With two RED ONEs back in from their M-X upgrades, and with the results of my previous test in mind, I ran a more rigorous examination comparing the Mysterium to the Mysterium-X.
In my previous side-by-side there seemed to be about a half-stop difference in the sensitivity of the M (Mysterium) and M-X (Mysterium-X) sensors; the M-X clips shot at a given exposure were all a bit lighter than their M-captured counterparts. This sensitivity difference made it difficult to compare the two sensors, as my two exposure ramps were half a stop out of step with each other—most noticeable in the apparent different clipping levels of the two chips, with the M-X clipping sooner than the M.
I also had a non-uniform magenta stain on the high-gain shot from the M-X, which I put down to excessive eagerness on my part; I had pulled our new M-X camera out of the box and thrown it straight into testing, without running a black-shading calibration pass first.
This time, I tested the cameras on their own terms: instead of dialing in an ISO rating in the cameras and on my light meter, and then setting the cameras exactly as my meter told me to, I trusted the cameras’ own internal meters: their false-color display mode colors anything that’s at 18% reflectivity green, so I could dial my exposure up and down until my 18% gray card turned green. I also metered the scene separately and compared the meter reading with what the cameras reported.
Procedure
I set up my test targets, illuminated with two Kinoflo Diva-lite 400s with 5600K daylight tubes. I used a single tripod, putting the camera under test onto it (however the cameras were using different baseplates, so there was a slight height difference between them, perhaps a centimeter. They also balanced differently front-to-back, and as I balanced them on the fluid head, one sensor wound up maybe an inch closer to the targets than the others, resulting in a minor magnification difference between the images).
I used three cameras: #2323 (M-X), #2324 (M-X), and #2325 (original Mysterium; upgrade delayed specifically so we could shoot these tests). All cameras were running build 30.5.0 firmware. All cameras were set to daylight balance: 5600K, Tint 0. The cameras were warmed up for at least half an hour, and then two successive black-shading calibrations were run with the bodies capped.
All images were shot at 4K HD, and processed in REDCINE-X build 252.
I put our 32mm T1.9 Ultra Prime on the camera being tested, and used it for all the exposure ramps. I first tested that its T-stops were properly marked by varying shutter speed and T-stop in inverse proportion and verifying that exposure remained constant; it did from T2 to T16, with a slight deviation at T22 (as an aside, I also found that while most of our Ultra Primes matched one another—the 16mm, 24mm, 32mm, 50mm, and 65mm—our 85mm is about 1/6 of a stop slower than the others, while the 135mm is 1/3 of a stop slower, and our RED 18-85mm zoom is 1/3-1/2 a stop faster than the Ultra Primes for the same indicated T-stop!)
The RED ONE “M” has ISO settings from 100 to 2000 in 1/3-stop increments; the M-X cameras run from ISO 100 to ISO 6400. I shot the chart with both RED ONE #2325 (Mysterium) and #2323 (Mysterium-X) camera at every available ISO setting, using the false-color exposure meter to set the neutral grays in my scene to the 18% reflectance level, which come in around 45% on the waveform monitor. I also opened the lens one and two stops for the ISO 100 test to really overexpose things, and I faked the settings for ISOs from 2500 to 6400 on the Mysterium RED ONE, since that camera doesn’t let me dial in those sensitivities. Since I simply closed the lens down a third of a stop each time by looking at the barrel markings alone, there may be more variance in those exposures than would be the case if I could have seen the exposure level in the EVF.
I shot a few setups using RED ONE #2324 alongside #2323, which went into RED for upgrading in the same box as #2323 but returned a week later. That way I could compare the two new cameras against one another and see if they tended to track the same way or if they were wildly different.
To double-check the high gain color non-uniformity I had seen on #2323, I put the 135mm lens on each camera, focused it to infinity, and taped a sheet of white paper over the front of it as a diffuser, so I could capture a solid field of neutral color. I then twiddled aperture and shutter speed to yield about 20% exposure at ISO 6400, and grabbed a clip of it.
