There’s been a lot internet chatter lately about the so-called “star-eater” issue with Sony cameras. If you haven’t heard of this, last August Sony issued a firmware update for the Mark II versions of all its full-frame E-mount cameras (a7 II, a7s II, and a7r II), and many people have reported that since then these Sony models have been making smaller stars disappear in nighttime photographs. It seems that the new firmware included some kind of noise-reduction algorithm (even with Raw files) that blurred or eliminated those smaller stars. And, unfortunately, there’s no way to revert these cameras to the previous firmware version. In June of this year Sony released new firmware updates for these models, but some initial reports indicated that this new firmware didn’t fix the “star-eater” issue.
I heard about all this before the June firmware update, and naturally I was concerned, since my main camera is the Sony a7r II, and I make many nighttime images with it. I was also puzzled, since I hadn’t seen any noticeable difference in my star photos since the firmware update last August. I went through my star photographs made with the “star-eater” firmware (Version 3.3 on my a7r II), and found plenty of small stars in all of them. But it was impossible to make any kind of side-by-side comparison, since there was no way to revert to the previous firmware version.
The inability to revert to a previous firmware version also made me wonder about how people had made their tests that seemed to confirm this star-eater issue. Did they have two copies of the same model camera, one with the older firmware, and one with the “star-eater” firmware issued last August?
I don’t know the answer to this, since most reviewers haven’t spelled out their exact testing methods. But the noise-reduction algorithm apparently kicks in with exposures longer than 3.2 seconds, and most of the tests I saw seemed to compare exposures above and below that threshold – that is, 3 seconds vs. 4 seconds. While those comparisons seemed to clearly show a problem, they weren’t, in my view, real-world situations. I don’t make star exposures at 4 seconds. I typically make star exposures at 15 to 30 seconds, and I think most people who photograph nighttime landscapes with stars also use exposures in that range.
So when the new firmware was released in June, I decided to do some of my own tests with my a7R II, comparing different exposure times, and comparing this new firmware (v4.0 on the a7R II) with the previous version (v3.3 on the a7R II) that caused so much controversy. This was pretty easy to do, since I’m lucky to live in a place where I can step out into my driveway on a moonless night and see dark skies with plenty of stars.
The test images shown below were all made with the same camera (a7R II) on a single night. I made exposures of 3 seconds, 4 seconds, and 20 seconds with the Version 3.3 firmware (the “star-eater” firmware issued last August), then updated the firmware to the latest version (Version 4.0, issued this June), and repeated the same exposure times (3, 4, and 20 seconds). I chose those shutter speeds because the “star-eater” noise-reduction algorithm seems to kick in with exposures longer than 3.2 seconds, so the easiest way to see the effect was to compare 3- and 4-second exposures. But since I (and most photographers) typically use longer shutter speeds at night (like 15 to 30 seconds), I included 20-second exposures in my tests to be able to compare the 3- and 4-second images with more typical, real-world photographs.
A few more things: First, some reports had mentioned that on older Sony models (a7, a7r, a7s) the “star-eater” effect only kicked in with Bulb Mode. Just in case, I tested my a7r II both ways, with a 20-second exposure set through the shutter-speed dial, and timing a 20-second exposure with the camera set to Bulb. I couldn’t see any difference with either firmware version.
Second, Sony released some rather confusing wording with the latest firmware updates. With the a7r II, they said, “Improves image quality when Long Exposure NR setting is OFF.” With the a7s II, they said, “Improves image quality when Long Exposure NR setting is ON.” So, just in case long-exposure noise reduction mattered, I tried 20-second exposures both ways, and again the photos were identical, both with the old (3.3) and new (4.0) firmware versions.
Third, I used Uncompressed Raw for all these tests, and turned off e-Front Curtain Shutter, to make sure the camera was producing the highest image quality possible. (I also compared photos made with Compressed and Uncompressed Raw, and with and without e-Front Curtain Shutter, and could see no differences.)
And lastly, I used Adobe’s default sharpening and noise-reduction settings in Lightroom for the screen shots shown below. And processing was minimal: I started with my typical default settings (Exposure at -1.00, Contrast at -33), and then moved the white point a little to brighten the stars. That’s it.
