Canon Picture Styles: Shooting Flat or Not?
When Technicolor released the CineStyle picture profile last year it immediately became a hit amongst Canon DSLR videographers. After all, this is Technicolor. These folks have extensive experience in color science, image processing and digital intermediate. So after this introduction the following may come as a surprise to you. The thing is, unless you know exactly why you are using CineStyle, then chances are you will get better results by not using it. This article talks about dynamic range and picture styles, and attempts to explain the Why’s behind the previous statement. Also, we are focused here on Canon picture styles but the principles apply to any DSLR or video camera with 8-bit video.
Dynamic range
Scene dynamic range specifies the ratio between the luminance of the brightest whites and the darkest blacks in a scene. Similarly, camera dynamic range specifies the ratio between the luminance of the brightest whites and the darkest blacks that the camera can capture. In photography this range is usually measured in f-stops (or exposure values, EV) which is a logarithmic measure (with base 2). Each successive stop is double the light. So a dynamic range of 10 stops means a contrast ratio of 210:1 or 1024:1.
Current digital sensors have some pretty good dynamic range capabilities. The ARRI Alexa and Red MX sensors capture dynamic range well in excess of 13 stops. The dynamic range of the APS-C sized Canon 550D/t2i sensor is about 11.5 stops and the dynamic range of the full frame Canon 5D mark II is around 12 stops.
The RAW data of a Canon DSLR camera is 14-bit. This means that the analog-to-digital conversion on the sensor yields 14-bit values (between 0 and 16383). This is enough to represent accurately dynamic range spanning 12 stops when encoding it linearly. This is all nice and good, but unfortunately for videographers that’s not the signal we get out of the camera.
Most digital cameras feature settings for picture styles. These are sometimes called picture profiles or creative styles (on some Sony cameras). The picture style defines some parameters for the image, most notably color handling. But in this article we are more interested in another property of custom picture styles.
The raw image of the camera is in linear space. Which means image pixels record luminosity in a linear fashion. The picture style specifies a curve which is applied to this input image in order to make it presentable. This is usually some sort of S-shaped curve to make the image more contrasty: shadows and highlights are compressed. Custom Canon picture styles further enhance this behaviour by allowing curves to be applied on top of the default curve. And these additional curves are what we are interested in here.
See, Canon DSLRs (and all DSLRs, for that matter) capture consumer grade video. By design, video from DSLRs is not meant to be tinkered with but displayed directly on consumer displays. Consumer displays are almost universally 8-bit displays, with some of them actually having 6-bit panels and dithering colors up to 8-bit. Unsurprisingly, consumer level video is 8-bit too. Blu-ray discs, for example, have 8-bit video. Canon DSLR video is also 8-bit. And that’s where picture styles come into play.
The picture style tells the camera how to put all that RAW dynamic range into 8 bits (with gamma encoding on top of it). Back in the 90′s Kodak made the Cineon system for scanning and digital intermediate of film. The digitized data was 10-bit and the guys at Kodak actually came to the conclusion that 8-bit was good for representing around 6 2/3 stops, or less than 7 stops of dynamic range. Encode more range, and you start getting banding issues in the grading process. Digital cameras now routinely output between 8 and 9 stops of dynamic range in 8-bit video and jpegs. This is mostly achieved by rolling off the highlights and pressing down the shadows. Most of the coding space is reserved for midtones. This is the well-known S-curve. So in this encoded dynamic range of 8-9 stops there is good detail in, say, 5-6 stops in the mids.
The camera sensor has lower signal-to-noise ratio in the blacks and higher signal-to-noise ratio in the whites. This is because noise is more or less evenly distributed on the sensor but there are less photons hitting the sensor in the darker pixels. For this reason, the 8-9 stops of dynamic range baked in the encoded video (or still picture) are taken from the upper end of the RAW dynamic range. This ensures cleaner image. Incidentally, that’s also the reason RAW images usually have more exposure latitude for pushing shadows up than for pulling highlights down.