I repeated that test the following morning with the two M-X cameras, with the 135mm lens aimed at a gray card but focused at infinity—again, I was aiming for a flat, featureless field, and shooting an out-of-focus gray card flatly lit with the two Kinos seemed like a good way. I grabbed a clip as soon as the camera booted, before the camera had warmed up; and again an hour later, after its temperature had stabilized.
All clips from a given camera were color-balanced to the setting used for that camera’s ISO 400 clip, and rendered to 1920×1080 ProRes422 HQ using the RED ROCKET hardware accelerator.
Then, since RED ROCKET doesn’t allow use of the DRX control to remove color casts from clipped highlights, I reprocessed the clips using REDCINE-X’s software codec with DRX 1.00 for all clips.
I took the clips into FCP, gamma-corrected them to fix the QuickTime RGB-YUV transcoding peculiarity, and output h.264 versions for viewing.
The Results
I’ve put together a couple of videos: one on YouTube, with a side-by-side comparison of the M and M-X sensors at the same ISO setting, and a QuickTime here on PVC (~43 MB, so don’t try this on your iPad 3G with AT&T’s limited data allowances!) that focuses on one camera at a time.
The PVC version is a 900×506 pixel-for-pixel extraction from the larger frame, with a small inset of the entire image, so that you can see pixel-level detail and noise as well as the whole picture. The clip’s hefty bitrate was chosen to preserve as much of the source image’s noise as practical. It’s saveable; you can grab it to your local disk, throw it into your NLE, and do your own side-by-sides or A/Bs of various ISO settings.
Both videos have the results of both RED ROCKET conversion with no DRX, and software conversion with DRX. The software conversions all have sharpness and denoise settings at zero; they look marginally softer than the hardware-assisted conversions as a result.
Observations and Conclusions
Both cameras show 18% gray at the same level when in RAW and in REDcolor/REDspace; there’s little change in monitor scene brightness when toggling between viewing modes at this ISO setting.
The Mysterium camera consistently reads a stop less sensitive than an external light meter; if the camera says a scene should be shot at T4, my meters will read the same scene at T5.6.
The Mysterium-X cameras were half a stop faster than the Mysterium camera for the same ISO rating, and half a stop slower than an external meter. If the Mysterium camera saw a scene at T4, and the meter saw T5.6, the Mysterium-X camera wanted to use T4.8.
M and M-X cameras have rather different native white points. Under Kinoflo daylight illumination, the M sensor’s three channels are well-matched, with blue clipping first, green right behind it, and red soon thereafter; yellowish blowout results. On the M-X, green hits clipping first with red and blue well afterwards, resulting in brighter magenta blowout.
M (left) and M-X (right) at ISO 25 (effective), no grading, no DRX.
M (left) and M-X (right) at ISO 25: note that green is at same level in both.
In the histograms shown, both cameras are two stops overexposed at ISO 100. All color channels in both cameras have hit their ceilings, with green at the same level in both, but with higher levels in R and B in the M-X sensor. If you look at the overexposed bits in the videos, you’ll see that even though the clipping level is generally the same for the two sensors(remember, green contributes about 70% of the brightness signal), the M-X sensor allows more useful and more accurate color detail in bright highlights since red and blue have a bit more reach.
Even at “normal” exposures, and with the cameras set to the same fixed color balance, the M-X renders a warmer, more magenta image:
M (left) and M-X (right) at ISO 320, not color-corrected.
To makes these images match, I rebalanced the M sensor’s pix to 6810K, 2.34 tint, while the M-X clips required 6207K, 5.95 tint; those are the corrections applied in the video clips.
Roughly speaking, the Kelvin adjustment moves pix between yellow (low K) and blue (high K), while tint biases images towards magenta (low tint) and green (high tint).
Don’t confuse tint with the hue control of the same name on NTSC analog TVs and some color-correction controls. RED’s tint control moves the entire image, including its white balance, towards green or magenta; the constellation on a vectorscope will move side to side along the green/magenta axis. An NTSC tint control spins the vectorscope constellation around the current white point, changing all the hues, but it doesn’t affect the white balance at all.
