Test: Effect of ISO on RAW histograms

[kallisti]

There has been some confusion about how ISO affects images shot in RAW. A recent thread about shooting in aperture priority mode got somewhat derailed by claims that ISO is "fake" and applied during RAW conversion, the claim being that one should just shoot at base ISO and then increase exposure in post. It got me thinking so I shot a series looking to objectively assess the impact of increasing ISO on image quality.

I used a color calibration chart as the subject. Not perfect for the test since the squares are discreet, but I don't have a gradient target to use as a test.

The series was shot with a Nikon D850 and 70-200 f/2.8E FL lens @ 190mm. This filled the frame with the color target and didn't require any cropping. The camera was on a tripod and I used a remote release. Not totally vital for this test as it isn't a sharpness test.

I tested from ISO 64 (the base ISO of the D850) to ISO 12,800.

I used an Einstein studio strobe for the light that was positioned roughly 45 degrees off axis and 45 degrees above the target to provide uniform light. The room was otherwise dark, so all of the light was coming from the strobe. All images shot at 1/250th second. In general, I tried to decrease the strobe output by a stop as I increased the ISO by a stop. Early images shot at f/8, but as I increased the ISO I reached the lower limits of the strobe's output, so had to stop down the lens to keep exposure constant. ISO 3200 was the breakpoint where I couldn't decrease the strobe output further and stopped down to f/11. ISO 6400 was at f/16. ISO 12800 was at f/22.

For ISO 3200, ISO 6400, and ISO 12800 I also took a shot at ISO 100 with the same flash output to test the hypothesis of whether it is better to use a higher ISO value, better to use ISO 100, or whether it is a wash between the two. Where the ISO 3200 pic was properly exposed, the ISO 100 pic with the same light reaching the sensor was 5 stops underexposed at the time of capture. 6 stops for the ISO 6400 pic and 7 stops for the ISO 12800 pic.

I used RawDigger on the RAW files to graphically quantitate the results. For those not familiar with the program, it can display a histogram with EV on the X-axis and number of pixels registering that EV value on the Y-axis. Ideally, one wants this graph to be continuous without any gaps. Gaps indicate areas where there is no pixel information for that specific EV value. Assuming that the subject has somewhat continuous color tonal values, any gaps (specifically ones where there is zero pixel information present) imply that the sensor failed to capture useful information that was actually present in the subject. Gaps are bad. More gaps are bad. Larger gaps are bad (like, really bad). They can result in banding or other artifacts. They represent gaps in the tonal continuity of the image as captured by the sensor. Bad, bad, bad.

The RawDigger histograms let you see how all the pixels on the sensor responded to the light they received in each color channel. The overall area under the curve is a graphical representation of how all the pixels recorded light at the time of image capture. Huge spikes with gaps on either side mean that a large number of pixels all recorded one value where presumably there was actually tonal/color variation in the subject. In other words, the sensor failed to capture the differences that were really there.

Including a pic of the target at ISO 100 to give a baseline. As ISO was increased by a stop, flash output was lowered by a stop to achieve the same overall exposure. As mentioned above, starting at ISO 3200 I had to stop down the lens as strobe output couldn't go lower.

Below are the RawDigger histograms. They are all labeled with the ISO and aperture used. Overall exposure was kept constant throughout the series to the best of my ability. The 3 exceptions being the ISO 100 shots corresponding to ISO 3200, ISO 6400, and ISO 12800 as outlined above.

Nota bene: pay special attention to the scales on both the X-axis and Y-axis. The X-axis corresponds to EV values. For the early pics at low ISO, the leftmost edge of the scale is -8 EV. For later pics at higher ISO it moves to -12 EV since a significant number of pixels registered in this range. The Y-axis corresponds to the number of pixels registering that EV. At low ISO values the pixel spread is smoother (a good thing). At higher ISO values a large number of pixels all registered the same value with gaps around them (a very bad thing).

This pic is slightly underexposed compared to others in the series. I used the ISO 100 pic as the baseline since it was then easy to adjust strobe output in full stops to ISO changes in full stops.

