HDSLRs for Video: Beyond the Hype

Marco Solorio One River Media Walnut Creek, California , USA © 2010, CreativeCOW.net. All rights reserved.

HDSLR,” “VDSLR” or whatever else you call them - the interest in “still” cameras that can also shoot HD video is growing fast...and so is their potential. I liken it to the DV revolution that began with the Sony DCR-VX1000, and again with affordable 24p production using the Panasonic AG-DVX100. Both were truly groundbreaking cameras that, at the time, were unparalleled in quality for the price. We are seeing the same revolution begin with HD video coming from these HDSLRs.

Amidst all of the interest in these cameras for HD shooting, it shouldn't be forgotten that these are DSLR cameras first, with still photography capabilities that are nothing short of amazing. The full-frame sensor and its 14-bit, 21 megapixel resolution is a recipe for breathtaking results in skilled hands. Having both tools in one unit is still mind boggling to me, even after over a year of using it.

The fact is that these HDSLR cameras should NOT be shooting beautiful video at all. Their compression rates are high, their chrominance sub-sampling is high, their re-sizing from full-frame to 1080 isn’t smooth, they’re 8-bit at the compression level, and choosing a bad quality lens just compounds all of that. Yet the HD video can still look breathtaking!

I have owned a Canon 5D Mk II for over a year now, and love it. It shoots 1080p, and boasts the only full-frame, 35mm sensor on the market — larger than the RED ONE’s sensor or super-35mm motion picture film, and about the same size as VistaVision. These full-frame sensors are HUGE, and suck up so much light that even a room lit with a single candle can produce beautiful results. I’ve shot 1080p video in such extreme low-light conditions that any other HD camera, including a Sony F900 or a RED ONE, would have quickly degraded to noise and artifacts.

LENSES

HDSLR cameras have added a new dimension to the idea of “revolution:” the vast choice in lens options is every independent filmmaker’s dream come true. No longer are shooters stuck with a fixed lens of dubious quality or limited focal length. No more wide-angle adapters, or telephoto adapters, or even going the extra step with secondary lens adapters. Now, you simply pick the lens you want, specific for the type of shot you want. The only limit is your budget. There are even manufacturers developing lens adapters to fit exotic cinema lenses (like PL-mounted Cooke, and Zeiss optics), which work well with smaller sensor cameras like the Panasonic Lumix DMC-GH1 and Canon 7D.

I’ve been a long-time Canon SLR shooter (both film and digital), so at this point, my lens arsenal is quite large, and includes Canon FD, Canon EF, Nikon F, and Sigma mounts. With my 5D Mk II, I purely shoot with Canon’s best “L” lenses, mostly consisting of their fastest primes and a few zooms.

Although the imagery from these still lenses is astounding, there are differences between them and film/video lenses that you should know about.

You can twist the focus ring on cinema lenses for days from lock to lock, for detailed focus pulls. The short twist on still lenses can be a little annoying if you're tight on a slightly moving subject (even a talking head) with very narrow DOF.

Annoying or not, at least these focus rings move only from start point to stop point. Compare this to the electronic focus ring on a lens like the beautiful Canon 85mm f/1.2L II. Although one of the absolute sharpest and fastest lenses on the market for photography (I know, I have one), a lens like this can seem downright evil for video: the electronic focus ring just doesn’t have that physical feel and feedback to it, and worse, will keep spinning forever, even when you can't focus anymore.

Controlling aperture with a physical ring on the lens itself is also ideal, preferably without f-stop click-points — rather than the push-pull mechanism found on some photographic lenses. Although ring gears can be temporarily added to any lens (I use them), they’re usually bulky, a drag to take on and off, and simply not as effective for follow focus and/or motorized control. The good news is that some companies can modify your photography still lenses to employ some of these cinema style features, including clickless aperture stops and permanent ring gears.

Although I prefer Canon optics, I definitely prefer Nikon mechanics in their F lenses. They employ physical lock-to-lock focusing and a physical aperture ring, making them a good candidate for such cinematic style lens conversions. A trick I’ve sometimes implemented is using Nikon lenses on my Canon body, by buying cheap Nikon F mount to Canon EOS EF mount adapters. There are some good buys on used F mount lenses if you’re on a tight budget but want more physical lens control.

WHEN IN DOUBT, HACK IT

One of the strongest movements in the HDSLR world today is the development of third-party firmware. Trammell Hudson is the guiding force behind the "Magic Lantern Firmware." He reverse-engineered Canon’s first firmware release to bring much needed features to the 5D Mk II. He has also laid some initial groundwork for the new 7D.

Magic Lantern firmware

The Magic Lantern firmware does not replace the camera’s own firmware and features, but merely adds his functions to them — more like a patch — in non-volatile RAM. In fact, the ML firmware needs to be “added” to the camera every time it is turned on, which for safety, is a good thing.

Why would he bother to hack in so many features, and why would you bother to reload it every time you shoot, when you could just buy a video camera? As Trammell notes, “If you can find a camera that a) shoots HD, b) has a 50 mbps data rate, c) has interchangeable lenses, d) has a 35 mm or larger sensor and, e) costs less than $150K…then buy that one instead.”

