 |
English - Español |
recommendations |
Picture Styles |
Sharpness - Bokeh |
|
HOME
Foto tutorial (English) Foto tutorial (Español) Equipment recommendations US-ES Flaat Picture Styles Old Picture Style Tests LENS TESTS For stills For video For video (lineskip) Conclusions ND filters Other filters BOKEH TESTS Summary Alternative scenarios Badly assembled lenses and image quality Lens mount compatibility chart ISO on different cameras DIY: DR test chart RGBWK Bayer sensors DoF/FoV CALCULATOR 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 MY VIDEOS Random rants Search engine |
In this post I will present you the Similaar suite of Flaat picture styles for Canon DSLRs. You can download them here. You'll be the judge, but my aim when creating them was: * Less noise in the final images than Technicolor, by using more efficiently the codec's color space (Technicolor uses just a portion of it) * Easier to grade, leading to nicer skin tones in final images (even with my limited grading skills) than Marvels Cine and other similar options * Only on Flaat_3 and Flaat_4, more recorded DR (less part of the sensor's DR being clipped in the shadows; this comes at cost of noise) There are four variants: Flaat_1 is for low-DR situations, and for people that are used to shooting Neutral -4 Flaat_2 would be my workhorse option; it's the closest one to CineStile or Marvels Cine Flaat_3 is the one I’d use if I needed some extra DR; but not for everything: it's more noisy Flaat_4 is for very excepcional cases where I need as much DR as I can get; very noisy The general advice for clean images would be: use the narrowest Flaat picture style that records enough DR for your shot. If Flaat_1 doesn't give you clipped highlights or shadows, that route will lead to less noise and smoother gradients than using Flaat_3. Below you'll find a detailed explanation of what they do, and lots of tests of how they treat the images. There's a video test here, and some real world stills tests here. There are 3 reasons these were born: * I was unhappy with all previous picture style options for Canon DSLRs: standard, neutral, portrait, etc, just don't seem to cut it; Technicolor CineStyle seems to use the available color space inefficiently and delivers more noise than other options; Marvels Cine requires a lot of work in post and is not easy enough to grade for my very limited skills (I end up with clay looking people nearly all the time). You can see my old post about picture styles here. * I had some free time (the shooting for my main project right now got postponed again: this is the third time, but it's a project with friends and we won't shoot until we can all be on set). * I was inspired by AbelCine's great blog posts about the different gamma curves available on the Sony F3, with and without the S-Log upgrade (here, but also here, here, here and here). Quick summary: this is the waveform of a DR test chart with a final image gamma (F3 S-LOG plus P2 LUT: HG8009G40)  and this is the waveform of same chart with S-LOG (ungraded):  Wide and flat, much nicer than what I can get with my Canon DSLR. So, given that, and being the constructive nerd I am, this is what I did: DR test chart: * I used my TV to create a relatively accurate dynamic range test chart; it only covers 11 stops of latitude, but for my T2i (550D) it should be nearly enough * With my Canon T2i (550D), I took 7 stills of said chart, at shutter speeds 1/3 stops apart from each other, in RAW mode (so I can easily develop them using different picture styles and settings) * I resampled those 5184x3456 stills to 1920x1280 (in order to approximately simulate the noise in video mode: first resample to 5184x1152 using nearest neighbor, then resample to 1920x1280 using bicubic) * I merged those developed and downsampled images in a very specific way, so I get a confirmation that the chart is relatively accurate (it seems to have some minor problems in the brightest sample, but the rest is fine), and a synthetic shot of a virtual chart that works in 1/3 stops and has two stops more dynamic range than my TV can provide * I inserted the resulting image on a 1080p timeline, and got the waveform for this virtual 1/3 stops chart under each picture style tested Skin tones tests: * I looked for an old RAW still which I could use for assessing skin tones (I chose one with difficult lighting but properly white balanced), and developed it using the same picture styles and settings * I resampled said still to 1920x1280 (same simulated video mode method) and cropped the relevant area (that's the ungraded skin test) * I brought those skin crops to the editor, and graded them using a simple luma or RGB curve and a saturation adjustment, just to see what issues arise when you bring some contrast back to the image (that's the "graded skin test", for which I also supply histogram and vectorscope, and the curve that I used for grading) With these tools, I could see what my usual picture styles do to the image: I could see that Technicolor does not digg much farther into the highlights than any other picture style, and it does bring back some more information from the shadows, but again it's just a tiny bit; and I could see that Marvels Cine comes with contrast=-3 for a reason: if you change it to -4, you get a weird gamma curve. Then, with those same tools, through a trial and error process, I created the Similaar suite of picture styles: Flaat 1 through 4. They are all modified versions of the Portrait picture style (not surprisingly, I find it delivers the best skin tones), with contrast = -4 and curves that try to do something similar to what S-Log does on the Sony F3. What's different in versions 1 through 4 is how aggressive those curves are (how far they digg into the shadows): Flaat_1 is just the vanilla portrait picture style, with a curve to smooth the gamma response Flaat_2 brings up the shadows a bit but it's not too far from Flaat_1 Flaat_3 pushes the shadows up some more (already quite aggressive: use with care) Flaat_4 includes a hugely aggressive curve that tries to maximize the captured dynamic range These picture styles use the available color space in an efficient manner, are easy to grade, and deliver nice skin tones. Even with my very limited grading