Note here that I had to change the EV scale as some pixels were registering lower than -8 EV. This is potentially a bad thing, though not horrible. There is also more gapping at the low end of the EV scale. Also not a good thing.

Note here that I had to change the EV scale as a significant number of pixels in the red channel (and a few in the green channel) were registering at -12 EV. This is a bad thing. Also the gapping at the low end is becoming more pronounced.

The ISO 100 shot for the same light level as the above ISO 3200 shot. This is horrid. Major gapping. Also note that the Y-axis changed fairly dramatically. Related to the gapping, massive numbers of pixels were registering the same EV value despite the fact (as seen in the ISO 3200 shot) that there is tonal/color information present in the subject, the sensor just wasn't able to capture it at ISO 100.

Again, the X-axis scale is -12 EV because an even larger number of pixels registered here compared to ISO 3200. This is a bad thing.

The ISO 100 shot for the same light level as the above ISO 6400 shot. This is even more horrid.

Same story as above with this shot at ISO 12800 but worse. All 3 channels are showing a significant number of pixels at -12 EV and the gapping is even worse.

The ISO 100 shot for the same light level as the above ISO 12800 shot. Yikes. Notice the scale on the Y-axis. This shot is effectively 7 stops underexposed. LR only lets you increase exposure by 5 stops. This image couldn't be salvaged under any circumstances.


Two takeaways for me from this series. First: it is a really, really bad idea to significantly underexpose with a low ISO hoping the image will be the same with some fiddling in post compared to one properly exposed using a higher ISO. That is just completely inaccurate, internet videos notwithstanding. The proof is in the pudding as shown above. Expose properly and accept a higher ISO value if you need to.

Second: for my Nikon D850, there are two break points regarding ISO. ISO 1600 isn't as good as ISO 800 (or lower), but it isn't horrible. ISO 3200 is a bigger breakpoint as image degradation is much more obvious at that point compared to lower ISO values and I would be hesitant to use it. Anything higher than ISO 3200 is unacceptable to me unless I really had no other option. The other option being to just not take the pic .

Sorry for the length of this post, but hopefully people will find it helpful

 

[v3rlon]

I need to create more accounts on Mac rumors so I can give this post all the thumbs up it deserves.



Sometimes Tony Northrop goes off on some urban legend like he just found “proof” the moon landings were faked.

Even his own video shows a clear difference.

 

[casperes1996]

This was an absolutely brillant post. 50 internet points to you, good sir/madam.

I would note however that the same findings have been released by FilmakerIQ on YouTube. They also include the mathematics of figuring out the sensitivity scale/range and whatnot. But your input with data points significantly add to their material as well, so thanks!

 

[kallisti]

The other important thing to note in the above histograms is the spikes and valleys in areas in the middle and on the right of the histograms and how they flatten as the ISO is increased.

As I stated in the OP, this color target may not have been the best subject for this test. There aren't tonal gradients present, instead the subject is made up of discreet color squares. So there *should* be peaks and valleys reflecting the actual color/tonal values present in the target. *Not* gaps, but peaks and valleys.

Notice what happens to these peaks and valleys as the ISO is increased. They flatten and merge together. This means that even in areas where there aren't gaps, the sensor is failing to capture the differences in tonal values that are present in the subject. Values are getting mushed together as the sensor is failing to adequately distinguish true differences.

This starts to happen around ISO 1600 and gets progressively worse as the ISO is increased.

So shadow detail starts to be lost with increasing ISO (as evidenced by the gaps at the lower end of the ISO histograms), but the ability of the sensor to distinguish true differences in tonal values is actually being lost across the entire EV range as the ISO increases. Resulting in loss in image quality.

 

[kallisti]

I shot the series again with a light modifier that reduced strobe output, letting me shoot the entire ISO range at a constant aperture. I performed the test with the Nikon D850 and Zeiss 135mm f/2 Milvus lens, Sony A7R3 and Zeiss Batis 135mm lens, and a Leica M240 and 135mm Leica lens.