So what does Magic Lantern add? Dozens of features that cinema lens shooters expect. Zebra patterns, for starters, with threshold levels that can be modified to your needs. There is also recorded lens data, live histogram, custom overlay crop marks for infinite aspect ratios, automated focus pulls, focus stacking, automatic HDR exposure bracketing, and much more. He is also working on timecode integration – in fact, I’m letting him borrow one of my timecode generator units to help with his development.

Magic Lantern Zebra pattern (checkerboard)

Best of all, the Magic Lantern Firmware is open-source and free to use under the GPL license. If you do use the software, please donate. I did!

AUDIO

One shortcoming of these HDSLR cameras is their lack of high quality audio recording. Ironically, the audio recording hardware isn't terribly bad, and the final audio recorded is actually uncompressed.

The monkey wrench is that these HDSLR cameras record the audio with automatic gain control (AGC) always on. AGC is a method of compressing and/or limiting audio input levels so they do not overload the camera's audio preamps. Think of it as an automatic volume knob for incoming sound.

This is normally a good thing as it safeguards the camera's internal audio hardware from inexperienced users that may override the preamps with signal that is too hot. However, an AGC system that cannot be turned off is a curse in any professional environment. It creates a noticeably higher noise floor that can be very annoying.

There are currently two ways to overcome this. The first is the oldest method in the cinematic book, and works with any HDSLR: using dual system sound, whereby the audio is recorded on a separate, dedicated device, where full control and sonic quality is maintained throughout the entire audio path.

Car-mounted rig from FilmTools, the Gripper 490 with 3/8 ballmount and 6" suction cup.

The second method combines the Canon 5D Mk II, the Magic Lantern Firmware and some additional hardware. In fact, the first features that Trammell added to Magic Lantern were to defeat the ugly internal AGC system.

The rig I use includes a JuicedLink CX231 micro mic preamp/mixer — a device with proven, low-noise isolated signal preamps — and a good quality microphone, specific for the sound environment I want to capture. I can hook up (for example), a Sennheiser K6/ME67 combo (super-cardiod shotgun microphone) into one of the two channels of the JuicedLink CX231 via a balanced XLR cable. The JuicedLink is then fed into the 5D Mk II into its line-level 1/8" audio input jack.

With the Magic Lantern Firmware, and ONLY with it, I can then monitor the live audio feed into the camera, as it’s recorded by plugging headsets into the 5D Mk II's external AV output jack, and visually monitor with color-coded peak-meter VU overlays.

Magic Lantern VU meters at top of viewfinder

We’re talking about a still camera, right? Amazing.

The result is the use of professional phantom powered microphones, an input trim mixer for both mic channels, an optionally defeated AGC signal path with audio and visual monitoring, ultimately recording to an uncompressed audio codec.

Firmware hackery at its best, for professional features that should have already been built in! Not out-of-the-box by any means, but an excellent solution to the problem nonetheless.

MORE CHALLENGES

The downside to the massive native resolution in these large HDLSR sensors is aliasing that results from on-the-fly downscaling to 1920x1080 (or 1280x720). This is roughly 2.1 megapixels, down-sampled, in the case of the 5D Mk II, from 21 megapixels!

The current workaround for any HDSLR is to try to avoid shooting against things like brick walls, chain link fences, power lines, etc. If you must, try to distance the subject further away from them, so that the backgrounds become defocused. In fact, even just a hair out of focus will completely solve the problem.

Unlike traditional CCD sensors that capture the entire image at once, HDSLR video is recorded by linearly scanning across the camera’s CMOS sensor one pixel at a time, which can lead to “rolling shutter” distortions, aka, “Jello-cam.” Although only microseconds, the scanning time it takes between the very first pixel and the very last pixel is long enough to create the visual distortion. When objects, or the camera itself, move faster than can be recorded in one pass, the image becomes skewed or squashed. The faster the motion, the heavier the skew and/or squash. Unfortunately for higher-resolution cameras, larger CMOS sensors, with their longer

Click thumbnails below for uncompressed, untreated still frames from footage captured for this article, captured with existing light

distance to scan, are more susceptible to this than smaller ones.

As technology advances, CMOS scan rates will become faster in these large sensors. There are also some new software solutions that can help "fix" the problem in post-production, but it's an extra step, adding both time and a very slight reduction in overall image quality due to interpolated pixel reconstruction. The workaround for now is to keep your pans and tilts slow.

Also, there is currently no support for live, uncompressed HD output from any of these cameras’ HDMI port. This will be a huge leap forward, as it will allow for real-time capture to a much higher quality format than the camera’s built-in encoding format. Using something like an AJA Ki Pro for direct ProRes recording would be invaluable, especially for chroma-key production.

THE FUTURE OF HDSLR TECHNOLOGY

I haven’t talked much about the RED ONE camera in this article, but if an HDSLR company really wanted

to build a “RED killer,” they could. What would that entail? For starters, they would need to address many of the limitations above: reduce the amount of rolling shutter skew, implement a better down-scaling solution, support live uncompressed audio and video output, and support timecode.

