The future of cinematography; or, how I learned to stop worrying and love the DSLR
Let’s get technical.
The human eye is a far superior instrument to the film or video camera. Over the years film stocks have gotten more sensitive, larger negatives and three-strip processes have been developed, and video cameras have improved immeasurably, but these innovations have not brought the baseline visual fidelity of cameras to the level of our own ojos. It’s been claimed that the resolution of the human eye is equivalent to 576 megapixels; while that claim should be taken with a grain of salt, as should any calculations of dynamic range, sensitivity, and field of view, all such assesments lead to one conclusion: our eyeballs are pretty damn good. For decades, the most common observation for first-time cinematographers has likely been, “wow, this camera needs a lot more light than my eyes.”
Of the several videocameras I’ve owned over the years, the best low-light performer was probably the Sony PD-150, because it was interlaced (half the vertical resolution meant faster exposure, essentially), which was exponentially (logarithmically?) more sensitive than what I’ve shot with in the past few years: a 35mm adapter in front of a 24p camera. The most common production problem on The West Side was not having enough light: not surprising considering our light kit consisted of about $100 worth of Home Depot shop lights. Along with a plethora of other production hassles (such as all of our footage coming out of the camera upside-down thanks to the 35mm adapter), The West Side was often nothing short of a production clusterfuck, and ever since we put the show on hiatus I’ve been looking for a better camera setup.
So it was much to my surprise when the Canon 5d Mark II DSLR still camera was announced with full 1080p, 38Mbit h.264 video capability. And when it was released into the wild and users started posting low-light clips, it became clear that the era of the ultra-sensitive, full-size CMOS chip had arrived, and the death knell of film, or at least its inevitable relegation to period pieces and specialty shots, had begun.
To get this out of the way: no, I am not saying digital sensors are better than film today. But all one has to do is extrapolate the arc of technology to know that the sensors will get larger, more sensitive, and sharper, in the same way that technology always advances, to surpass film very soon. It’s telling that post houses are already adding grain to digitally-captured images in an attempt to, essentially, degrade a too-clean image into something that feels more filmlike. In addition to the standard march of technological progress, unforeseen innovations will take digital imaging even further than film ever imagined. To wit: Sony has recently introduced backlit CMOS sensors for better low-light performance, Vision research announced a new Phantom camera that shoots up to 1.4 million frames per second (try that on film!) which can also do two exposures in one frame for increased dynamic range, and other camera manufacturers are experimenting with what amounts to subpixels, enabling the equivalent of real-time HDR imaging. Soon digital sensors will outstrip film in dynamic range, and once that’s the case, the proverbial fat lady can begin taking advantage of post-show craft services.
First, let’s backtrack: RED, Jim Jannard’s startup, was founded based on a big-CMOS-chip approach, and its RED ONE camera delivered on most of their promises. History will likely note RED as the progenitor of large CMOS-based digital cinema, and rightfully so, especially if their forthcoming cameras deliver on expected specs. By no means am I trying to take anything away from RED when I use the 5d Mark II as the harbinger of a future of large CMOS cameras; ideally, I’d own a RED right now, what with its RAW workflow and variable frame rates. Actually, I’d wait and buy a full-frame Scarlet, but I’ll be pleasantly surprised if Jannard et. al actually ship it in the next year, and I’ll be even more pleasantly surprised if I have ten grand lying around (for the “brain” alone) when they do.
But in terms of sensitivity, the RED ONE isn’t a breakthrough like the 5d Mark II is, and a basic RED ONE kit costs $30,000. Which brings me to the not-so-new-anymore Canon DSLR (which I’d preordered once upon a time, but canceled my order due to lack of availability and changing plans). I would wax lyrical about its massive sensor size — the 5d Mark II’s 36x24mm sensor is closer to Vista Vision than it is to standard 35mm motion picture film (see comparison graphic here) — but now that it’s nine months later and I’m only now buying the camera, this point is widely known. Suffice it to say that the 5d is really hundreds of times cheaper than the next full-frame HD camera, and tens of times cheaper than the next available super-35 camera (the RED ONE). Pricewise, the next nearest option is a Sony EX-1 with a 35mm lens adapter attached, which clocks in around $8k (and is far less sensitive to light, but does have a number of other advantages).
