Part I: The State of HDR Film Emulation LUTs
Part II: Moving Beyond Traditional Film Print Emulation
Part III: CineD Review of Dehancer Pro
Part IV: Dehancer Print Film Profiles
Part V: Negative and Print Clarified
During an episode of Grade School, a weekly YouTube Live session where he answers questions on all things color grading, Cullen Kelly untangles the confusion surrounding the order of operations, print vs. negative film emulation and film texture, among many other topics.
And another question from Jim: Why do people try do print emulation in the same order as the original process? For example, halation on the negative, etc., when digital is a completely different process, the order should be irrelevant.
Yeah, I’m glad we asked that. That really goes to something that I have strong creative convictions about. Put it this way. I think that there is a pretty firm upper limit to the utility of a forensic level film emulation, and I’ll give you guys a spoiler alert: after having spent time with some of the most knowledgeable people about this stuff in the world, chief among them Dr. Mitch Bogdanovich, who helped design these films for many, many decades for Kodak; after having spent a long time talking with Dr. Bogdanovich and a number of other people who, if they don’t know, no one does, about this process, the more you dig into it and the more you set about the goal of like, I want to take the light that this digital sensor saw and I want to transform every inch of it into the exact same equivalent sequence of manipulations that that light hitting a film negative and going through a film system; from that point forward I want to forensically, precisely model that out – every link in that chain as Jim is suggesting – it’s actually not really possible because at every single turn there are assumptions that we have to make and judgment calls that we have to make because it is so very complex and because, as Jim points out, they’re different systems. They fundamentally see light in a different way. We can account for a lot of that. We can come up with a very good visual reproduction of what that system might have done under a particular set of conditions and that’s another whole topic we can talk about. It’s like, well right off the bat, are you trying to model what a specific roll of film did on a specific day after being shot through a specific lens and processed through a specific bath and printed to a specific manufactured film print, like an individual unit of film print? Or are you trying to get some kind of idealized best fit, like if I average together a hundred different film negatives capturing the same thing in and 100 different film prints capturing the same thing, I’m not going to get the same result every time because it’s an analog format. Do I average those all together and is that a more meaningful model? That’s like the surface, the very outer layer of the kinds of questions you have to scratch your head at if you’re trying to get this crazy forensic level of emulation. So, to answer Jim’s question, I think the reason why people do that is because they’re misunderstanding, they’re misaligning their goals and thinking that, oh, if first of all it’s possible to get this one-to-one forensic recreation – and there’s an underlying assumption there that if I do, I will get the best visual result. Neither of those things are actually true. I understand the temptation to believe them and the temptation to seek them. It only took me a decade of fooling around with this stuff to realize that they’re not true but I promise you they’re not true. So, to that extent, that’s why I think we do it that way, because at least it was in my case for the many years I was working in that way. It’s mislaid expectations and assumptions that like, ‘Oh, if I do this, I’ll get the best possible results,’ and it’s possible to do this in a pure, technical way, in a pure, objective way – it’s not. We have to make assumptions, we have to make inferences and best guesses and judgment calls at every turn. So, that leads me personally to say, okay, the best thing I can do here is get a really dimensional understanding of what that system is doing, using not only the evaluation of imagery like we’re looking at here in the color page but the evaluation of like, what does it do to a cube like we just looked at in Fusion? What does it do to a 1D ramp? There are other, even more sophisticated tools that we can use to really get a good window into what exactly is that thing and what are all of its particulars but then I think the most creatively beneficial thing that we can do is to zoom out a little bit and say, ‘Okay, I’m going to creatively, subjectively borrow the pieces that I find most creatively advantageous,’ as opposed to perpetually looking in the rear view mirror and saying ‘Well we’ve peaked, nothing’s ever going to look better than film and now all we can do is the extent to which our images look good is the extent to which we are able to one-to-one reproduce the results that we would get from a system that is dead or dying.’ I think that’s a pretty depressing thought and I think we can do better and there’s all kinds of stuff that come with film systems that are not desirable. The variability that I just talked about – that’s not desirable. Who wants, who who among us would sign off on ‘How would you like to have a system that gives you one result on one run of negative and print and a different result the next day?’ That’s not optimal, that’s not ideal, that’s just unavoidable with a film system. Another one is that we started the conversation with like, well film prints, there’s no such thing as an HDR film print. Film prints max out at an absolute peak of about 48 nits. What if I want something brighter that still looks good? Am I just out of luck because all I can do is grope at this system that hasn’t evolved past that in a number of years? There’s all kinds of undesirable aspects of film as well, and I think the only way we can borrow the best of the film tradition is to recognize it’s not perfect, it’s just the results of about a century of really smart people working really hard to make images look really good and if we steal the best parts of that tradition and leave behind the parts that they weren’t able to push further – because this is something else that you can hear about from Dr. Mitch Bogdanovich and any other number of the engineers and scientists who are actually working on the problems of image reproduction with film systems. There’s lots of complaints. They’re like ‘We can’t get this any better, we tried; right now we’ve got to ship this print stock and it’s going to have to do.’ There’s all kinds of compromises and things they wish they could do better that we now can do better, that they would have killed to have the ability to do; so if we’re not taking advantage of that stuff, in my opinion, that’s just dumb. Anyway, lots of thoughts about that but that’s my answer to the original question: it’s because I think people are fundamentally misunderstanding what’s useful and valuable about mimicking or borrowing elements of a traditional film system.
