Some thoughts on RED color space, exposure, and more..
I apologize for the lack of good content on this blog as of late. It’s been pretty slammed here in NY for the past few months, which has been great but hasn’t left me much time for the blog. Sooo many jobs on the RED lately and so many questions being asked.. Number one being – which is the most “correct” ISO/ASA setting to shoot the camera? Reading the R1 manual and message boards yields no truly solid answer to this question BUT there are a handful of truths about the way this camera works that can help guide you to a workflow that will make sense for you.
Sorry, I don’t have any snazzy pictures or images for this post. Just lots of words strewn around without any semblance of grammar. Sorry. Have fun.
Anyone who has ever worked with the R1 can attest that there are no hard and fast rules for exposure. In fact, RED has offered so many tools for evaluating exposure that sorting out what’s what can be downright confusing – RGB “traffic lights” indicate clipping in the RGB video path but not RAW, the weird little color bar to the left of it indicates clipping in RAW but not RGB, the False Color meter is not calibrated to VIEW RAW so is therefore unreliable when viewing RAW but works great with RGB… WTF? Sadly enough, all of this isn’t clearly spelled out in the manual and if you want the info you have to spend hours sorting through thousands of posts on Reduser.net. Anyone who has spent time on that site knows that there are far more questions asked than are ever answered!
So back to the main point of this post - how to rate the camera and what really is the true sensitivity of the sensor? There are so many different ideas on this - rate the camera at 320 for REC709 but 80 for RAW, treat the camera as if it were loaded with 250D, 200ASA for tungsten… the list goes on and on. So what is it? How do you rate the camera? How do you know if you’re actually clipping or not? How do you know if you’ve successfully exposed your way out of R1’s treacherous noise floor? Stu at ProLost has recommended RED publish some sort of exposure guideline to help clarify all the confusion. I couldn’t agree more. This situation is what it is, so test test test and come up with your own guidelines because we’re in the wild west with motion picture RAW and the rules are being made up as we go.
According to Graeme Nattress, one of the guys who writes the Redcode,
“ISO is totally dependent on what LUT is used to map which code value to mid grey…
That's why the ISO of RED is natively set at 320 so you don't clip the highlights like some video camera, and it's also dependent on the precise curve you use to develop the RAW image with.”
So basically by rating the camera at 320 ASA, you are building in a little protection for your highlights. In my experience with the camera, it only wants 3 things in life – not to clip, not to have noisy blacks, and to be 5000 Kelvin. Everything else is totally relative. The smart folks at RED realized that the number one issue with digital is that the knee just can’t hold as much information as film and it simply does not roll off the excess in a visually pleasing way. Their solution to this issue was to nominally rate the camera at 320ASA thus building in the white clip protection via consistent underexposure. As everyone who has tested this knows, if you meter at 320 and view your monitor path in RGB color space such as REC709, checking a gray card with False Color will not perfectly put it in the middle GREEN zone like it’s supposed to. You pretty much have to open up a stop to turn the gray card green so perhaps 160ASA is the true native sensitivity for the camera.. Who knows! There is infinite debate about this over at Reduser and everyone’s got an answer to the question.
Whatever the answer is, I’d rather give the sensor a little less and hang on to a little more information then the other way around. Whatever your RED technique is, the goal for exposure is to find a nice happy place in the sensor where you aren’t clipping the whites and you’re safely out of the noise floor in the blacks. If you do that, you can literally do whatever you want with your exposure in postproduction. I’ve done substantial histogram comparison tests and it’s REMARKABLE how flexible the Redcode RAW is.. But if it’s not there it’s not there and there’s nothing you can do to get it back. And on the other side of the histogram, sensor noise is not something that can be easily removed. So the correct exposure for the RED is one that is a compromise between sensor noise and clipped whites. Sometimes you’ll clip and sometimes you’ll have noise. There’s just no way around it but once you’ve figured out your own exposure techniques, big mistakes can be substantially minimized.