Which is why I happily purchased a 5d Mark II (and sold ye olde DVX on eBay); it’s high time I get back to lensing. The tipping point in this decision? Tramm Hudson’s brilliant Magic Lantern firmware, which enables a number of “features” absent on the 5d Mark II, like manual audio levels, headphone monitoring, cinemascope crop marks, overexposure zebras, and a whole host of other handy/necessary features. Before the camera was released I was among the first to wonder if it could be hacked, and now it has; while we still don’t have 24p, there have been a wealth of exciting features already added — and more importantly, there will be more to come. A new era of end-user customization has begun, and the idea of a “hive mind” figuring out how to modify a widely-available camera to achieve high-end results is a game-changer. In fact, if Canon would sell its bare full-frame CMOS sensor at wholesale to a group of motivated digital filmmakers (such as those found on such forums as DV Info or Cinema 5d), I’d bet the various members could pool their resources and create a camera out of available lens mounts and video processing chips, perhaps tethered to a laptop. It’d be the world’s first crowdsourced camera. Of course, this is essentially what RED has done, only under the guidance of a well-funded founder — in fact, several of RED’s employees were plucked straight off of video forums — and the fact that RED is a computer attached to a sensor is borne out by the fact that the camera takes 3 minutes to boot up, as if it were running Windows 95.
The fact that I’m talking about a still camera versus a video camera would have been unthinkable just a couple of years ago, but the DSLR/video-camera convergence is more than just a matter of a shared CMOS chip: beyond the technical side of things, this convergence is going to profoundly affect the careers of anyone involved in still or motion photography. When RED last year announced their DSMC — “Digital Stills and Motion Camera” — my first reaction was to wonder if 6k resolution on a S35 chip was really necessary, even after one accounted for the loss in “true” resolution that bayer filtering brings with it. To this day most pro video cameras are based on three CCDs, one each for red, green, and blue, but the RED cameras (and 5dMkII, and other DSLRs) only have one CMOS chip, which allows for larger sensor sizes at lower pricepoints. In order to separate the RGB signals, however, a bayer mask is applied and 4k becomes more like 2.8k. Of course, the reason they’re doing 6k on the next-gen cameras is to get to a “real” 4k of resolution, but I’m of the opinion that increased dynamic range and better low-light performance (the advantages of larger photosites with less pixels on a same-sized chip) contribute to a better image more so than simply increasing resolution. Considering that 1080p is 1.9k, and that many theaters are equipped with 2k projectors rather than 4k, as a filmmaker I’d rather have a more sensitive camera with increased headroom for color correction and other post-processing than an extra 2k. Of course, by moving to full-frame sensors, as RED is doing with the higher-end Epic and Scarlet “brains,” you get both: more pixels and more dynamic range. But in my musings at the time, what I didn’t think of was that RED wasn’t just targeting the motion picture industry, but the photography industry; seen in this light, their preference for more pixels, even on the lower-end S35 Scarlet, makes sense and is incredibly forward-thinking.
Now that magazine covers have been shot on the RED ONE, RED’s plan is clear: they’re shooting for nothing less than a takeover of the digital imaging industry, from motion to stills and back. When you’re shooting 24-72 FPS on a 6k sensor, why settle for strobe photography? You can now set up continuous lighting and choose any frame from the whole shoot to use as your still. And if the current RED is good enough for a magazine cover, imagine the next ones.
Filmmakers, you are now photographers; photographers, you are not filmmakers. That’s what it means to me, at least; I believe it will prove to be easier to transition from a filmmaking background (producing moving pictures at 24 frames per second, dealing with music, sound, acting, story arcs, suspension of disbelief, etc) — to producing a well-lit and well-composed still image (with perhaps more attention paid to hair and makeup) than vice-versa. Any photographers who disagree, that’s fine, but while we engage in a debate I would recommend shorting the stock of any company whose core business is producing strobe-based photography equipment.