The case of flat picture styles
Flat picture styles try to emulate how film negative captures the scene. Modern negative stocks can capture full detail in some 11-12 stops of dynamic range. Compare that to the 6 detailed stops in 8-bit video meant for display. On top of that, there is even more exposure latitude in the rolled off highlights above the shoulder and in the dark tones below the toe. Because equal negative density is allocated to each stop, the film is said to work in Log space (from logarithmic). This may be somewhat confusing at first, but bear in mind that the f-stop axis is already logarithmic: each successive stop is double the light of the previous stop. Having that much dynamic range captured in full detail allows the cinematographer a lot of latitude in the way the scene is shot. Exposure errors can be fixed and decisions about the overall tone of the shoot can be deferred to post-production. The film is then printed on a contrasty release stock which returns contrast back for presentation. (You can read a more detailed introduction on film dynamic range here.)
Flat picture styles aim to allow for similar exposure latitude and post-production flexibility. To increase the recorded dynamic range in general, and to have good detail throughout this recorded dynamic range. Flat picture styles do this by distributing the coding space equally over the exposure range: each stop of light is being encoded by the same number of values. So, an ideal Log distribution is a straight line. For example, if we want to encode 10 stops in 8 bits, we will have around 25 values to encode shades in each stop. In practice, the curve is rarely totally flat because there is too much noise in the low end of the range. Lifting the blacks also lifts the noise. So allotting a full bucket of values for them would be a waste of coding space. That’s why the extreme darks get compressed a bit even in flat picture styles.
One clarification, before we go further. Flat picture styles should not be seen as a way to increase dynamic range. In fact, they may not increase DR at all and still be useful. Usually there is some dynamic range increase over the factory picture styles. These additional stops (or a fraction of a stop) are almost always in the shadows, and they are noisy, compressed and lacking color fidelity. There is no real detail in there; just some notion of tonal change, at best. The real gain is in the stops which are already there at the bottom of the curve in the factory picture styles. The flat picture style just expands them and makes the detail in these stops available in post. In that sense, flat picture styles increase the usable dynamic range.
As you may have guessed, it is not only dances and songs in the case of flat picture styles. They have a serious weakness. The 8-bit coding space simply sucks for representing extended dynamic range. The more dynamic range you encode, the less precision there is in tonality distribution in the usable dynamic range. That means that video shot with flat picture styles breaks faster when pushed in post; posterization (banding) happens sooner when stretching the range around. In a sense, what we get in exposure latitude, we lose in tonal precision and gradation. So, flat picture styles offer a trade-off, and not the solution to all problems.
The flat picture style is in essence a production format. The same way as Super 35 negative film or RAW video are production formats. Flat picture styles are meant for post and not for presentation. Usually, the image needs to get some contrast treatment to make it presentable.
Technicolor CineStyle
The Technicolor CineStyle picture style was developed by Technicolor in order to assist their digital intermediate process. Cinematographers often utilize different cameras on a shoot, including Canon DSLR cameras. And Canon DSLR video doesn’t mesh well during DI with scanned film negative footage and footage from high-end digital cameras that shoot Log space video. That’s why they developed CineStyle to have the output of the Canon DSLR in Log space. It came as a surprise to Technicolor that so many people actually donwloaded CineStyle. The recommended settings for Technicolor CineStyle are -4 Contrast and -2 Saturation. And it also comes with an optional LUT which can be used in most NLEs or compositors and applies a S-curve to the image to make it presentable.
I’ve measured the dynamic range and values distribution of CineStyle through multiple exposures of a Danes-Picta BST13 reflective grayscale chart with steps of 1/3 stop (a copy of Kodak Q-13). This test does not pretend for scientific accuracy but nevertheless gives a pretty good idea of what’s happening with the image. The test was done with a Canon 550D/t2i but the results should apply to any Canon DSLR camera.
There are a couple of peculiarities about Technicolor CineStyle. Both can be seen in the chart.
First, somewhat controversially, CineStyle sets the black point at 16. The values [0..15] are practically unused. Second, and not as obvious, the whites are brought down significantly. There are still values recorded all the way up to 255, but a whole lot of the value range is dedicated to the highest 1/2 stop. This is quite the opposite of the typical knee and knee slope approach to handling extreme highlights. It is worth noting that there is no official info from Technicolor setting things straight about this. There is a notice on their site, saying “a detailed white paper and case study will be developed in the near future”. But it’s been there for ages with no white paper on the horizon (at the time of writing this).