With this camera, I am absolutely a light meter guy. My personal preference for verifying exposure is to use the REC709 color space on the video because it is a nice compromise in terms of latitude between RAW and Red Space. Red Space while it looks very nice and punchy is not very useful for evaluating your image because the gamma is so lifted. Camera RAW obviously has the most latitude – roughly 9 stops by most accounts. But I personally don’t trust the VIEW RAW option for the monitor path and here’s why – the sensor is recording linear light information where gamma = 1. The VIEW RAW feature has to be converted to video therefore some arbitrary gamma has to be applied for it to be displayed. What is the gamma that is applied to VIEW RAW? RED hasn’t come out and said what it’s actually doing so who knows what you’re really looking at. Also, there’s no way to VIEW RAW in the Red post production software so again, there’s no way to go back and look at what you were seeing before. From what I’ve come to learn, more problems are encountered by trying to evaluate exposure based on VIEW RAW than simply metering and looking at the video in a normal color space.
More on that:
REC709, to my eye has about 7.5 stops of latitude whereas Red Space has about 6. My tried and true technique for evaluating exposure with the R1 is to use the REC709 monitor path, set the camera and meter to 320, and open the stop a bit here and there if the histogram needs it. I then use the False Color meter to double check myself. What my meter tells me and what False Color tells me in this setting is usually pretty spot on. Depending on the situation, I generally expose skin tones so that they appear in the gray or green zone in the False Color view. This would correlate to about 40-70 IRE on a waveform monitor, which is typically where you would want them. I’ve found this technique to be consistently reliable. The thing to remember about the fear of clipping is that RAW has a lot more latitude than video. If you’re not clipping in RGB video, then there’s no way you’re clipping in RAW. You can rest easy on that.
I shoot a lot of tests. That’s how I’ve developed these ideas. I don’t own a R1 but whenever I work with one I find a few minutes here and there to shoot a few tests so I can continue to learn as much as I can about what I think is ultimately the future of motion picture images – RAW.
Most of the exposure tests I’ve done have been in daylight because I find that keeping key light as close to 5000 Kelvin as possible yields the most consistent and least noisy results. This is one of the few hard truths about the R1 – it is a daylight-balanced camera and the difference in quality when we work within this parameter can’t be ignored. Tungstens just look a little muddy to me. Especially on skin!
Here’s another very important thing to keep in mind about exposing the Mysterium sensor:
When you adjust ASA or gamma in the camera or in the computer, all you’re doing is affecting the PREVIEW, the down converted monitor paths. This is accomplished by compressing the information either to the left or to the right in the histogram but it has NO effect on what’s being recorded from the sensor. The point of the exposing in RAW is to preserve information. You can set the ASA higher if you’re worried about clipping white or lower if you’re worried about low light noise but it ultimately has no effect on the actual sensor data that’s being recorded!
Experiment with various exposures on set and in Red Alert and you will see that you can create identical histograms with a combination of settings in the software and lens T stops. Set up a controlled still life scene and light it up. Ambient light reading indicates T4 at 320ASA and False Color in REC709 verifies the exposure. Make sure that you have plenty of room in the highlights so that you won’t clip when you overexpose. Shoot the scene and then shoot the same thing again this time overexposing one stop and underexposing one stop. Take these files into Red Alert and set up the correct exposure for Color Space: REC709 and LUT: REC709, 320ASA. Now do the same for the overexposed and underexposed shots. The histograms will indicate the change in exposure. Now in the overexposed shot, drop the ASA down to 160. In the underexposed shot put it up to 640. Now check the histograms – they should be virtually identical or at least very very close.
What this proves is that as long as you expose so that you’re not clipping and getting out of the noise floor, you have total control over your exposure after the fact. You have successfully captured all of the information that the R1’s sensor was able to give you and you can now do whatever you like with it. Think about it, this is a total paradigm shift in the way we create digital motion picture images.