Won't share all of the results (though the findings were similar between bodies, but the different sensors didn't perform exactly the same).

***Something important to note for Sony shooters: as opposed to Nikon (and Leica) who have lossless compressed RAW, Sony compressed RAW is a lossy format. This is really inexcusable on Sony's part. Below are the RawDigger histograms for properly exposed images at ISO 100. The first image is using compressed RAW (with a file size of 41.22 MB). The second image is using uncompressed RAW (with a file size of 81.81 MB). There are very obvious differences between the histograms, favoring the use of uncompressed RAW. The increase in the file size makes this less than optimal however from a storage perspective (imagine the size of your LR catalog effectively doubling if you are a Sony shooter opting for uncompressed RAW to maximize image quality).

Sony compressed RAW

Sony uncompressed RAW.

Fewer and smaller gaps in the uncompressed RAW compared the compressed RAW image. Note the increase in size of the spikes around the areas of gapping in the compressed image. This means that the pixels in the spikes were all registering the same EV value rather than distinguishing the true tonal variations that were present in the subject. Same light and exposure settings for both images. Sony's compression algorithms for RAW just seem to suck.

 

[Painter2002]

There has been some confusion about how ISO affects images shot in RAW. A recent thread about shooting in aperture priority mode got somewhat derailed by claims that ISO is "fake" and applied during RAW conversion, the claim being that one should just shoot at base ISO and then increase exposure in post. It got me thinking so I shot a series looking to objectively assess the impact of increasing ISO on image quality.

I used a color calibration chart as the subject. Not perfect for the test since the squares are discreet, but I don't have a gradient target to use as a test.

The series was shot with a Nikon D850 and 70-200 f/2.8E FL lens @ 190mm. This filled the frame with the color target and didn't require any cropping. The camera was on a tripod and I used a remote release. Not totally vital for this test as it isn't a sharpness test.

I tested from ISO 64 (the base ISO of the D850) to ISO 12,800.

I used an Einstein studio strobe for the light that was positioned roughly 45 degrees off axis and 45 degrees above the target to provide uniform light. The room was otherwise dark, so all of the light was coming from the strobe. All images shot at 1/250th second. In general, I tried to decrease the strobe output by a stop as I increased the ISO by a stop. Early images shot at f/8, but as I increased the ISO I reached the lower limits of the strobe's output, so had to stop down the lens to keep exposure constant. ISO 3200 was the breakpoint where I couldn't decrease the strobe output further and stopped down to f/11. ISO 6400 was at f/16. ISO 12800 was at f/22.

For ISO 3200, ISO 6400, and ISO 12800 I also took a shot at ISO 100 with the same flash output to test the hypothesis of whether it is better to use a higher ISO value, better to use ISO 100, or whether it is a wash between the two. Where the ISO 3200 pic was properly exposed, the ISO 100 pic with the same light reaching the sensor was 5 stops underexposed at the time of capture. 6 stops for the ISO 6400 pic and 7 stops for the ISO 12800 pic.

I used RawDigger on the RAW files to graphically quantitate the results. For those not familiar with the program, it can display a histogram with EV on the X-axis and number of pixels registering that EV value on the Y-axis. Ideally, one wants this graph to be continuous without any gaps. Gaps indicate areas where there is no pixel information for that specific EV value. Assuming that the subject has somewhat continuous color tonal values, any gaps (specifically ones where there is zero pixel information present) imply that the sensor failed to capture useful information that was actually present in the subject. Gaps are bad. More gaps are bad. Larger gaps are bad (like, really bad). They can result in banding or other artifacts. They represent gaps in the tonal continuity of the image as captured by the sensor. Bad, bad, bad.

The RawDigger histograms let you see how all the pixels on the sensor responded to the light they received in each color channel. The overall area under the curve is a graphical representation of how all the pixels recorded light at the time of image capture. Huge spikes with gaps on either side mean that a large number of pixels all recorded one value where presumably there was actually tonal/color variation in the subject. In other words, the sensor failed to capture the differences that were really there.