In fact, I could see adding a multi-pin “accessory port” on the camera body that tethers out to a multi-port connection, which would then add XLR input, audio monitoring, video monitoring, TC in/out, and anything else that might be needed.

From there, we need to go beyond 30 FPS in 1080 HD. Having true variable frame-rates in 1080 HD would be ideal, like 1-60, 1-120 or even 1-240 and beyond if they can do it without melting the hardware. Likewise, going higher than 1080 HD would really open everyone’s eyes; 2K, 4K, and even beyond. The sensor is certainly capable of it, even if recording moving pictures in that resolution has not been enabled.
A flip-out screen would also be nice.

Seriously, most of these suggestions are features already used in existing products.

Of course, the idea of a “RED killer” is somewhat misleading. RED will not be going anywhere as far as the industry as a whole is concerned. But for many shooters, the remarkable quality of HDSLR cameras mean that investing in a RED camera is a far less compelling option. I'm an example of that. Investing in a RED with the necessary options to shoot even a single frame is not cost effective for my business model. If a RED ONE dropped in my lap from the sky, would I use it? Absolutely.

And would I still use the HDSLR for shooting HD? Absolutely. Am I ditching my Sony EX1 anytime soon? No way. But with some of the advantages my HDLSR gives me over a RED, even if just at a “paltry” 1080 HD resolution, for a much lower price, with all the lenses and bells and whistles I use in my camera rigs, there’s simply no need for us to consider RED right now.

CONCLUSION....AND GETTING STARTED

Despite the technology’s shortcomings (and there are many), shooters, developers and manufacturers are all climbing aboard the HDSLR bandwagon because of the great potential these cameras have now, and in the future.

Even if you are only interested in experimenting with them for video, these HDSLR cameras are practically a no-risk purchase: great quality, a lot of manual control, and at a price-point that can't be beat. For many people, the Canon 7D would be a good place to start. It has full 1080p resolution, employs a gamut of frame-rate options and can take advantage of lighter, less expensive EF-S lenses. My suggestion would be to pick one decent quality zoom in the 28-70mm range, and one decent quality prime, maybe in the 50mm range, buying as fast of each of these (i.e., letting in as much light) as you can afford.

For a 7D body, an inexpensive zoom and an inexpensive prime, you'll still only be around $2500 all in, a far cry from the price of any other HD option, let alone one with a removable lens system. Don’t like it? Sell it right back into the hungry market for used photography gear — the money you’ll lose could be justified as a “rental cost” — or you can keep it is a great still camera.

The most important thing to remember about shooting with HDSLRs is true for every camera, but ESPECIALLY so with HDSLRs. You have to understand your shoot, and know how the advantages and limitations of your camera play into those specific circumstances. Otherwise, you'll either shoot yourself in the foot on production and lose a client, or you will miss out on an opportunity to seriously raise your production values.

I have a feeling that we’ve only seen a snippet of what’s to come from the world of HDSLR cameras, and from here, our jaws will continue to drop.

AND NOW FOR THE MOVIES....
	As mentioned above, I decided to grab model, Jackie Rivero, and visit nearby Broadway Plaza in downtown Walnut Creek, California to capture some images for this article. I brought some of my lenses, but not all. Enough to quickly use the camera in natural low light; we did not use any controllable lights and used what was around us. Unlike the low-light testing in the Agave Nightclub video below, this time around we have manual control to make the shots a little more interesting, namely, faster shutter speeds. To give the video a slightly over-cranked (slow-motion) feel, all the raw footage was conformed from 30p to 24p. It really smooths things out. Click here to play.
	Shot and edited January of 2009, shortly after the camera was received in December of 2008, PRIOR to manual control functionality. This video was shot using the Canon 5D Mark II on a modular shoulder mount rig. The only lighting used was what was available in the nightclub. No controllable or camera-mounted lights were used. The prime focus (no pun intended) was to see how well the camera would record this low-light scenario before using the rig in a real production environment. To give the video a slightly over-cranked (slow-motion) feel, all the raw footage was conformed from 30p to 24p. Click here to play.
	Shot and edited in July of 2009, mostly with the Canon 5D Mark II (firmware 1.1.0) and some B-camera with our Sony EX1. This trailer only has one EX1 clip in it. Can you figure out which one it is? Lenses used for this production: 24mm f/1.4L II, 35mm f/1.4L, 50mm f/1.2L, 85mm f/1.2L II, 135mm f/2L, 16-35mm f/2.8L II, 100-400mm f/4.5-5.6L, Canon 2X extender and Kenko macro rings. Some steadicam, some tripod, some shoulder-mount. Production was shot the weekend immediately following the public release of the Canon 1.1.0 firmware update in July of 2009, and proved invaluable. We wanted to mostly shoot with high shutter speeds, something otherwise impossible prior to 1.1.0 firmware update in low-light conditions. Click here to play.