Image Quality

Our latest test scene simulates both daylight and low light shooting. Pressing the 'lighting' buttons at the top of the widget switches between the two. The daylight scene is manually white balanced to give neutral grays, but the camera is left in its Auto setting for the low-light tests. Raw files are manually corrected. We offer three different viewing sizes: 'Full', 'Print', and 'Comp', with the latter two offering 'normalized' comparisons by using matched viewing sizes. The 'Comp' option chooses the largest-available resolution common to the cameras being compared.

Raw performance

Although at first glance the a9 appears slightly softer than the Nikon D5, part of this may be due to its slightly higher resolution and the D5's known-to-be-weak AA filter (not to mention: the different lenses). Detail levels are quite similar to the Canon EOS-1D X II, with similar levels of color aliasing. The smallest lines of text in particular look very similar among the three. The currency shows similar output between the a9 and D5, and the playing cards show similar levels of color aliasing - suggesting parity in detail capture. When light levels drop, the Sony shows very similar noise performance to the class-leading D5 and pulls very slightly ahead of the 1D X II. At the very highest ISO values, the Nikon pulls ahead of the Sony, but these settings are all generally so noisy, it's a difference you're only likely to appreciate side-by-side.

Compared with the Sony a7R II, you can see that the a7R II is slightly sharper as well as being much more detailed than the a9. The a9's magnified live view doesn't help achieve perfect focus (its lower resolution output makes it harder to achieve focus), and no number of attempts yielded results as good as the a7R II at the pixel level. Part of this may be due to the different lenses, but the results suggest the a9 has an anti-aliasing filter (which the a7R II lacks), which makes sense for the pro market it's targeted at. At higher ISOs, the 42MP camera shows comparable noise but still more detail when viewed at equivalent size - impressive given its high resolution.

JPEG output

When it comes to JPEGs, we can see that the a9 still has blue-shifted greens and yellows not as deep or saturated as a Nikon D5. Reds and yellows are also slightly less rich than the 1D X II, which is one of our current favorites for JPEG color rendition. Colors are generally more more muted in comparison, with the interesting exception of the bright blue patch. This is also apparent at high ISO values, where the Sony also shows additional color bleed along edges compared to the D5. There does appear to be some color evolution though: yellows look a bit better than the a7R II, particularly at high ISO, and in real-world shooting we've found more magenta and less greenish tinge to images - a good thing in our opinion.

At base ISO, clever sharpening means the text sharpens up nicely compared to the Raw, without any strange artifacts, and actually appears to show more detail than the D5 in most areas. The a9 produces a very sharp image without the downsides of oversharpening, with no visible noise or 'haloing' around edges (something the D5's large-radius sharpening produces). Sony handily beats Canon in retaining fine detail (despite starting with similar detail in the Raw file), while avoiding halos. It's safe to say Sony's sharpening is the most sophisticated we've seen.

It's safe to say Sony's sharpening is the most sophisticated we've seen.

As the ISO values climb, both the a9 and D5 leave behind some luminance grain while all chroma noise is removed (this is also similar to the a7R II's behavior, though the a9 looks cleaner). While the Canon 1D X II appears to be performing similarly in this region of few features, its noise reduction is not as intelligent or context-sensitive: obliterating low contrast detail. At ISO 12800, the a9's noise reduction removes some small details in the dead leaves pattern the D5 leaves behind, but again does a much better job than the Canon. Overall, the Sony has one of the best noise reduction engines we've seen, essentially trading punches in detail retention with the D5, which we also consider to be excellent at high ISO. Which you prefer will be a matter of taste: the a9 leaves a bit more detail behind, while the D5 yields a slightly cleaner image. Canon unfortunately leaves noise behind while sacrificing low contrast detail - a sort of worst-of-all-worlds approach.

Note on banding

Early on, Diglloyd reported banding in blue skies in black and white conversions that relied heavily on the red channel. We've looked into this and while the banding does appear in the raw red channel when it's heavily underexposed (blue skies), its pattern follows that of the masked phase-detection AF rows of the sensor - where the blue pixels are masked and then binned for readout (there are 336 total masked rows).*

We took this image (you can download the Raw yourself if you click-through) from our Sony 12-24G gallery and converted it using ACR with and without noise reduction. The red channel of the latter clearly shows a banding pattern to it and, interestingly, if you hover back and forth over the 100% 'No AF overlay' and 'AF overlay' crops below, you'll see the pattern matches that of the phase detection rows (indicated as black lines across the image). We're not sure how masked blue pixels end up yielding these bands in the red channel (they're in the red channel of the Raw file before demosaicing), but the artifacts are certainly present. Click on the large image for full-size to see the artifacts clearly.

ACR - Zero NR ACR - defaults No AF overlay AF overlay

While this may not show up in typical shooting, it can affect certain black and white conversions of deep blue skies. Phase detection rows using masked pixels can also cause stripes across your images when you're shooting wide open directly into the sun or, more generally speaking, when you have a lot of flare in your shot like this one. This is not an issue unique to the a9; we've witnessed it in almost mirrorless camera with masked PDAF pixels, such as in this Fujifilm X-T2 shot (where it shows up as weave pattern due to the X-Trans pattern) or this RX100 V shot.**


* Our friends Bill Claff, Jim Kasson, and forum member Horshack are to be credited with much of these findings, and for the detection of phase-detection rows.

** Bill Claff detected all phase-detection rows on the RX100 V sensor and generated an overlay that lined up perfectly with the stripes in this shot, implicating the masked pixels.