English - Espaņol
Sharpness - Bokeh
for Canon DSLR cameras
Foto tutorial (English)
Foto tutorial (Espaņol)
Equipment recommendations US-ES
Flaat for Canon
Flaat for Nikon
Flaat for the BMC
Flaat for NEX-5N
Old Picture Style Tests
550D video lineskip
APS-C vs Full Frame
Badly assembled lenses and image quality
Lens mount compatibility chart
ISO on different cameras
High ISO on the 5D3
DIY: DR test chart
RGBWK Bayer sensors
Notes on DoF-FoV
Notes on crop-DoF-FoV
Custom Cropmarks for Magic Lantern on the Canon 550D
How many megapixels do I want?
How many megapixels can I see?
Quick Monitor Calibration Chart
This is old information now: after all this, I have created my own suite of Flaat picture styles.
Below you'll find my original post, plus two important updates.
Here's what I'm going to do...
I chose one RAW picture, from the ones I took at a wedding I attended some time ago. It's slightly underexposed, as some people say you should do when shooting video with a DSLR (as clipped highlights will look really nasty and are impossible to recover), has some nice skin and some deep shadows.
The camera used was a Canon EOS 550D, and the lens is a Leitz Elmarit-R 35mm f/2.8.
I "developed" this RAW image with Canon's Digital Photo Professional, using different settings and picture styles. I then cropped an interesting area (with some nice skin and some deep shadows), and resampled it to 33% using a bicubic algorithm (the video mode is roughly a 33% resample too, but it uses a much nastier nearest-neighbor-like algorithm).
The table below shows the results from developing this picture with different picture profiles: Portrait, Neutral, and the recently released Technicolor CineStyle.
For each picture style, there are three samples: one with standard contrast ("contrasty": 0 on Portrait and Neutral, silly +4 on Technicolor), one with reduced contrast ("flat": -4 on Portrait and Neutral, non-advised 0 on Technicolor), and one with very, very low contrast ("ultra flat": -4 plus curve with lifted shadows for Portrait and Neutral, -4 for Technicolor).
The rest of the picture profile settings were set at my usual values: sharpness=2 (many people prefer 0, but I normally use 2), saturation=-2, tone=0. No further modifications or adjustments have been applied to the images.
The Technicolor CineStyle profile is really powerful: it can bring up shadows that even the aggressive curves on "extra flat" Portrait and Neutral profiles are crushing to black. So, if you need extended dynamic range, Technicolor CineStyle is an amazing tool. But it kills your skin tones, and delivers grey skin tones and clay-looking people. You can recover nice skin tones in post, in theory, but my skills are very limited, and I never could get nice final images from superflat footage.
And this is what the Technicolor magic becomes even more revolutionary: the Technicolor CineStyle profile also comes with a LUT (look-up-table) that gets your recorded image and converts it back to something similar to a reasonably flat, neutral picture style. So now I can get beautiful skin tones in my final images, but the recorded files have all the dynamic range that the camera can offer. All I have to do is import my footage to the editor, use the free Red Giant Magic Bullet LUT Buddy to apply the Technicolor LUT to the footage, and do my color grading by adding filters below the LUT layer. IT'S NEARLY PERFECT (the only drawback being that the codec's 8 bits of color now have to be spread over a larger dynamic range, so there will be more noise in the final images, and codec and sharpening artifacts will be more visible).
But wait: there's more: I opened the .lut file on a text editor, and it's just three S curves, for RGB, and they're all the same! So I got the Technicolor Cinestyle ultra flat image, applied some curves to it, and got basically the same result as with the LUT! This means that this ultra flat picture style is so nice that you can get back your pretty colors with a simple luma S curve. Amazing.
MY HONEYMOON WITH THE TECHNICOLOR PICTURE STYLE IS OVER
I ran some more tests, and found the following: even with Technicolor CineStyle in my bag of tricks, I get my best results if I try to get the footage in-camera as close as possible to what I want my final images to look like
* If I know my final footage is going to look like "neutral with contrast=-2", I get a lot less noise and in general a better image if I shoot like that in the first place, instead of shooting with Technicolor CineStyle and then applying the LUT (or equivalent curve); but at least on this case it's just a matter of applying a LUT (or equivalent curve).
