We sent some files to our friend Jim Kasson for analysis, and he confirms that the Sony a7R III is definitely still a Star Eater, despite several claims to the contrary that have been published online over the past week. That said, while his analyses don't suggest so, our images shot in the field suggest at least some improvement (more on this in the 'real world' section below).

Looking at Kasson's graphs, one can clearly see the noise reduction kick in at 4s and above in Kasson's Fourier transform energy plots. Indeed, in our own shots of the stars with the a7R III and latest a7R II bodies, while the a7R III looks slightly better, all bodies only show stars that are larger than one pixel with a few neighboring pixels. This suggests that smaller (single pixel) stars are indeed 'eaten' or dimmed due to a spatial filtering algorithm.

At a 3.2-second exposure, the 'spacial filtering' (Star Eater) is mild to non-existent: noise (in dB) is similar at all frequencies, so your stars shouldn't be affected. Credit: Jim Kasson

But as soon as you hit 4-seconds, spacial filtering kicks in, which you see here as a drop in noise (in dB) at the highest frequencies (the right side of the graph, where 0.5 f / fs is the Nyquist limit). That means the smallest details have noise reduction applied to them, causing similar Star Eater problems that were seen in the a7R II. Credit: Jim Kasson

Real-world examples

We did our own night sky shootout of an a7R III vs. an a7R II v3.30 vs. an a7R II v3.00. Below, you can compare the a7R III vs. an a7R II with v3.00 firmware (which Jim confirmed to have similar noise reduction in his analyses):

Many parts of the image, especially where stars are densely concentrated, show little to no difference between the Mark III and Mark II (v3.00). But if you look at sparser lower contrast regions, the Mark III does show an improvement: there are stars that aren't there in the Mark II shot, and there's generally more 'pop' to the stars. This is encouraging, but it's also difficult to rule out how the spatial filtering algorithm interacts with shots that might be ever-so-slightly focused differently.

Should you care?

While it's possible Sony may have tweaked the spatial filtering algorithm, the difference is often subtle, and we're still not seeing any pixel-sized stars in any of our shots (indeed, Jim Kasson visually shows us that single pixel-size detail is removed at 4s and above). But on the other hand, we still got some very nice starry skies out shooting in Sedona. Have a look at the full a7R III shot below and compare it to a similar one from the a7R II v3.00.

A real-world nightscape shot with the Sony a7R III and a Sigma 14mm F1.8 lens. Sure it may be munching small, single-pixel stars, but many stars do remain, while hot pixels are suppressed. Whether or not this matters to you is something only you can decide. Click to view at 100%.

15s, F1.8, ISO 3200, LENR off.

While we have yet to do a direct comparison of a star field vs. a camera that doesn't employ such an algorithm, we can say this with confidence: while a lot of stars still survive 'Star Eater', the a7R III continues the trend of noise reduction that dims or erases small stars at exposure longer than 3.2s. There's some sense in this: Sony is ostensibly using this algorithm to avoid hot pixels that might otherwise riddle long exposures. But some nightscape and astro- photographers wish to be given a choice as to whether or not this form of noise reduction is applied. Alternatively, the noise reduction could (reversibly) be applied in post.

And so, this seems a missed opportunity for Sony, which could have offered a choice to its users. Instead, the forced noise reduction may give pause to dedicated astrophotographers who can opt for cameras without this issue (see a similar energy plot for the Nikon D850 at 30s). Other photographers happy with the number of stars still in their shots likely won't care.

Other photographers happy with the number of stars still in their shots likely won't care

We'll be investigating further to see if the improvements we noted in portions of the Mark III image over the v3.00 Mark II one are real, so stay tuned.