zoom non-macro lens, just for fun. That’s why the target area only takes a fraction of the frame. The backlight is ~D50 sunlight.

Using dcraw, convert the image to a tiff:

dcraw -v -T -o 5 -j -M -6 -W -g 1 1 -w -c chart-capture.nef > chart-xyz.tiff

Explanation of the options can be easily found online. The important bit is the raw gets converted with the linear gamma. Note how because of that the result looks much darker:

Then, for an easier target detection, I cropped the image. This is not strictly necessary, but makes things simpler:

Now take the measurements from the scanned capture. scanin is a tool coming with ArgyllCMS. Again, the explanation for all the switches can be found online.

scanin -v -dipoan -G 1.0 -p chart-crop.tif it8Wolf.cht F210418.cie

The it8Wolf.cht is the target layout desctiption file, comes with ArgyllCMS. The F210418.cie is the target batch measurements data file, available from the vendor of the target. The result is an .ti3 file, not yet directly usable. In the meantime let’s check the diagnostic output of the target detection:

Everything seems to be located nicely.

Finally, creating the ICC profile. This is a matrix profile, which is preferred over LUT profiles on small numbers of patches:

colprof -v -D"My Profile" -qm -am -u chart-crop

The profile gets placed to the appropriate system folder, Photoshop gets started up, and here is the image with the profile applied:

Similarly the profile can be applied to any other frame, scanned in the same conditions. Is this profile any good? Let’s check:

profcheck -v2 -k chart-crop.ti3 chart-crop.icc
> Profile check complete, errors(CIEDE2000): max. = 5.832972, avg. = 1.372846, RMS = 1.638548

For a matrix profile of medium quality on a limited number of patches (IT8 is not a big target) the results are quite fine. Subjectively speaking, the image could be brighter, but that’s down to the original target exposure, and also to the -u setting of colprof. This disables the automatic exposure correction by the profile, personal preference.

As you can see, the procedure is not excessively difficult, and in fact, is easily automatable.

1 Like

slow clap Wow, once again thank you for the thorough response. I am blown away and will have to think on this and analyze it a few times to fully absorb all of it!

Thank you!

Also, I will post some examples soon. I finally got my settings recipe dialed in for aperture priority on the R5 + lens + light source for bulk slides. This has had a very, very helpful effect on limited post time for more accurate color.

  • center-weighted avg. metering,
  • f.5.6 ish (up to +/- 2/3 if it is flat or needs DoF)
  • min shutter 1/125, max any
  • +2/3 EV,
  • D+ (highlight protection, forces ISO 200 which is a sacrifice I make for volume/sped, could be talked out of this).
  • not framed at 1:1, more like 1:1.7 or so? Limited by projector + lens combo. Yields a 24MP crop but can get to 30MP if I push closer.
  • and using Adobe Neutral profile and controlling contrast and vib/sat sliders later
  • they all get -1/3 to -1 exposure and 0-30% highlights and they are rock solid looking.
  • minor tweaks to shadows to taste

Still want to make color profiles though to be discerning.

This gives a strong, fast automatic exposure using ETTR, great shadow detail/noise, a good baseline on color, and never blowing highlights. Also works great for the occasional -1.3/0/+1.3 bracket on problem slides. Should probably go even more ETTR after deeper analysis digging into the RAW files, too. Camera histogram is a liar as we know. I get around 700-1500 prepped slides scanned per hour this way depending on my personal focus. Selects get made and those are scanned vertically at 1:1 45MP.

@nicnilov I am pretty happy, generally with my light source for my bulk scans on my Dr. FranKodakenstein Projectoscanner™ (still workshopping this one :crazy_face: ) but I have bought some 35W/20W Solux halogen bulbs to test for closer to 100CRI spectrum in the future after I am done cam scanning the first wave here.

Generally will be trying to increase my my light, alignment and lens quality as time goes on, and especially interested in jumping from 14bit to 16bit at some point with a higher MP count for medium format stuff I have to do.