Including a pic of the target at ISO 100 to give a baseline. As ISO was increased by a stop, flash output was lowered by a stop to achieve the same overall exposure. As mentioned above, starting at ISO 3200 I had to stop down the lens as strobe output couldn't go lower.

Below are the RawDigger histograms. They are all labeled with the ISO and aperture used. Overall exposure was kept constant throughout the series to the best of my ability. The 3 exceptions being the ISO 100 shots corresponding to ISO 3200, ISO 6400, and ISO 12800 as outlined above.

Nota bene: pay special attention to the scales on both the X-axis and Y-axis. The X-axis corresponds to EV values. For the early pics at low ISO, the leftmost edge of the scale is -8 EV. For later pics at higher ISO it moves to -12 EV since a significant number of pixels registered in this range. The Y-axis corresponds to the number of pixels registering that EV. At low ISO values the pixel spread is smoother (a good thing). At higher ISO values a large number of pixels all registered the same value with gaps around them (a very bad thing).

This pic is slightly underexposed compared to others in the series. I used the ISO 100 pic as the baseline since it was then easy to adjust strobe output in full stops to ISO changes in full stops.

Note here that I had to change the EV scale as some pixels were registering lower than -8 EV. This is potentially a bad thing, though not horrible. There is also more gapping at the low end of the EV scale. Also not a good thing.

Note here that I had to change the EV scale as a significant number of pixels in the red channel (and a few in the green channel) were registering at -12 EV. This is a bad thing. Also the gapping at the low end is becoming more pronounced.

The ISO 100 shot for the same light level as the above ISO 3200 shot. This is horrid. Major gapping. Also note that the Y-axis changed fairly dramatically. Related to the gapping, massive numbers of pixels were registering the same EV value despite the fact (as seen in the ISO 3200 shot) that there is tonal/color information present in the subject, the sensor just wasn't able to capture it at ISO 100.

Again, the X-axis scale is -12 EV because an even larger number of pixels registered here compared to ISO 3200. This is a bad thing.

The ISO 100 shot for the same light level as the above ISO 6400 shot. This is even more horrid.

Same story as above with this shot at ISO 12800 but worse. All 3 channels are showing a significant number of pixels at -12 EV and the gapping is even worse.

The ISO 100 shot for the same light level as the above ISO 12800 shot. Yikes. Notice the scale on the Y-axis. This shot is effectively 7 stops underexposed. LR only lets you increase exposure by 5 stops. This image couldn't be salvaged under any circumstances.


Two takeaways for me from this series. First: it is a really, really bad idea to significantly underexpose with a low ISO hoping the image will be the same with some fiddling in post compared to one properly exposed using a higher ISO. That is just completely inaccurate, internet videos notwithstanding. The proof is in the pudding as shown above. Expose properly and accept a higher ISO value if you need to.

Second: for my Nikon D850, there are two break points regarding ISO. ISO 1600 isn't as good as ISO 800 (or lower), but it isn't horrible. ISO 3200 is a bigger breakpoint as image degradation is much more obvious at that point compared to lower ISO values and I would be hesitant to use it. Anything higher than ISO 3200 is unacceptable to me unless I really had no other option. The other option being to just not take the pic .

Sorry for the length of this post, but hopefully people will find it helpful

Thank you for taking the time to post such a detailed post. I am have been trying to get more into photo editing and have often wondered what the point of meddling with aperture mode was when you can now artificially "adjust" ISO in photo-editing apps. This deserves much more upvotes!

 

[kallisti]

One more set of examples...

I screwed up when shooting the series with my Leica M240. Near the end of the series I increased the strobe output by a stop but didn't decrease the ISO. Shooting a series like this is pretty boring. Mea culpa.

The interesting thing about this mistake is that it provides an example of why people say to "expose to the right". In other words, overexpose at the time of capture without blowing out highlights.