All right, on a related note, Defender is asking if can you talk about negative emulations – what’s the difference between print and negative emulation for creating looks?
Yeah, so we can definitely talk about that. So, here’s the first thing to understand when we talk about, you know, we tend to use the acronym PFE or we talk about print film emulation a lot, we’re really talking about negative plus print film emulation, generally speaking. We need to understand that in a traditional film system, the entire thing is a black box that’s designed to work as a unit and that’s kind of one of the challenges of it. A traditional film system is designed for the following: it’s designed to take light in and it’s designed to put light back out in the form of a projected print that looks good; that has a naturalistic and pleasing reproduction. It’s a black box. So, the idea of like, I’ve already talked a bunch today about like ‘Oh, wouldn’t it be fun to sort of be able to cut into that organism and extract like oh I just want this piece of it but not that piece,’ and it is fun and it can be cool. One of the least sensible axes to cut into that organism along is the axis or the dividing point between a negative and a print because there’s a ton of the behavior of a film negative that is done in preparation for the fact that it’s going to be going to a film print. So, film negative was never designed to be pleasing or to produce great visual results in and of itself. Nor was a film print ever designed to be pleasing and produce optimal results in and of itself. It is designed to expect a negative input and a negative is designed to receive a subsequent print transformation. So, that’s the first thing to understand. Now, what we can do, and what is very worthwhile to do is, we can still be modular in the sense that like, ‘Okay I’m going to be in a film system and I’m going to have separate modules for emulating negative and emulating print. I’m just simply going to understand that if I don’t marry my negative to a print or if I don’t marry my print to a negative, I can’t necessarily expect great results because that’s not what the system I am mimicking or modeling was designed to do with these photons, with this light.’ So, they kind of marry together. So, that’s kind of principle number one to understand about neg emulation. And now if we think further about negative emulation, all kinds of other interesting questions come into play that kind of go along the lines of what we were just talking about, like what are the great things and also the limitations of a film negative. Here’s one that is not popular and people don’t like to talk about: film negative doesn’t have the dynamic range that a modern digital sensor does. It just doesn’t. So, what do we do with that? Do I want to mimic the lower dynamic range of a film negative? That’s not the craziest thing in the world but it’s also another example of the type of subjective question we have to answer if we’re going to emulate film negatives – like, what parts do we want to emulate? Do we really want to emulate that part? I don’t necessarily see a benefit to rolling off my highlights prematurely that otherwise would be preserved and available to me in a grade. I’m not sure what the benefit is there. Unless I’m really trying to quote something, if I’m trying make something truly feel like it was captured in the 70s or in the 30s, or whatever, that might be a useful application of such a transform. But that’s another thing that we tend to get upside-down and confused about. It’s like no, you want it to feel that way, like I want my movie to feel like the French Connection or some great 70s film but that doesn’t mean that I actually want to inhabit every single limitation and shortcoming that that film was plagued by unless I’m trying to convince the audience that this is a lost shot from the French Connection. What’s more likely is that I’m trying to steal what’s best and what is signature and characteristic of the French Connection and translate it for a modern audience, for a modern display and for a modern palette because if I don’t do that then I’m just quoting something and really not being much of an artist. So there’s a bunch of considerations like that when it comes to a film negative, like all right, what are we going to do about the the dynamic range problem – and it happens on the top and on the bottom. At the bottom of a film negative there’s an entry point called the D-Min, the minimum amount of light that is needed to stimulate that negative to get a response. If you go below that light, that luminance level, you have a shelf, and spoiler alert, that shelf is above pure black: you can’t give it one single nit above pure black is not enough to stimulate most film negative – some modern film negative maybe but probably not, so that’s another thing you have to think about is, ‘Do I want that kind of hard cut in the toe?’ And just to visually demonstrate that, that would be something like this – this is like what the D-Min toe of a film negative looks like, roughly speaking. Kind of exaggerated, a little bit further, you can see what I mean. So, let’s do something like this: let me set a control point here; so you can see how my blacks are getting quote unquote ‘printy’. We call those ‘printy’ because of the D-Min of of negative and print films; if you go below the D-Min, you’re not stimulating the negative with enough light for it to respond, for it to react, because remember, it’s an organic material, so that’s another thing we have to evaluate like, ‘All right, do I want that in there or not?’ So, this is all a very long-winded way of saying that for me personally, when I start with a one-to-one recreation of a film negative but take away that reduction in dynamic range, I don’t think I want that. Take away that curvature and that lifting of the toe. I may indeed want that but I’d like to do that later; I’d like to have creative control over that later. I don’t necessarily want to do it in two places so do it at the negative end at the print stage I want to do it in one take away any you know like the other piece that we didn’t even really touch on is like what’s the effective gamut of a film negative? How many colors can it see and where are the edges of that? Maybe it’s like well, if I can capture more color and have that color available to me when I’m grading, wouldn’t I want that? Especially if I’m going to be feeding into a look downstream anyway, that’s going to shape it into what I creatively want. So, for me, thinking about the film negative, that’s another piece I’m like, all right, let’s take that away. So, what’s left other than something simple like a matrix which I could probably just incorporate as part of my creative look stack. So, this to me kind of goes back to Jim’s question: everybody has their own way of thinking about this and I really do understand and appreciate the idea of wanting to modularly and tightly implement a recreation of film in terms of like here’s my neg module and here’s my halation and here’s my grain and here’s my print. If that floats your boat and you really like the results you get, who am I to tell you that’s wrong? For me, by the time I subtract the things I don’t want from the neg, all that’s left is stuff that I don’t see any reason not to do as part of my creative transform on the print side. Whether or not I want to call that a print film emulation or not, I just don’t see a benefit to doing such a transform at the head of my node stack and maybe this is all a very long-winded way of saying something even simpler, which is that I don’t see the benefit of imposing technical constraints on my image at the head of the node graph, nor do I really see the benefit of imposing a creative look at the head of my node graph. I’d rather that happen at the tail and I’d rather it happen once, not twice. So, very long answer, but those are kind of some of the principles and ideas there that all start from like, you know, it is possible to take the data from a film negative and say well, what’s the D-Min, what are the D-Log E curves, the response curves for the cyan, magenta and yellow channels… you could pretty tightly model out if I neutralize the light that came from my Alexa and I move it into some common color space, what would the response to that light have been if I had shown it to a film negative? We can pretty well model that out in a one-to-one way. Then the much more difficult question arises of like, well, what aspects of that model do we want? Do we want all of them or some of them and then how do we implement that?
[In a later episode of Grade School, Kelly clarifies that in a digital workflow, the negative to print model isn’t obligatory, and when designing what he refers to as his own digital print stocks, there is no need for one transform at the head of the node tree and another one at the tail. This is in contrast to plugins like Dehancer Pro, which have independent operations for the film negative and film positive and DaVinci Resolve’s built-in legacy film look LUTs, which are expecting to be fed a film negative, meaning that you shouldn’t expect optimal results, not to mention that the look won’t travel to HDR].
Rafa is here asking could you talk about film texture emulation when filming with digital cameras?