Though, generally lab and at-home testing results vs the real world execution of such things are often just differences of degree rather than magnitude. So perhaps I should be happy for awhile and not let perfection be the enemy of the good (or even great). Time will tell.

For more on some lenses I’ve recently come across a few solid links, 1 of which I’ve posted on NLP forum elsewhere. The other two, @Harry posted in another thread and were quite surprising to me. These focus on 35mm 1:1 reproductions for camera scanning.

1- 1X LENS TEST 2020 — Close-up Photography
2- 1x Macro Lens Test 2022 www.Closeuphotography.com — Close-up Photography
3- How Good a Macro Lens do you Really Need for DSLR/Mirrorless Camera Scanning? - pixl-latr

Very interesting to read your method, as you would expect it’s similar to the one my friend uses except that he uses different software. I wonder if you could please explain what bearing the choice of IT8 target has on this - so Ektachrome, Fuji etc., even Kodachrome when such targets were available?

Would you see a difference in results given that the target might not match the film type of the originals?

Alternatively if you bought each of the currently available 35mm IT8 targets (5 from coloraid.de) would the resulting profiles produce markedly different results?

Thanks

Thanks nicnilov - I’ve been scanning for decades and still found lots of useful info in your post.

But re scanning dense areas I think your points above are more theoretical than practical. I’ve found multipass and analogue gain to give very little improvement in dense area repro. The speed of the scan beam and the max brightness are fixed and that puts a limit on what can be achieved this way whereas a DSLR exposure has no limit.

The difference I see with a DSLR is orders of magnitude better - and, of course, far quicker. Modern DSLR sensors also have a far greater dynamic range and, while that can to an extent be improved with the above techniques it still doesn’t hold a candle - or a candela :slight_smile: - to the DSLR output.

Generally this all looks valid, especially for preview bulk scanning. One point to consider is the automatic exposure. Is there really a reason for it? The maximum needed exposure is defined by the deepest shadows, which on slide film are within density 4.0. The minimum exposure is defined by the density of a clear film base, 0.05D. Density range of 4.0 in linear contrast terms is 10000, which in dynamic range terms is about 13.2 stops.

Your camera’s dynamic range at ISO 50 is 13.33Ev, which is a tight match. It means for the slides having the extreme dynamic range it would be rather difficult (and not desirable anyway) to take a single exposure accurately matching the image, calling for bracketing. For the slides not having such dynamic range any reasonable exposure will do, and it does make some sense to go ETTR.

However, when making a profile it is important for the target exposure to be correct, that is, the histogram reasonably spread and the middle gray being at 18% or 50 L* in L*a*b*. In order for the profile to be applied with precision, the image’s exposure should also be in normal range. Meaning if we take an overexposed image and apply a color correction via the profile, the particular densities of the image will be affected by parts of the profile curves destined for densities that are lower. If there is a significant non-linearity at different parts of the profile curve (which is not unusual), the result can be a misplaced correction.

Because of that it is advisable to scan most frames at a fixed exposure. This gives a more stable baseline, more suitable for profile-based correction.

Can’t comment on D+ without experience, but from the descriptions all it does is selectively boosts gain in shadows by the higher ISO. R5 dynamic range at ISO 200 is 12.18EV, which is a full stop lower than the maximum. I’m not sure this is a good tradeoff, given that the same f/stop can be extracted in post without affecting the noise. It’s easy to see how this would be beneficial for jpegs or video, but not for the raw files. Needs further investigation.

Source: Canon EOS R5 - DXOMARK

Consider using Camera Neutral instead. Adobe has its own views on how an image should look.

Nice choice. Solux can be driven to the vicinity of 5500K but only just, see Making a camera profile with DCamProf. It’s not a big deal, but is something to remember when configuring the profile generation.

Bit depth question is a difficult one. On the one hand, 14bit can be argued to be sufficient. On the other, (monochrome) tonality improvement on 16bit is visible, but still is not comparable with analogue print in smoothness. Also, 14bit in camera specs does not mean 16384 discernible levels. R5 sensor only has about 308 levels per channel, meaning about 29M colors as opposed to the theoretical 4 trillions.