I'd been trying to push exposure to the right of the histogram for all of my series, but it's sometimes hard to tell at the time of capture just looking at the histograms on the back of the camera. But this mistake showed that I wasn't really fully exposing to the right on my Leica tests.

Both pics taken at f/11, 1/125th sec (the fastest flash sync on the Leica), and ISO 200 (the lowest ISO on the Leica). Strobe output varied by a stop between them. This is easy enough to see on the histograms by looking at the EV value on the right of the graph in the green channel. +1 EV vs +2 EV. The second pic (with the green channel going to +2 EV) was exposed with 1 stop more light at the time of capture.

Exposing to the right mitigates the gaps in the shadow areas (i.e. the leftmost tail of the EV histogram).

Digital sensors have trouble in the shadow areas of the EV curve (again, the leftmost tail of the EV histogram). At roughly -5 EV you start seeing significant gapping which reflects loss of shadow detail.

When processing images in post, changes in exposure shift the entire RAW histogram to the right or the left. Exposing to the right at capture lets you add some underexposure in post while retaining shadow detail. Underexposing at capture means that the RAW data has gaps in the shadow areas that can't be fixed in post because there is no pixel level data there.

As a practical example, pretend that there are shadow areas in a subject that have detail in them. Reviewing your image, these areas look black. So you decide to boost the shadows in post to bring out the detail.

If the RAW data for the portion of the image you want to boost is relatively free of gaps, you will be able to see that detail as you boost the shadows because there are pixels that registered the different tonal values in that EV range.

On the other hand, if the RAW data for that portion of the image is composed of large gaps then there is no pixel data for that range of EV values. So boosting the shadows makes the black area look lighter, but it is still without detail (because there is no pixel level data present in the RAW image for those different EV values). Instead you will see artifacts (like banding).

The bigger picture (and question) for this entire thread is: does it really matter?

Which is an important (and complicated question).

How do the differences in the RAW files impact my ability to manipulate the images in post? How does this impact the final images I will see (or share or print)?

The easiest answer is that it is always better to have a higher quality file to work with (more data, more accurate data). Regardless of what you ultimately do with an image, starting off with better data keeps your options open both now and later.

Similar to the decision to shoot RAW or JPEG. RAW keeps your options open now and in the future. By the same token, a better quality RAW file is more useful than a lesser quality RAW file.

 

[mollyc]

I really appreciate all the effort you are putting into this! Thank you for sharing. For those of us without enough technical know how (or coffee) with the result of your last test and the "mistake" is the takeaway to ETTR in camera and then drop exposure in post as much as you want?

 

[kallisti]

I really appreciate all the effort you are putting into this! Thank you for sharing. For those of us without enough technical know how (or coffee) with the result of your last test and the "mistake" is the takeaway to ETTR in camera and then drop exposure in post as much as you want?

Thanks!

Yes, expose to the right when possible. The "mistake" pic isn't horribly underexposed by any stretch. It's probably close to what the camera would have metered. It's perfectly usable. It just has slightly less usable data in the shadow areas (specifically in the red channel, but also in the blue channel) compared to the second pic where I exposed to the right. From a practical perspective, the differences in these histograms would likely not be noticeable depending on what you are doing with the image in post and the ultimate output of the image.

But exposing to the right will give you the most options with your image in post. The caveat being that there is a risk that you expose too much to the right and blow out one or more channels on the right of the histogram. Meaning that you lose usable data in highlight areas that can't be recovered in post. It's similar to the gaps in the shadow areas but worse--for blown out highlights there aren't gaps in the histogram at various EV values, there is just lack of any usable sensor data above a certain EV value.

The "blinkies" you see on the camera's rear LCD when reviewing an image in-camera can kind of help you avoid this, but they are often too conservative and will "blink" even if the area in question really isn't blown out. I *think* this is because the histograms on the camera's rear screen are using an internal JPEG version of the image to produce the histograms and aren't generated from the actual RAW sensor data. So the histograms aren't accurately displaying the data in the RAW file, they are displaying histograms of what a JPEG file would look like. Yes, this is true even if you are shooting in RAW and not RAW + JPEG. Annoying.

