Stunning Good Looks

RED ONE Build 15 EI Test

The idea behind this test was to underexpose a neutral reference and see at what point the RED’s noise becomes too much of an issue. Using Colorista in Final Cut Pro, I put 18% gray at 50 units on the waveform monitor for each exposure, effectively “printing up” each exposure to see how much “grain” (noise) showed up. Some of this is exaggerated by compression artifacts, but I think you can get a good feel for what’s going on.

I’m not sure how this would look on a film out, but to me there isn’t a ton of difference between noise levels at 320 and 640, although I think I’m more comfortable rating the camera at 320. I start seeing reduced contrast at EI 1280, which I really wouldn’t want to try on a paid job.

Compressed via FCP Compressor, H.264.

RED ONE Build 15 Flesh Tone Latitude Tests (tungsten)

In my quest to figure out just how much overexposure latitude I can get out of a RED under tungsten light, I decided to do a test with a flesh tone reference and a RED with build 15 loaded. It turned out to be almost a completely different camera.

Previously I was only able to see two stops of overexposure latitude on a Kodak 18% gray card under tungsten light. That was using build 14. This build seems to yield 3 stops of latitude before clipping flesh tone, which already has more red in it than a gray card. Also, instead of the red channel clipping and causing highlights to turn cyan until the other channels clipped, I see no indication of that at all now. The clipped highlight is very clean and holds color a lot longer than I expected.

On top of that, the DRX highlight recovery tool in Red Alert now seems to cause more problems than it solves. When I used DRX on these clips they turned green, the exact opposite of what they’re supposed to do--which is to take red channel clips that have turned cyan and use the other two unclipped color channels to rebuild detail in the clipped channel, while using the white balance meta data to blend all of that into a neutral highlight.

I don’t know what they did to keep the highlights so clean, but I’d be willing to take this camera into a real world situation now. The only thing that bugs me is that the red channel, which used to be the cleanest color channel by far, now seems very noisy. Maybe that’s a side effect of whatever was done to clean up the highlights.

I’ve now had independent confirmation from a couple of people who watched my RED build 14 exposure tests and have had great luck treating their RED as a daylight film stock. Apparently clipped highlights are significantly less of a problem and noise in the shadows drops way down. It’s a much smoother look.

If you haven’t had a chance to test your RED before shooting to see how well you can recover clipped highlights, it might be a good idea to go HMI or daylight if at all possible. I’m hoping to test recovering clipped highlights and blue filtration by the end of the week, so stay tuned.

Meanwhile, look back a couple of posts in this blog if you need to learn how to recover clipped highlights in Red Alert, and remember that the same control in Red Cine, called “highlight,” does work--but you need to “jiggle” or double-click another control before the new value takes effect. That is slated to be corrected in the next build.

Like a lot of people with Final Cut Pro or After Effects, I’m trying to become a crack colorist in the privacy of my own home. It’s unlikely to happen anytime soon as I just don’t have the training or the time to take on a second full time career. But--that doesn’t mean I can’t do a little bit of damage to my own footage when I want to experiment. Or when I’m too cheap to hire someone else to do it.

There’s been much discussion on the Cinematography Mailing List about rookie colorists and why it’s a bad idea to try to do these things yourself on a wing and a prayer, but I figure the more I work on my footage the more I’ll learn about giving others the kinds of images that they can improve on my behalf in the future. I’ve done the same thing editing my own footage, and I’ve learned a hell of a lot not just about shooting for editors but also how to fix the kinds of mistakes that just happen on sets when we’re working against time and budget.

I don’t have a color critical monitor, and as much as I’ve tried to calibrate my Apple Cinema Displays to look something like what I might find on a badly tuned TV set in a third world nation, I can’t quite get close enough to feel comfortable. If I’m working with a piece of footage that’s been horribly screwed up in post by someone else (I once shot some jeans spots where the client decided the whole thing should have a blue cast because, well, jeans are blue, and isn’t that a good enough reason?) I can bring it back to normal, but creating rich sophisticated looks on a questionably-calibrated computer LCD is a little frustrating.

The one thing I can’t screw up too much is luminance, and the single most beneficial thing I can do to my 8-bit HD footage is to pop in a Power Window and reduce bright unclipped areas that I couldn’t control during shooting. This is super simple in a telecine suite on someone else’s dime, but where I live and work HD footage is made on the set and never touched again. That’s fine, we do good work anyway, but it’s nice to have the opportunity to take footage to the next level. And since I’m only affecting brightness, which I can judge reasonably well on a computer screen and via waveform, I’ve got a lot more confidence controlling luminance alone than if I decided to cool the shadows and warm the highlights with any degree of finesse.

I’ve got Digital Film Tools 55mm, Apple’s Color, and Magic Bullet’s Colorista, and all can do vignettes (frequently called Power Windows because that’s they’re designated on a high-end Da Vinci color corrector). So far, for ease of use, Colorista wins hands down. It’s incredibly simple to quickly create a Power Window.