Source: Canon EOS R5 - DXOMARK

Yes. However if you are going to build a profile in the future, keep the scanning setup consistent so that the profile would be applicable to the previously scanned images.

Different films use different emulsions, with result being the difference in color response. This response comprises the “film character”, or “the look”. One and the same object shot on different films may be rendered in a slightly different color. When we are applying a film profile, we are not looking to achieve a “realistic” image, or the look of the object in reality, as this isn’t possible. What is possible is to measure the color response of a particular film (superimposed on the response of the scanning equipment) and then apply it to a scanned frame.

Because the color response of different film brands varies, sometimes significantly, mixing and matching of films and profiles is going to give imprecise results. The deviation will typically be visible, but likely not that significant, and can actually be pleasing, just won’t be that a precise match. In any case, applying a non-matching film profile often is way better than keeping the image without color management. Deviations introduced by the scanning equipment can be far larger than the difference between film profiles.

The results will be different, noticeably but not necessarily significantly. Whether the difference will be definitive depends on the scanning goals.

It seems your are talking about areas so dense, that the scanner’s backlight power at maximum is not sufficient for the sensor to detect the values. This should mean the film density exceeds the scanner maximum density specification (which is not unusual for slides). There is no arguing a scanner has a performance limit, beyond which it just can’t do much. Speed of the scanning is a function of the selected scanning resolution. Therefore, since the backlight flow is continuous, the decisive factor is not the duration of the illuminance but its power. If the power is not sufficient to penetrate a given film density, the sensor will not get any data.

So, while multipass and analogue gain do give very real improvements when the film density is within the scanner specification, naturally, they won’t help much when the density is beyond that.

As a matter of fact, the exposure on an Epson scanner using VueScan can be increased to a complete overexposure. This won’t help with higher density images for a different reason, which is the light scatter within the scanner. The flare becomes so uncontrollable, it doesn’t makes sense to scan at such configuration. This doesn’t happen to the same extent with DSLR, but still remains a major quality factor.

Where/how did you get that information?

Source: Canon EOS R5 - DXOMARK

2^8.27 ~= 308. If dxomark data is to be believed, and if I’m interpreting it correctly.

Strange, compared to what I see here:
https://photonstophotos.net/Charts/DXOPDR.htm#Canon%20EOS%20R5

…and from the same page as you:

You are looking at the dynamic range which is measured in Ev (f/stops, log2). Select the Tonal Range tab on dxomark site. Sorry for not being specific enough before.

The interpretation seems to be correct in terms of what is said in DXOMARK camera sensor testing protocol and scores:

The best image quality metric that correlates with color depth is color sensitivity, which indicates to what degree of subtlety color nuances can be distinguished from one another (and often means a hit or a miss on a pantone palette). Maximum color sensitivity reports in bits the number of colors that the sensor is able to distinguish.

Yet, the chart does look weird, and I’m not sure I quite grasp the difference between the normalized Print and Screen charts on their site, and how the Tonal Range chart correlates with the Color Sensitivity. My statement of 308 levels per channels may be incorrect, but in any case the camera’s capability to distinguish tones is under 10 bits per channel, which is 1024 levels. I can’t substantiate this at the moment though, this was a research I did quite a while ago.

Or maybe it isn’t incorrect after all. Guys over optyczne.pl tested R5 and found this (translated by Google):

In the chart above, we can see that for the lowest native sensitivity, the number of tones reaches around 326, so we get an 8.4-bit data record. Regardless of these criteria, the R5 is unfortunately inferior to its competitors. For example, Nikon Z7 got 9 bits, Panasonic S1R 8.7 bits, Sony A7R IV - 8.8 bits. Increasing the sensitivity obviously degrades the tonal range and at ISO 1600 in the EOS we get the values ​​of 7.1 bits, which gives about 135 tonal transitions. With the maximum available ISO of tonal transitions, we only have 16, which is 4 bits.

The difference between 308 and 326 could be due to the testing methodology.