 

[admwright]

Thanks!

The "blinkies" you see on the camera's rear LCD when reviewing an image in-camera can kind of help you avoid this, but they are often too conservative and will "blink" even if the area in question really isn't blown out. I *think* this is because the histograms on the camera's rear screen are using an internal JPEG version of the image to produce the histograms and aren't generated from the actual RAW sensor data. So the histograms aren't accurately displaying the data in the RAW file, they are displaying histograms of what a JPEG file would look like. Yes, this is true even if you are shooting in RAW and not RAW + JPEG. Annoying.

Yes, it is my understanding that the screen display is based on a jpg, if you are using raw and do not mind a "strange " looking display then you can try uni-white balance. This attempts to balance the channels to give a more accurate histogram for ettr.

 

[kallisti]

Yes, it is my understanding that the screen display is based on a jpg, if you are using raw and do not mind a "strange " looking display then you can try uni-white balance. This attempts to balance the channels to give a more accurate histogram for ettr.

Interesting. I wasn't aware of uni-white balance. Thanks for the information

I found a site that has a good explanation of why the in-camera histograms don't correspond to the data in the RAW file and how to adjust the JPEG settings in your camera to get more accurate histograms:

http://thedoctorsimages.com/uni-white-balance-histogram-nikon-d850/

 

[robgendreau]

I do a lot of outdoors stuff and there are often highlights that CAN get clipped. Like some white in a cloud, or something I am gonna edit out anywhere, or I just don't care. Somewhere (DPRevew maybe) had an interesting article about using waveforms instead of the histograms we currently see on cameras, where the wave forms corresponded to areas on the screen/VF. So you could notice that while you were exposing to the right and over the edge, turns out that's in an area that doesn't matter, and you'd still be getting good data about the exposure levels on other localized areas in the image.

 

[Dovan]

There has been some confusion about how ISO affects images shot in RAW. A recent thread about shooting in aperture priority mode got somewhat derailed by claims that ISO is "fake" and applied during RAW conversion, the claim being that one should just shoot at base ISO and then increase exposure in post. It got me thinking so I shot a series looking to objectively assess the impact of increasing ISO on image quality.

I used a color calibration chart as the subject. Not perfect for the test since the squares are discreet, but I don't have a gradient target to use as a test.

The series was shot with a Nikon D850 and 70-200 f/2.8E FL lens @ 190mm. This filled the frame with the color target and didn't require any cropping. The camera was on a tripod and I used a remote release. Not totally vital for this test as it isn't a sharpness test.

I tested from ISO 64 (the base ISO of the D850) to ISO 12,800.

I used an Einstein studio strobe for the light that was positioned roughly 45 degrees off axis and 45 degrees above the target to provide uniform light. The room was otherwise dark, so all of the light was coming from the strobe. All images shot at 1/250th second. In general, I tried to decrease the strobe output by a stop as I increased the ISO by a stop. Early images shot at f/8, but as I increased the ISO I reached the lower limits of the strobe's output, so had to stop down the lens to keep exposure constant. ISO 3200 was the breakpoint where I couldn't decrease the strobe output further and stopped down to f/11. ISO 6400 was at f/16. ISO 12800 was at f/22.

For ISO 3200, ISO 6400, and ISO 12800 I also took a shot at ISO 100 with the same flash output to test the hypothesis of whether it is better to use a higher ISO value, better to use ISO 100, or whether it is a wash between the two. Where the ISO 3200 pic was properly exposed, the ISO 100 pic with the same light reaching the sensor was 5 stops underexposed at the time of capture. 6 stops for the ISO 6400 pic and 7 stops for the ISO 12800 pic.