Yes, yeah definitely. So, this is something that I’ve been doing a lot of research and exploration in and I think there’s a a good link that I’ll share with you guys after our session today on this subject but I personally think that the ideal place to start with texture is actually before you even get into grain, to think about what is the overall sharpness or softness, or to to use a fancy term that I’ve talked about once or twice here in Grade School, the modulation transfer function: literally, how sharp or soft is my image at the bottom, middle and top of it. So, that’s kind of, in a more objective sense, if we’re talking about wanting to model the texture of a film system, what we really would want to start by modeling is the sharpness or softness of that entire system’s rendering from end to end and get a good reproduction there. And then from there, you could use grain or other additional textural elements; or you could use halation, which is sort of part of those spatial or MTF considerations, just done on a per-channel basis because all halation is, is the MTF of a particular layer of the film. Although halation is defined by the the bouncing of light off of the backing and back for a second time onto one or more layers of the film, so this may be slightly more complex, but those are all useful things. Let me give it a bit of a practical, creative bearing. I’ve shown you guys my halation stack before, but let’s go back to this tasty film print thing that I’ve been looking at and let’s just do a quick halation and grain thing. So let’s look at Isabella’s highlights up here: I’ve got that kind of highlight bloomy thing happening here and I can decompose this and show you that all I’m doing is linearizing the image and then applying a blurred version, just adding it on top of the image, which is roughly what you would get in the model of an actual negative and then upstream we could do any number of grain solutions that we might like. I don’t think I have this grain card pulled in so it’s not very happy with me… let me find some grain, hang tight, let’s get some grain up in here. Okay, so we’ll add this as a matte and now I’m going to select that grain card as my matte and now I’ve got some grain going on here. And as you can see, if I decompose this node, what I do have right now in a pretty broad way, is I just have some softening going on which I’m just doing a straight up blur on it here, that’s definitely not embodying all of the intricacies of the MTF curve of the film system that I might theoretically be kind of trying to model… but again, my whole thing when I’m grading here is I want to get the essence and this to me feels like the essence; this feels like the character, this feels like the thumbprint. I love the idea of getting a forensic model but I’ll put it to you maybe in another way: the distance between this and the tightest, most forensic model you could possibly devise for a film system, including halation, grain and negative and film print colorimetry, I don’t know how great that distance is; if I’m less than 80% of the way to that ideal with what I have right here, I would be pretty surprised. I think I’m pretty much there, so it kind of becomes a question of time-in versus benefit out. I think I’m kind of there to my eye as an artist, and again that should be what we are anchoring everything around when we’re grading today. I don’t want this generation – like the generation of artists that I’m a part of and certainly those that are going to follow me and then follow them – it would really bum me out for our high water mark to be like, ‘Boy, we nailed the film emulation… we got down to like the forensic details, we really got it right!’ That would be kind of a bummer, wouldn’t it? We want to push further, we want to push past that, and that has to start with being willing to exert some autonomy and say, well this is deriving inspiration from that thing but I’m using my eye to evaluate whether or not I have successfully navigated myself to a good visual result or not. And in the same way like I’ve kind of got both of these stacks that I just pulled in here calibrated to very roughly, not in like a super tight sense, but let me just uh flip these back off into their kind of compounded states and turn this grain back on. I’ve got these kind of roughly calibrated to do what what I feel like I see when I’m looking at a piece of printed film of well-shot, well-exposed, well-developed printed film. I’ve kind of got these calibrated to the amount of halation and grain that I see in those and usually the first thing that I do when I drop these in – I almost never use them at full strength – I’m going to drop it to like 35 percent grain and maybe like 50 percent halation, just a kiss, just a little bit of it, so that I can get the flavor in there without necessarily noticing it or being like married to what the film system would have done. I don’t care what the film system would have done except to the extent that I like it, if that makes sense. I want to steal the good and leave the rest behind because my job is not to be a creative stenographer and uh sort of like take the pieces and model the film system out one-to-one because that’s what I’m supposed to do and it’s against the rules to do anything else. I just want to steal what’s good and leave the rest. That’s the the essence of creativity in my opinion, so that’s kind of how I would think about the the textural pieces there and you could play with like where are you getting your grain cards from and what space are you applying them; what’s the scale of that grain; what’s the motion of that grain; how are you compositing it in: those are all variables that we can talk more about in future Grade Schools we’ve talked about in the past and there’s more than one good answer there and it really just comes down to like how are you achieving best results and if you’re not yet satisfied by the results, why not use the actual model the way it would have happened in a film system as your guide to get to a better result but it’s there to help you not to punish you type of thing. If the way that it works in a film system and modeling that out gets you to a good spot, great. If that’s not the way that it works in a film system but modeling it that way gets you to a good spot, also great.
Part I: The State of HDR Film Emulation LUTs
Part II: Moving Beyond Traditional Film Print Emulation
Part III: CineD Review of Dehancer Pro
Part IV: Dehancer Print Film Profiles
Part V: Negative and Print Clarified
The Daejeon Chronicles 