At high gains, the M-X sensor has about two stops more usable range; I see roughly the same amount of image noise on the M-X at ISO 6400 as I see on the M at 1600 (mind you, that’s very subjective; the noise has different characteristics on the two cameras). Put another way, the M-X has about two stops more usable dynamic range between its clipping-limited highlight detail and its noise-limited shadow detail.
(Where exactly that noise-limited point is, and thus what the dynamic range number is specifically, is a religious issue I won’t go into, thanks all the same: the noise levels seen in a 512×288 web video, processed through Arri Relativity noise reduction, will be rather different than the levels seen in a 4K output straight from REDCINE-X with sharpness cranked up and no noise reduction applied.)
Indeed, those with a high noise tolerance may find that the M-X sensor isn’t noise-limited, it’s color-uniformity-limited.
M (left) and M-X (right) at ISO 6400, not color-corrected.
Even after two black shading calibrations done with the camera fully warmed up, the M-X sensors show noticeable shading differences across the frame at high gains. It’s quite visible in M-X pix at ISOs above 3200, whereas M images aren’t visibly affected:
M (left) and M-X (right) at ISO 6400, flat-field shading.
The non-uniformity appears to be heat-related, as clips recorded from a just-fired-up camera are much more uniform, though the pattern is still present. Both our M-X RED ONEs show this increase in non-uniformity as they warm up, and each has its own characteristic pattern:
RED ONE #2323 M-X ISO 6400 shading at power-on (left), after 1 hour (right).
RED ONE #2324 M-X ISO 6400 shading at power-on (left), after 1 hour (right).
Don’t panic: this isn’t unique to RED M-X sensors. Many digital cameras show shading non-uniformity at high gains; I’ve seen it in quite a few video cameras at all price points, and in a couple of pricey still cameras, too. Practically speaking, it’s unlikely to be an issue at any sensitivity setting up through ISO 3200—though if you’re matching cameras for stereo 3D or other critical operations, you may want to stay below ISO 1000 or so. As always, don’t take my word for it; test for yourself, preferably before an important shoot!
Other observations:
• DRX does a good job of correcting blown-out highlight errors, but it also increases color saturation slightly throughout the tonal scale. DRX also isn’t available when using the RED ROCKET accelerator. It’s a good idea to avoid blown-out highlights in general, so you don’t need to use DRX—but it’s nice to have it available, just in case.
• RED’s recommended operating point of ISO 800 does seem like a good suggestion; it gives you about 1.3 stops more highlight headroom (so bright, strongly colored things like neon signs, car brake lights, and such hold their color more faithfully) as well as a stop more shadow detail.
But you can also shoot it at ISO 320, just as before, and have a much cleaner tonal scale overall with the same headroom as before—very nice for greenscreen work, for example. And when the lights are low, the camera is quite usable at ISO 1600 or even ISO 3200.
• When flipping between M and M-X clips, Tim Blackmore and I both noticed that the M-X sensor captured more accurate, higher purity colors. It’s a subtle thing, best appreciated by loading M and M-X clips in two windows, stacking them one atop the other, then flipping between them (or loading them on two tracks in a NLE and then toggling the visibility of the upper layer). Between the new “color science” in build 30 firmware and the M-X sensor, I expect that RED color will be a lot truer to nature than has been the case previously; I’m eager to shoot some people tests and see how the camera renders real-world subjects.
See also:
LightIRON Digital’s M-X tests. These compare Mysterium using the older REDspace look with Mysterium-X and REDcolor, but they’re interesting nonetheless.
16 CFR Part 255 Disclosure
No material connection exists between me and RED, other than as a customer. My employer, Meets The Eye LLC, purchased three RED ONEs two years ago on my recommendation, and we have since purchased two M-X upgrades (soon to be three), three RED ROCKET hardware decoders, two 18-85mm zoom lenses, and a selection of accessories. I do not personally own any RED products nor do I have any financial interest in the company.