skills. Just use a luma curve (or RGB curve and desaturation if your software doesn't have luma curves) as the first step of your grade, and move on from there. Ease as pie. Even I can do it. Most of them deliver more latitude than any other option I've seen (including Technicolor CineStyle and Marvels Cine). But they do that by bringing up the shadows: you get to see what's there, but it will be noisy. In the case of Flaat_4, it brings over one stop back from previously clipped shadows, but even at ISO 100 the heavily lifted shadows may display quite a lot of noise: you'll have to push them back down in post, but at least they're not clipped (like they are with every other option), so you have more control on how you treat them. Also, beware: when shooting with the most extreme Flaat options, the reduced color space and bitrate that's left for the midtones and highlights can create (in some scenarios) some banding and/or macroblocking (e.g. on bright skies and such). I'm carrying in my camera Flaat 1 through 3. The other one is too extreme and I only use it for very special situations. You'll find the detailed results at the botoom of this page, but before that this is a comparison of the gamma curves induced by different picture styles (as extracted from the virtual chart that works in 1/3 stops): Flaat 1 vs Flaat 2 vs Flaat 3 vs Flaat 4 
Flaat 1 vs Neutral with contrast=-4 
Flaat 1 vs Portrait with contrast=-4 
Flaat 2 vs Marvels Cine 
Flaat 2 vs Technicolor CineStyle 
Using Flaat may require you to make some changes in the way you shoot. Specifically, if you look at what you camera actually sees (viewfinder, histogram, waveform, false color, etc) in order to judge exposure, you'll have to adapt to the modified gamma curves. This means, for example, that with Flaat_3, skin tones will have to fall in IRE 65%-75% instead of 60%-70%. But don't worry: this doesn't require you to throw any more light at your subjects, and doesn't get them any closer to clipped highlights either. But if you see them at IRE 75% and dim your lighs so they fall at IRE 70%, your final images will have more noise and you'll think it's my fault; we don't want that, so remember: they're probably not overexposed, it's just a different gamma curve. The big problem is that the target IRE for skin tones is different for each Flaat 1-4. The solution for this would be to apply a preview LUT and expose normally, but these cameras won't let you do that (maybe your field monitor will). An alternative that resembles that is to use Flaat_1 while setting your lights and exposure (skin tones will fall in the usual area), then (if that's what you want) switch to a more aggressive Flaat picture style and start recording. If you try to do that by hand, you'll probably forget to change picture style more often than not. But Magic Lantern is your friend: it will make that change automatically for you (set the desired Flaat as your "Rec PicStyle" in ML). Problem solved. Thanks, Magic Lantern! You can DOWNLOAD the Similaar suite of Flaat picture styles for Canon DSLRs here. These picture styles are heavily optimized for my T2i. I think they'll work just as well on all the Canon DSLRs that share that same 18 Mpix sensor and Digic IV processor (7D, 60D, 600D-T3i, 550D-T2i), but I'm not so sure about the 5D2 or 1D4. If you have one of those, you can test yourself (and maybe keep me posted: similaar.feedback@gmail.com). And now, for the detailed results... These are the contenders: * Flaat 1 through 4: with contrast=-4 (and you shouldn't change that), saturation=-2 * Portrait, Neutral and Standard: with contrast=-4, saturation=-2 * Technicolor CineStyle: with contrast=-4, saturation=0 (as they distribute it) * Marvels Cine: with contrast=-3, saturation=-2 (as they distribute it) * Marvels Cine: with contrast=-4, saturation=-2 (as many people use it) In all cases, sharpness=2: I think shooting with sharpness=0 and then adding it in post creates more noise, because it makes the macro-blocking from the poor codec more visible, so I always add at least some sharpness in-camera. You can use the Flaat pictures styles with different values for saturation and sharpness if you wish, but I think deviating from contrast=-4 is a bad idea: the gamma curve will definitely not be linear anymore, it will have some funky shape that I'd rather avoid. Also, I think lowering saturation too much in camera will lead to increased noise, as the color space of the codec is not efficiently used, but I'm not too sure about this one yet. The "final look" reference will be Portrait with contrast=-1, saturation=-2. That looks great to me, but you have to nail everything in camera because it leaves very little room for tweaking in post. Below you'll find a ton of charts and tests. If you prefer real world tests, you may rather go here: Real World Tests With Flaat Picture Styles. I also shot some video tests (with skin tones, high-DR and low light scenarios), specially useful for looking at noise (and also for checking that my method of "take a still, develop it with different picture styles" is valid). Watch it here.
With Marvels Cine, Neutral_4, etc, I hardly ever get such nice results (usual problem: clay-looking people). I was only able to get this here because I hade the Portrait_1 image as reference. Without that, I inevitably get lost. Flaat avoids these issues most of the time: applying a luma curve (or RGB curve and desaturation if your software doesn't have luma curves) to the original footage to get the contrast back in the areas I want is usually a good enough starting point, and leads to nice results.
The issues you see in these histograms arise mainly because the graded image was encoded back to an 8-bit codec. Working in a deeper color space would avoid most of this jaggedness. This histograms are nevertheless useful to identify where problems may arise. Regarding the curves: for Portrait and Flaat, it's just a luma curve. For the others, it's an RGB curve and a saturation adjustment.
|
|
Similaar is a participant in the Amazon Services LLC Associates Program and the Amazon EU Associates Programme, two affiliate advertising programs designed to provide a means for sites to earn advertising fees by advertising and linking to amazon.com and amazon.es.