Test Results with the Version 3.3 Firmware
So with that out of the way, let’s delve into the results. Using the Version 3.3 firmware issued last August for my a7R II (the “star-eater” firmware), I made exposures of 3 seconds at 20,000 ISO, and 4 seconds at 16,000 ISO (both with an aperture of f/1.8 on my Rokinon 20mm lens). In this test it was pretty easy to see the star-eater effect. Here is a 100% (1:1) screen shot, with the 3-second exposure on the left, and the 4-second exposure on the right (click on the image to view at 100%):
The initial impression is that there are fewer stars on the right, in the 4-second exposure. But if you look closely (again, click on the image to really view it at 100%), you’ll see that everything has been blurred slightly in that right-hand photo, so those tiny stars are less visible, but most of them are actually still there. This is significant, because with a longer exposure those tiny stars will become short streaks, and become more visible (more on that shortly). However, this comparison does show clear evidence of the star-eater issue. Though it doesn’t really make those tiny stars disappear completely, it does make them less visible, which amounts to the same thing.
Something else you’ll notice is that the right-hand photo is less noisy then the left-hand photo – a much bigger difference than you’d expect to see from just changing the ISO by a third of a stop. So there clearly seems to be some noise-reduction algorithm kicking in with exposures of 4 seconds or longer that softens the image, reducing the visibility of stars. (Though there is an upside to this: less noise.)
Again, though, I don’t typically shoot stars at 4 seconds. So here’s a comparison of the same 3-second exposure (at 20,000 ISO and f/1.8) to a 20-second exposure (at 5,000 ISO and f/2.5):
There’s considerably less noise in the right-hand photo, as you would expect when reducing the ISO from 20,000 to 5,000. But there’s also a certain softening effect, which I attribute to the camera’s noise-reduction algorithm. And, of course, the longer exposure has turned the stars into short streaks. That’s normal for star photographs like this; the stars may look like streaks at 100%, but in any normal print size they look like points. (Most people use the 600 or 500 rule for this, but I use the 400 rule, which is a bit more strict. You divide the focal length – in this case 20mm – into 400, and the result is the maximum number of seconds you can use before the stars become streaks in a normal-size print – like, say 20×30. For a 20mm lens, 400 divided by 20 is 20, so the maximum length of the exposure is 20 seconds.)
More importantly, if you look closely (again, click on the image to see it at 100%), it’s pretty hard to find any missing stars. Since those tiny stars have turned into short streaks, they’re larger, and therefore more visible in the 20-second exposure than in the 4-second exposure. In most cases, the small stars are actually more visible in the 20-second exposure than in the 3-second exposure – despite the “star-eater” noise reduction taking place.
And here’s another very important point about all this: With a 100% view of files from a 42-megapixel camera, we’re looking at an extreme enlargement. In any normal-size print those tiniest stars wouldn’t be visible anyway, with or without the “star-eater” noise reduction algorithm.
There are a couple of factors that might have affected these tests. First, it’s possible that stopping down the aperture from f/1.8 to f/2.5 might have increased sharpness, and the visibility of tiny stars. Perhaps, but these screen shots were all taken from near the center of the image, where this lens is quite sharp even at f/1.8, so I don’t think that’s a factor.
Also, reducing the ISO from 20,000 to 5,000 decreased noise, and noise can obscure fine details (like tiny stars). So everything else being equal, a photo made at a lower ISO would be more likely to show small stars, and that might explain why the 20-second exposure reveals as many (or more) tiny stars than the 3-second exposure, despite the “star-eater” algorithm. Unfortunately I don’t know any way to compare photos of stars made at 3 seconds vs. 20 seconds without changing the aperture and ISO. The only way to make a true apples-to-apples comparison would be to revert the camera back to an older firmware version (prior to v3.3) and use the identical shutter speed, aperture, and ISO for comparison photos made with different versions of the firmware. Since I can’t do that, there’s no way to know for sure how much of a difference the change in ISO contributes to these comparisons.
We can see, however, that a 20-second exposure at ISO 5,000 shows lots of tiny, faint stars, even with the “star-eater” (v3.3) firmware. And again, when comparing the 3-second exposure to the 4-second exposure, with identical apertures and a minimal change in ISO, we can see that the 4-second exposure shows nearly all the tiny stars visible in the 3-second exposure, so we could expect to see those tiny stars become even more visible as they start to streak during a longer, 20-second exposure.