Various explanations exist for this range distribution. Mike Seymour advocates the idea that lifting the black point is reducing noise by allowing negative values (with respect to black at 16) and having the smoothed average exactly at black (read his post for details). This makes sense but there are problems with the idea. For one, even at high ISO settings with lots of sensor noise there are only rarely any pixels with values below 16 (and none below 14 in my tests). Another popular explanation says that the h.264 codec doesn’t like small values. That’s why lifting the blacks to 16 preserves them from being butchered by the codec. Finally, there is the idea that lifting the blacks and limiting the whites is simply done in order to put all the important info in the broadcast legal range of 16-235. Even though there are values all the way up to 255, important whites are pulled down below the 235 limit.
My take on this matter is different. I am convinced that the lifted black point is entirely connected to film and digital cinema Log workflows. For example, in 10-bit film scan formats like Cineon/DPX the black point (D-min) is placed at 95 for reasons that I won’t delve into here (it is concerned with the uneven grain structure of unexposed/underexposed film). Converted to 8 bits this would be around 24. Incidentally, D-min is around 4 2/3 stops down of mid-gray (18% gray). In my measurements of CineStyle’s distribution 4 2/3 stops below mid-gray came at around value 25. This theory also explains the limited precision (due to the relatively small inclination) in the straight portion of the curve resulting in apparently badly used precision in the extreme whites. Both the lifted black and the low gamma of the curve are needed for seamless intercutting of footage from Canon DSLRs with film scans and digital cinema Log footage without any significant import pain. Which is the reason for the creation of CineStyle in the first place.
And there is, of course, the possibility that Technicolor did the blacks lifting for very specific reasons connected to their DI workflow. Hopefully, the white paper will materialize some day to clear this point. But whatever the reason, limiting the useful value range doesn’t look like a good idea when recording video in 8 bits. This decreases the already precious space allotted to each stop even more and invites banding artifacts in post. Still, many videographers use Technicolor CineStyle because, well, it’s Technicolor. But there are better alternatives in the Canon DSLR world. At least in the likely case that you are not doing your DI at Technicolor.
One side effect of the lifted black point is an increase in perceived noise. Pushing black up to 16 also lifts the noise always present in the shadows and makes it brighter and thus more visible. This is not noise created by CineStyle, the noise exists in other flat picture styles too. But being darker makes it harder to notice.
Other flat Canon picture styles
Even before Technicolor CineStyle there were flat picture styles attempting to increase the useful dynamic range. The easiest option is to use one of the lower contrast factory Canon picture styles – Neutral or Faithful – and decrease the Contrast setting down from the default 0 to -4. You can see Faithful at -4 Contrast in the chart above. The curve of Faithful becomes flat later than CineStyle but keeps being flat all the way to the limit of 255. This means more range for each stop in the mids and the highlights, compared to CineStyle. CineStyle has more detail in the shadows and the extreme highlights though. Neutral looks identical to Faithful in terms of dynamic range. The difference is in the color treatment. Neutral is closer to Standard, and Faithful is closer to Portrait. Canon claims that Faithful offers colors closest to the original scene colors. Note that the difference in contrast between Standard, Portrait and Landscape on one side, and Neutral and Faithful on the other, is small. In fact, Canon says it is mostly due to color tone balancing.
When the basic styles at -4 Contrast are not enough for the job, other custom picture styles may come in handy. Marvels Cine is one of the oldest attempts for a flat picture style. It is currently at version 3.4. This version is based on the Neutral Canon picture style. And it is created to allow for correct exposure evaluation through the Canon DSLR LCD screen. This is a common disadvantage of flat picture styles: by lifting the range they brighten the LCD screen of the camera. This makes exposure judgement through the screen difficult. Some people actually compose and setup exposure in another style (Standard, for example) and switch to the flat style for recording.