I used RawDigger on the RAW files to graphically quantitate the results. For those not familiar with the program, it can display a histogram with EV on the X-axis and number of pixels registering that EV value on the Y-axis. Ideally, one wants this graph to be continuous without any gaps. Gaps indicate areas where there is no pixel information for that specific EV value. Assuming that the subject has somewhat continuous color tonal values, any gaps (specifically ones where there is zero pixel information present) imply that the sensor failed to capture useful information that was actually present in the subject. Gaps are bad. More gaps are bad. Larger gaps are bad (like, really bad). They can result in banding or other artifacts. They represent gaps in the tonal continuity of the image as captured by the sensor. Bad, bad, bad.

The RawDigger histograms let you see how all the pixels on the sensor responded to the light they received in each color channel. The overall area under the curve is a graphical representation of how all the pixels recorded light at the time of image capture. Huge spikes with gaps on either side mean that a large number of pixels all recorded one value where presumably there was actually tonal/color variation in the subject. In other words, the sensor failed to capture the differences that were really there.

Including a pic of the target at ISO 100 to give a baseline. As ISO was increased by a stop, flash output was lowered by a stop to achieve the same overall exposure. As mentioned above, starting at ISO 3200 I had to stop down the lens as strobe output couldn't go lower.

Below are the RawDigger histograms. They are all labeled with the ISO and aperture used. Overall exposure was kept constant throughout the series to the best of my ability. The 3 exceptions being the ISO 100 shots corresponding to ISO 3200, ISO 6400, and ISO 12800 as outlined above.

Nota bene: pay special attention to the scales on both the X-axis and Y-axis. The X-axis corresponds to EV values. For the early pics at low ISO, the leftmost edge of the scale is -8 EV. For later pics at higher ISO it moves to -12 EV since a significant number of pixels registered in this range. The Y-axis corresponds to the number of pixels registering that EV. At low ISO values the pixel spread is smoother (a good thing). At higher ISO values a large number of pixels all registered the same value with gaps around them (a very bad thing).

This pic is slightly underexposed compared to others in the series. I used the ISO 100 pic as the baseline since it was then easy to adjust strobe output in full stops to ISO changes in full stops.

Note here that I had to change the EV scale as some pixels were registering lower than -8 EV. This is potentially a bad thing, though not horrible. There is also more gapping at the low end of the EV scale. Also not a good thing.

Note here that I had to change the EV scale as a significant number of pixels in the red channel (and a few in the green channel) were registering at -12 EV. This is a bad thing. Also the gapping at the low end is becoming more pronounced.

The ISO 100 shot for the same light level as the above ISO 3200 shot. This is horrid. Major gapping. Also note that the Y-axis changed fairly dramatically. Related to the gapping, massive numbers of pixels were registering the same EV value despite the fact (as seen in the ISO 3200 shot) that there is tonal/color information present in the subject, the sensor just wasn't able to capture it at ISO 100.

Again, the X-axis scale is -12 EV because an even larger number of pixels registered here compared to ISO 3200. This is a bad thing.

The ISO 100 shot for the same light level as the above ISO 6400 shot. This is even more horrid.

Same story as above with this shot at ISO 12800 but worse. All 3 channels are showing a significant number of pixels at -12 EV and the gapping is even worse.

The ISO 100 shot for the same light level as the above ISO 12800 shot. Yikes. Notice the scale on the Y-axis. This shot is effectively 7 stops underexposed. LR only lets you increase exposure by 5 stops. This image couldn't be salvaged under any circumstances.


Two takeaways for me from this series. First: it is a really, really bad idea to significantly underexpose with a low ISO hoping the image will be the same with some fiddling in post compared to one properly exposed using a higher ISO. That is just completely inaccurate, internet videos notwithstanding. The proof is in the pudding as shown above. Expose properly and accept a higher ISO value if you need to.

Second: for my Nikon D850, there are two break points regarding ISO. ISO 1600 isn't as good as ISO 800 (or lower), but it isn't horrible. ISO 3200 is a bigger breakpoint as image degradation is much more obvious at that point compared to lower ISO values and I would be hesitant to use it. Anything higher than ISO 3200 is unacceptable to me unless I really had no other option. The other option being to just not take the pic .

Sorry for the length of this post, but hopefully people will find it helpful

Thanks for sharing. Useful one.