RED’s larger sensors gain them not only more resolution for the still-imaging world, but also introduces them to another market: IMAX (including IMAX 3D). The problem with an IMAX camera is, essentially, that it’s ridiculous: the size and weight of the cameras, the amount of audible noise they produce (good luck shooting indoor dialogue scenes), the 3-minute load times, the cost… they’re not practical unless you’re shooting The Dark Knight, and even then they’re not really practical. Watching that film on Blu-ray, I noticed in the very first shot you can see the helicopter they’re shooting from in the reflection off the building in the left of the frame (maybe a lack of edge-to-edge monitoring?), and there’s a surprising amount of flickering in some of the outdoor IMAX shots. Not to mention the distribution requirements of IMAX — a limited number of theaters, the impractical reel sizes, and a much higher cost per print. To my knowledge The Dark Knight had the most successful IMAX run in history, but its run may represent the peak of the mountain, unless another film surpasses it soon; after that, a 4k (or better) projector in a similarly-sized theater with RED EPIC 645-originated footage will be exponentially cheaper and more practical to the point where one might be able to see dozens of feature films a year at the quality level of IMAX (exclusive of the IMAX segments of The Dark Knight, feature films shown in IMAX theaters today are simply larger projections of the same resolution image). The medium-format RED 645 will be adequate not only for stills work on higher-end fashion shoots but also for IMAX-type acquisition; not to mention what a RED 617 would look like, although I pity the fool pulling focus on one of those (its sensor is an outlandish 168x58mm). One starts to wonder how many stuck or hot pixels a CMOS sensor at this size will have (coming soon to the post-production plugin market: sophisticated stuck-pixel removal): I’m currently sending back my 5d Mark II thanks to 16 (out of 21 million) stuck/hot pixels, which would be acceptable if it weren’t for the one red pixel that’s persistent during video capture. Of course, in addition to CMOS quality-control issues, there’s the question of when RED will actually ship these larger-format brains — they’ve got their hands full with a very aggressive planned product lineup, not to mention opening their ranch outside vegas and moving their operations there… I’d be surprised if the 645 and 617 are finished any less than 18 months after their original target date… which, in the scheme of things, may not be a big deal at all considering their ambitions.
Back to the cheaper side of things. Now, I’m obviously not getting a $250k camera for $2.5k. The 5d Mark II, while the promising first sub-$10k entry in the next generation of full-frame, single-chip CMOS video cameras, has its share of problems (beyond the ergonomic), and they are not insignificant: the 21-megapixel camera quite necessarily drops down to 2 megapixels when shooting video (1080p), but it does so by simply skipping lines, which leads to ugly aliasing artifacts. To my eye it also blows out quite harshly to “electronic white,” without the smooth roll-off of Sony’s Hypergamma or Panasonic’s Cinelike knee settings (effectively giving images an appearance of low dynamic range), although I hope to make liberal use the camera’s Highlight Tone Priority in conjunction with Canon’s Picture Style Editor, which allows you to setup custom settings and curves before shooting, to alleviate much of this). Also, its 4:2:0 chroma subsampling is less than ideal and could make aggressive color correction difficult. It’s also highly compressed h.264 video, but I wouldn’t note this as a negative considering Nikon’s competing models use a far inferior MJPEG codec, and I was actually surprised that Canon stuck such a high bitrate on its first video DSLR; considering HDV is only 25Mbit, and is theoretically meant for video work unlike the 5dMkII, I’ll take the h.264 (unless, of course, someone figures out how to enable the MkII’s HDMI-out, in which case all sorts of new possibilities open up). Finally, its footage looks very video-y, with crushed blacks (even after Quicktime 7.6.2 alleviated some of the processing issues) and an over-sharpened, over-noise-reduced look seemingly unsuitable for narrative work. Furthermore, the CMOS readout on the 5d is slow, not unusably so like the Nikon D90, but slow enough to exhibit clear jellocam when whip-panning. The viewfinder is positioned for reviewing still pictures and is horribly un-ergonomic for shooting video, the camera only outputs 480p while recording… the list of issues goes on and on. But this laundry list aside, the chief issue with the camera is its fixed framerate of 30p, which is nearly useless for anything theatrical and less than ideal even for web video: 24p makes sense for internet distribution not only because it’s the standard motion picture frame rate, but also because a lower framerate makes more sense in a bandwidth-scarce distribution scheme.
But the camera is inexpensive and, all told, an incredible deal (that sensor!). Plus I have some ideas about how to defeat many of these limitations. So here goes nothing.
In summary, film, it was nice knowing you. Yes, we will see major Hollywood films shot on film for years, and probably the lion’s share of best-looking and award-winning pictures, even; but the incontrovertible fact of a march towards larger digital sensors, increased sensitivity, and more dynamic range, all of which we’re just starting to witness — not to mention the future of digital distribution — means that its days are numbered. One day we’ll have to explain to our kids just where the term “filmmaker” came from… and perhaps even the terms “cinematographer” and “photographer,” as we’re all just visual imagists now.
If you’ve had enough of the written word, and I don’t blame you after all that, below is a comparison of CMOS-originated DSLR and RED ONE footage.