In this shot, from a Microsoft Zune spot shot on a Varicam (with a Pro35, Zeiss Super Speeds, and while riding a Steadicam operated by Tim Bellen) I’ve always been bugged by not being able to cut the laptop brightness down a bit. We were shooting in a cafe in Santa Cruz, and our morning ritual saw the crew standing around on top of a train trestle for two hours waiting to shoot the first shot while the day’s creative was re-written. As a result we ended up in this cafe shooting day/interiors after the sun went down. It ended up being a 17 hour day, and at some point our mission became doing the best we could before the poor Steadicam operator went numb from the waist down. (Every single shot for two days was a Steadicam shot. Every. Single. Shot.)

In this case the ambient light was established by erecting a couple of 12x12 gryffs in front of the windows on one side of the restaurant and bouncing PARs off them to recreate the daylight look. After that it became fairly simple to wheel around some Kino Flo Image 80’s to quickly shape whatever area we were shooting in. In this case we propped an Image 80, with 216 on it, on a table in front of our actress and started rolling. She looked great; the laptop looked hot but it wasn’t horrible, and my hope was that the viewer would be more interested in her than the laptop.

These stills are a little deceptive because the laptop looks better here than it does on DVCProHD, but take my word for it--the brightness competes a bit with the actress’s face. Not a lot, but enough that I wanted to try to focus a little more attention on her.

Colorista’s controls are very simple: you pick a vignette shape (ellipse or rectangle), place a top point and bottom point, set a width using a slider, feather, and done. You can see the vignette as a shape alone or as a red mask on top of your footage. Tracking is done in the usual way in Final Cut Pro, by plotting key frames on a timeline. (I have After Effects but haven’t learned it yet. Who has time???)

I’m going to go through all of this Zune footage and look for opportunities to focus attention by darkening corners and edges. I used to have this done to film all the time but I’ve yet to get any of my HD footage in front of a professional colorist. I’ve got great DIT’s who paint my cameras phenomenally well, but being able to dodge-and-burn HD images after the fact is a wonderful, wonderful thing. It will only ever end up being seen on my reel… but that’s where it counts.

(I’m in the process of reading “The Art and Technique of Digital Color Correction,” by Steve Hullfish. It’s a great primer on the art and craft of coloring. Highly recommended.)

New from RED: how to correct those bothersome clipped highlights I’ve been freaking out about for the last few days. Currently the procedure only seems to work in RED Alert using the DRX slider. This process is intended to work within REDCine using the “highlight” function but at the moment neither Adam Wilt nor I can get it to work (Intel Mac, build 90). That’s being looked into right now.

DRX works to reconstruct highlights in a clipped channel using information from the other channels. It then blends the reconstructed pixels and the original pixels together to create the most natural-looking effect.

When shooting, try not to clip more than one color channel at a time if you can avoid it. One channel isn’t hard to reconstruct. Two make it more difficult. It’s pointed out to me that two channels will rarely clip at the same time, so for part of the image only one channel will be repaired, and in other parts of the image two will be repaired (although the quality may suffer when two channels are clipped).

In RED Alert, open the R3D of the shot in question and look at the histogram. If you’ve got at least one channel that isn’t clipping, you’re in reasonable shape. If you have two channels that aren’t clipping, even better.

Using the exposure slider, back the exposure down until the curves are just touching the right side of the histogram. Then dial in the DRX slider until things look right. Make sure the matrix is turned ON with your desired white balance in place before using DRX, otherwise the algorithm won’t know what white balance you want and won’t know how to reconstruct the channel(s).

That’s it. It’s that easy. The hard part will be keeping an eye on the camera histogram when shooting to protect the quality of the highlights. It’s a strange new world, this land of RAW, and waveform and vectorscopes aren’t the only tools with which we need be familiar. The histogram is our new best friend, as that will be what tells us the quality of the data we are capturing.

A huge “thank you” to Graeme Nattress of RED for his help in solving this issue. I hope we’ll be able to bring you more info on this subject, and others, soon.

RED ONE Build 14 Latitude Tests

My goal was to see if RED’s 5000k chip is limited in exposure latitude under tungsten lighting conditions due to a tendency for the red channel to clip early. This seems to be the case, but I’m told there’s a post fix for this problem that I hope to learn about in the near future.

Here’s what I did:

-Shot a Kodak 18% gray card, with some texture to it, at different exposures to see where the camera clipped and to see where significant underexposure noise occurred.

-Shot two tests, one under tungsten light and one under tungsten + full CTB, to see how the camera did under both kinds of light.

-Originally set exposure by setting gray at 50 units using the camera’s Rec 709 output, which turned out to be a stop slower than REDLog would have me believe. Zone 4 on Rec 709 turned out to be Zone 5 in REDLog. (The ASA appears to be a true 320.)

-Took the darkest three tones for each lighting situation and boosted them to 18% gray value to better see noise; also isolated each color channel to see where the noise was coming from.

-Included histograms for each clip from Red Alert.

-Captured in RedCode28, 4K 2:1, 23.98 fps and 1/48 shutter.

I wasn’t able to check underexposure latitude as far down as I wanted because of the ambient light in the test location.

The process was: open .R3D in Red Alert and export clip; then reset white balance to 5600k and capture the histogram to see what the daylight-balanced chip was doing in each situation. The clip was scaled and output as a ProRes HQ Quicktime, 1280x720, and assembled on a ProRes HQ timeline in Final Cut Pro 2. It was then output via Compressor using H.264 VBR encoding at 1k/sec. bit rate, where I tried to keep the file size down without letting the noise get lost or exaggerated by the compression process.

Enjoy!