The Flaat family of picture styles is more recent. It includes 4 picture styles – Flaat_09, Flaat_10, Flaat_11, Flaat_12 – claiming 9, 10, 11 and 12 stops of dynamic range respectively. But remember what we said above. Dynamic range and usable dynamic range are two different things. Digging too deep in the darks is not recommended due to both noise extraction and loss of tonal precision in the mids. These styles have the flattest curve from the shadows to the highlights compared to both Technicolor CineStyle and Marvels Cine. Flaat_10 and Flaat_9 are the styles most will be interested in, as they offer good flat tonal distribution without excessive noise in the shadows. The Flaat picture style family comes in two variations: one based on Neutral and one based on Portrait. This is handy, considering how color handling is the main (but sometimes forgotten) reason for selecting a specific picture style in the first place.
Skin colors are one area which can be affected negatively by flat picture styles. Having the right skin hues won’t help much if tonality is lost due to insufficient tonal precision in this range. Lacking gradation in the range of the skin tones is a highway to plasticky skin. This is the reason Marvels Cine v3.4 supposedly keeps the linearity of the values (does not flatten them) in the most common range for skin.
Not shooting flat: getting the desired look in camera
The position of this section in the article may seem illogical. Why haven’t we started with not shooting flat in the first place? I believe that only by getting acquainted with the shortcomings of flat shooting one can truly appreciate the alternative.
Getting the desired look in the camera has been a popular mantra amongst cinematographers for decades and also a reason for pride, considering how this is a significant demonstration of skill and knowledge. Utilizing specific film stocks to achieve a specific look was a main part of the process. In the digital realm picture styles can be thought of as the analog of film stocks. By selecting a specific picture style and playing with the style’s parameters one can create various looks. While the factory Canon picture styles offer some decent variety and will work for lots of people, there are tons of custom picture styles online, each targeting a specific look, including simulations of popular film stocks like Kodachrome, Velvia, Ektachrome.
The point is, getting as close as possible to the final look during the shoot will limit the need for adjustments in post. And post hurts the image. The heavily compressed 8-bit video simply is not meant to be abused in post.
DO’s and DON’Ts
- Getting the look you want in-camera is the best solution. This means less tinkering in post, with less possibilities for breaking the image. It just so happens that this is not always possible. Nevertheless, when shooting in controlled conditions and for simple projects that should be the default option.
- Shooting flat is a trade-off between usable dynamic range and tonal precision throughout that range. Try to use the most contrasty picture style that does not clip or limit important highlights and does not excessively compress blacks. This ensures the best possible tonal precision.
- Don’t bother with CineStyle unless you intercut with film or digital cinema footage. For better tonal precision stick to Neutral/Faithful (and contrast set depending on the scene) or Flaat_10.
- Flat picture styles are mostly useful in scenes with lots of contrast and lots of details. Having large gradients in the frame may pose problems with banding in post when you try to add some contrast back to the image. Don’t be afraid to shoot flat, just be ready for the consequences. Shooting flat when the scene is low contrast makes no sense as you will compress an already limited range even further. Avoid that.
- Flat picture styles can be thought of as a production format. But it may still be acceptable to shoot flat if your goal is low contrast look. Even then, it is easier to take contrast out in post than the opposite.
- Don’t be afraid to play with the Color Tone parameter of your picture style. This tweaks skin tones from Magenta/Pink to Yellow and can help when skin does not look right.
- Mixing picture styles for a project can work, but may also introduce matching problems in post. Stylistic differences between sequences is a good reason to use a different picture style. Difference in scene contrast is a good reason for that only if the two picture styles match in color.
- Picture style color is at least as important as dynamic range, if not more. There are so many videos with crappy grading out there that it is not even funny (hello, Magic Bullet Looks). Test picture styles before committing to them. Canon’s Digital Photo Professional can apply any picture style to a RAW image, so it is easy to see for yourself. Don’t just use a style because it is popular, or because you’ve seen a nice video shot with it. Test. Know your image before you’ve even taken it out of the memory card.
Very nice and clear article. I have 3 custom profiles
1. neutral contrast -4 (prolost settings) when Magic Lantern shows zebras in both highlights and shadows I switch to
2. cinestyle. Didn’t use is very often so not much experience but I will switch to flaat 10
3. flaat12 for even more dr. It never came to this up till now. So never used it.
Bart