Test Results with the Version 4.0 Firmware
After upgrading the a7R II’s firmware to the version released in June (4.0), I did the same tests. Here’s a comparison of a 3-second exposure (at f/1.8, 20,000 ISO), and a 4-second exposure (at f/1.8, 16,000 ISO):
The right-hand photo looks a little better here – not quite as soft as the 4-second exposure with the Version 3.3 firmware, and the tiniest stars look a little sharper. You can still find a few stars in the left-hand photo that aren’t visible in the right-hand photo, but not many. So the latest firmware does seem to be a bit better.
Next, here’s a comparison between a 3-second exposure (f/1.8, 20,000 ISO) with a 20-second exposure (f/2.5, 5000 ISO):
Here, again, I think the results on the right are a little better than they were with a 20-second exposure using the previous firmware (v3.3). The stars aren’t quite as soft, and I can’t find any stars that are visible on the left that aren’t visible on the right. Again, the slight streaking makes the tiniest stars more visible, and, in fact, there are many stars that are almost invisible on the left that are clearly visible on the right.
(Note that the stars had moved during the time it took to upgrade the firmware. I tried to adjust my screen capture to show the same group of stars with the firmware Version 4.0 tests as with the firmware Version 3.3 tests, but the stars rotated slightly.)
As I said above, I didn’t see any noticeable difference in my star photos after upgrading to the “star-eater” firmware last August. I found plenty of small stars in my nighttime images. These test images seem to show why: even with that “star-eater” firmware (v3.3 on my a7R II), very few stars were actually lost in a normal, 20-second exposure. With the new firmware (v4.0 on my a7R II), this is even less of a problem. So with this camera, the a7r II, I have no concerns about the “star-eater” issue.
However, most of the complaints I’ve heard about the “star-eater” issue have come from owners of the a7s II, so it may be a more significant problem with that camera. The a7S II is a very popular camera for night photography because of its low-light capabilities. It only has 12 megapixels, which is part of what makes its low-light performance so strong, because each photo site on the sensor can be much bigger, gathering more light, and reducing noise. But because each pixel on this camera is larger, a tiny star may only cover one pixel, rather than several, and could be more easily erased with the softening created by Sony’s noise-reduction algorithm. Plus, this camera’s lower resolution means that each pixel has to be enlarged much more for a given size of print (compared with the 42-megapixel a7r II), so any loss of stars (when viewed at 100%) would be more apparent in a print. I don’t own an a7S II, but I’d like to see some real-world tests with this camera that include exposures of stars between 15 and 30 seconds, and compare the old and new firmware. (For the a7S II, the “star-eater” firmware issued last August is Version 2.2, while the latest version as of this writing, issued this past June, is Version 3.0.)
I do think Sony went too far in trying to reduce noise in the firmware issued last August. The latest firmware issued in June is better, at least with the a7R II, but there’s still room for improvement. And maybe they could give us choices about whether to use this noise reduction or not, or better yet, adjust the strength of it. But again, having said all that, I have no real concerns about photographing stars with the a7r II. It’s a great camera for star photography.
— Michael Frye
P.S. The star-eater issue may be a more serious problem with true astrophotography – done with telescopes, and a star-tracking device (equatorial mount). Tracking stars means they will remain small points, rather than becoming streaks, even with long exposures (that’s the whole point of the tracking equipment). So it’s possible the star-easter issue would be more of a problem when using a star-tracking device. I don’t have star-tracking equipment, but I’d like to see some tests from someone who does.
Michael Frye is a professional photographer specializing in landscapes and nature. He is the author or principal photographer of The Photographer’s Guide to Yosemite, Yosemite Meditations, Yosemite Meditations for Women, Yosemite Meditations for Adventurers, and Digital Landscape Photography: In the Footsteps of Ansel Adams and the Great Masters. He has also written three eBooks: Light & Land: Landscapes in the Digital Darkroom, Exposure for Outdoor Photography, and Landscapes in Lightroom: The Essential Step-by-Step Guide. Michael has written numerous magazine articles on the art and technique of photography, and his images have been published in over thirty countries around the world. Michael has lived either in or near Yosemite National Park since 1983, currently residing just outside the park in Mariposa, California.