* If my final images are going to have less contrast than that, I'll have to fight with the color correction tools in post, and it's not an easy task: it's much easier than it was with previous ultra flat picture styles, but I still struggle to get nice images and sometimes still end up with greenish/greyish skin tones or having to make compromises in order to avoid that; the LUT (or equivalent curve) is a great starting point, and makes the task much easier, but it's still not foolproof: it requires skills that I just don't have (yet) (hopefully).
* If my final images are going to have more contrast than that, all from the previous point applies, plus noise gets really, really bad.
I always leave some room to be able to tweak the final images in post, but leaving too much and then having to push and pull a whole lot creates an awful lot of noise in my final images (my guess: codec artifacts being blown up); the more I mess with my footage, trying to take it to a place where it doesn't want to go, the nastier it looks.
I totally understand this are just reflections of my lack of skills with the color correction tools, but that's me and I can't change that but very, very slowly.
Smarter and more skilled people won't have this problem; and for people with different workflows, it will be the everyday option (people working with clients, or more in general when the people shooting on set don't have full control over how the final images will look like).
Even with all this, I'm not completely back to square one: now I have an "extended DR" mode in my camera, which I can use if needed; my most common picture style, though, will go on being "portrait with contrast=-4"; for some extra DR, I'll go on using "portrait with contrast=-4 and curve pushing shadows upwards"; and then when I need as much DR as I can get, conserving not only shadows but highlights too, I'll go for Technicolor CineStyle.
UPDATE 2: switching between CineStyle and Portrait picture styles is problematic. I've ended up using only Technicolor CineStyle, but with 3 different settings for Contrast and Saturaion (as explained at the end of this page).
After applying the LUT, the ultra flat Technicolor CineStyle footage (shot with contrast=-4) becomes very close to what you get from the camera if you shoot with Neutral picture style and contrast=-2:
The problem with using this picture style is that, if you don't need all that DR (because what's in front of your camera just doesn't require it, or because you know you'll end up crushing some shadows to black in your final images) then you're spreading the 8 bits of the H.264 codec too thin. These are the histograms from the previous two images; the first one represents "trying to get in camera something close to your final image", the second one is "shoot with the Technicolor picture style and then get your final image as you want it in post":
CineStyle is a picture profile that expands the dynamic range of the camera. But the image still has to be stored using an 8-bit codec, which means it has to assign a round number, from 0 to 255, to the value of each channel. Later, you can apply the LUT to quickly get a non-flat image; what this does is that it expands those values (0-255) to a new set of values, again (0-255), but with less concentration in the middle and more concentration in the ends. And this way you get something that is pretty close to what you would have got by using neutral with contrast = -2 in the first place.
For example, what was X with neutral-2 becomes Y with CineStyle, then becomes Z when applying the LUT:
This allows for much greater flexibility in post (because you can use a curve that's different from the LUT), but creates image quality issues because you're spreading your (0-255) values too thin.
Consider the picture I used for my test. Shot with neutral and contrast = -2, it has a nice histogram with no clipping in the highlights or the shadows. Which means I don't need the extra DR provided by CineStyle. So, if I shoot with CineStyle, all I get is a more concentrated histogram, with everyghing moved towards the center and flat zeroes both on (0-30) and (225-255). So, in effect, instead of using (0-255), I'm using (31-224): it is nearly as if I had moved from an 8-bit codec to a 7-bit codec. When I expand that histogram back with the LUT, I get my original (0-255) range from the neutral-2 setting again, but now the histogram has holes: there is nothing with a value of 53, or a value of 57.
Now, if I had a setting with sunlight and some deep shadows, neutral-2 would lead to a lot of 255 and 0 values, whereas CineStyle will deliver a nicer gradation in highlights and shadows. But if I don't need that extra DR, I find it better not to use it.
In summary, after some experimentation, I use the following four picture styles, depending on how much dynamic range I need:
And I try to get as nice an image as I can in-camera.
There are other interesting opinions in this thread
As I found out, switching between Portrait and CineStyle is problematic, as they are so different (one of the main reasons is that CineStyle doesn't use the 0-16 luma range, so I have to reset my zebras every time I change picture style... which means I ended up avoiding changes of picture styles, or disabling the zebras).
My solution has been to use 3 different picture styles, all based on CineStyle, but with different values for contrast and saturation:
(those are my reference saturation values, but sometimes I lower them a couple of notches if I have one single channel clipping, most often because of some bright red clothing)
These histograms are what I get after processing the same image with Portrait -4 -4 and with these CineStyle settings:
And these are the results from those settings (in the Portrait example, I lifted luma to get comparable pre-grading results):