ORIGINAL 2015 VOICEOVER
In April of 2019, I released an overhauled version of the 'Display Prep Demo' with newly mastered images and a rewritten voiceover. Here is the text of the old voice over from the original 2015 release.
Hi there. My name is Steve Yedlin. I’m a cinematographer and I shot this demo to compare 35mm film and Arri’s digital camera, the Alexa. Each under some selected lighting conditions: samples from the many varied, nuanced, and challenging situations that can come up in motion photography.
This demo is different from some other head to head tests that I know of for a few reasons.
Firstly, many head to head comparisons claim an impartiality that’s simply impossible to deliver. This demo does not claim to be impartial: it has a pointed agenda, and I’ll be transparent about that agenda and the techniques used to accomplish it. More on that later.
Secondly, many format tests are built upon an intuitive certainty that the formats being compared each have an inherent "look" and that the very purpose of the test is to find out what that look is.
I believe that's a problematic presumption to build upon, and I think the premise itself should not be accepted without question. I think a more likely model (which I'll put to the test rather than presume) is to think of the camera merely as a data collection device and the so-called “look” comes primarily from how the data is prepared for viewing. The manufacturer’s recommendation on that preparation is just one of many options for the look, and is not an inherent attribute of the capture device.
Approaching motion imaging with the assumption that a camera format has an innate look is a trap: because if you go in believing the look is inherent, it means that you're not doing anything to control it, and THAT means you're not taking control of the sprawling variables. You're presuming that an out-of-control version of the image chain is the only version of the image chain.
So, while I do believe that cameras are not all the same, I believe that the important differences are technical and not aesthetic: how much noise does it have, how sensitive to light is it, what kind of contrast range can it handle, how subtly and coherently can it gather color data... things like that. If a camera system records ENOUGH information (information of the right kind), you can prep it for viewing any way you want… to have any “look” you want, and you're not stuck with some mystically enforced attributes.
In the pre-digital days of photochemical capture, it may have FELT as though there was an inherent look to a capture format, but that was merely because there was only one reasonable and available method for prepping film to be viewed: you simply had to process the negative and print it. Sure, you could do a little forced processing or you could choose between a couple of available print stocks, but that was all there was for flexibility. For the most part, the available paths for prepping it for viewing were so narrow as to give the impression that the whole image chain had an unalterable look.
These days, when the preparation for viewing is done digitally instead of chemically, the image data collected by the camera can be translated for display any way you like, or perhaps more often these days: in ways you DON'T like if you fail to take careful control. The now sprawling options of the digital image chain create pitfalls as well as opportunities.
But if you know what you want after controlled tests and are able to describe it precisely enough to represent it as a mathematical expression, you can get any desired look within the limits of the image data collected. Of course it is necessary to isolate variables and control the entire [imaging chain] to do this correctly.
And that brings us back to the agenda of this test. Although there are many contemporary looks, I personally really love the “look” (or method of display preparation) that’s traditionally been associated with film. Traditional camera negative printed onto photochemical print stock. You may ask: if the new technology allows any style at all, why do I prefer the old one that used to be the only option? I don't know -- I just like it. Maybe I'm not very imaginative, but I think it looks great and has very complex skin tones, I think that the work that Kodak and others did on it for over 100 years is of great value to the cinematic richness, and I also think that almost no one shooting digitally today is actually getting this look, whether by choice or by circumstance. In fact, I think that even many films today that are acquired on film are starting to embrace display characteristics that have customarily been associated with digital capture.
Anyway, I love that look of traditionally printed film. So, for over 10 years now, I’ve been developing my own empirical analysis of film’s response characteristics and also my own math for preparing digital cameras to coherently emulate those display characteristics.
So: the agenda of this test is to convince you that I can use these image science methods I've accumulated to prepare digitally acquired image data for display in a way that emulates the characteristics that we recognize as filmic.
In this demo of Alexa next to film, the claim is not that the two images are literally identical in every nuance (obviously they're not, as you can see here), but that the film response has been well enough characterized and emulated in the Alexa footage, that neither the film nor the Alexa footage looks more filmic than the other. That the substantial part of the look comes from display preparation and not from the capture format.
In attempting to do this, I've allowed myself the liberty to use all these techniques I've acquired for emulating film's complex color response, its tone curve, and even its grain structure.
Although on the one hand I've allowed myself this liberty, on the other hand, there is a liberty that I've not allowed myself. And this is perhaps the most important attribute of this test. I have prohibited myself from adapting the preparation of the Alexa footage to these particular shots. If you were going to shoot an actual movie and not a test such as this one on Alexa and wanted a traditional film look out of it, you wouldn't have actual film of all the same exact shots to use as a reference, so I've not used the film as a reference here. I could have eliminated the extremely small color differences you see here if I'd not restrained myself in this way, but I wanted you to see how close I could get these two formats without referencing one to the other.
I've merely prepped the Alexa footage exactly as I would do if I was applying my usual film emulation and had no side by side reference. Also, I've used the same version of that algorithm on all the Alexa shots. There was no shot-to-shot adjustment of the film emulation, no adapting to the individual shot -- just the same mathematical transformation applied to everything. The same one I always use when there's no film reference.
That's the gist of the test, so now a couple of quick side notes:
Firstly, since you're actually viewing the film footage digitally and not as a traditional print, the film itself, sort of like the Alexa, has also been prepped for viewing to simulate as well as possible how it would have looked if it had been traditionally printed.
Secondly, although the COMPLEX look, such as the tone curve and color response was not adjusted shot to shot, there was a color correction step at the end of the process where film and Alexa were adjusted shot-to-shot and side by side for a final polish. But this was very pointedly limited to overall color balance and luminance balance, and did not interfere with the complex properties of the film emulation. So, this final polishing shows that the film emulation can stand up to color correction: that it's not limited to working at only one color or density setting and doesn't fall apart with adjustments.
Okay, those are my footnotes. So now, to recap:
We've got film next to Alexa shot under the same circumstances and with the same lenses. Each format has been prepped for display to best match the properties of traditional film acquisition and photochemical printing. Most importantly: each of the two formats has been prepped separately in its complex look, based on pre-existing empirical data and not on subjective decisions based on these particular shots. Then only after the complex look has been applied, there's a subjective color polish that's been limited to only the simplest overall balance.
Okay, there you have it.
I'll also quickly mention that the two formats are shown here in a jumbled way, so that if you'd like, you can do a blind taste test.
If you do so, though, please don't treat it as a parlor game to see if you can pick out the film, but as an even handed way to to reflect on the question: are these actually two different "looks" or has film been well enough characterized that we can say that the substantial attributes of the look come from how the image is prepped for display and not from the capture format.
Now that you’ve watched the comparison with me talking non stop, I suggest that you turn the sound off and re-watch it silently with all your attention on the image.
Thanks for watching.