RED Day report by Adam J. Wilt
RED Leader Jim Jannard at (Ren-Mar) RED Studios on Saturday.
Stage 4 at (Ren-Mar) RED Studios following the Q&A&S session.
"Work proceeds apace on the new M-X (Mysterium-X) sensor. Ted Schilowitz told me that the new sensor
had higher sensitivity; rating the M-X any slower than ISO 800 resulted in excessive clipping.
Upon reading this, Jim Jannard wrote me with a correction:
This sensor is EXACTLY like the original Mysterium sensor but with a much lower noise floor.
ISO 320 looks spectacular with the new M-X sensor and has exactly the same highlight protection as the original.
The pixel size is the same for the two sensors… just much less noise.
DR is defined as signal/noise so it isn’t a big surprise that there is more DR in the new sensor."
Jim showed a clip shot the night before at ISO 1250, lit by firelight, and one shot at ISO 2000, lit by a match.
There was a little bit of noise in the ISO 2000 clip, but it wasn’t objectionable, and the ISO 1250 looked pretty clean.
In the demo area, two RED ONEs were set up with M-X sensors. It’s hard to say from looking at the LCD and 720p outputs,
but to my eye the M-X at ISO 800 looked about at least as clean as the original M (Mysterium) sensor at ISO 320.
When I put the LCD into false-color mode, the shadows at ISO 800 even looked a bit cleaner than M shadows at 320;
the dark blue and purple false colors showed less flicker and
noise on the M-X than I’m used to seeing on our RED ONEs with the M at ISO 320.
There’s a new “color science” that goes along with M-X; it includes a new REDcolor matrix and REDgamma tonal curve,
with a new FLUT (floating LUT) setting that lets you vary the midgrays at an ISO setting without affecting clipping,
so that you can really calibrate the camera’s output to agree with your lightmeter for a given ISO setting.
Graeme Nattress described it to me as a LUT that looks for clipping in any of the channels and applies a rolloff as needed
(much like the shoulder of a film’s H&D curve, if you will) to prevent the clipping from occurring.
“It’s not as good as having an experienced colorist hand-grading each shot, but it’s a lot quicker,
and a lot better than hard-clipping the highlights”, he said. Graeme said he could explain FLUT in more detail to me,
but then he’d have to kill me; I’m happy to wait until RED can send up an M-X-equipped camera for testing,
at which point I can really analyze the practical effect of the FLUT. From what I saw in the demo area, it looks quite promising.
"Jim Jannard also described the three most important steps for getting good RED footage,
since he’s tired of hearing how “RED lacks dynamic range” (when it’s badly exposed),
“RED is soft” (when it’s badly scaled),
and “RED’s color is hard to grade” (when the grading isn’t done properly).
1. Expose it properly.
When you use the histogram and “traffic light” guides (and the new over/under bars in the upcoming M-X firmware),
you can get perfectly good exposures. Most of the film shot on RED to date used older builds, like firmware versions 15 and 16,
which had even more limited latitude and higher noise than current builds, yet they look pretty good. Yes, the current sensor doesn’t have film’s latitude,
but the M-X is better, and even the current sensor gives fine results when you work within its limits, as I can state from my own experience.
2. Do a full-res deBayer.
Especially if doing a non-integral rescaling, like 3K to HD or 3K to 2K, make sure you use the RED tools to do a full-res decode.
Yes, it takes longer than a half-res decode, but it’s worth it. Half-res can work when generating a half-size output (like 4K to 2K, or “4K HD” to 1080p HD),
but for anything else, you definitely want a full-res decode.
3. Use the white balance sliders in the raw decode before using RGB color balancing.
When you use the white balance controls in the raw decoding tools,
whether in RED’s own tools, the RED tabs in Apple Color, Assimilate Scratch, or other apps, you’re getting the full range of the raw data running through the color matrix.
Do your primary correction using these tools, instead of the more traditional RGB lift/gain/gamma sliders, to exploit all that raw information.
A problem folks run into, RED says, is ignoring the color balance sliders in the belief that the RGB sliders will work just as well.
They won’t: once you convert the Bayer-mask raw data to 4:4:4 RGB, the matrix has been baked in, and if you want to radically alter the color temp of the scene,
it’s too late to exploit all that raw data. Kelvin and Tint work on raw data and contribute to the decoding;
RGB sliders work after the fact and have to deal with whatever the decoding has already done."