Sony FS700 Slog in 8bit comparison
The new firmware v3 for the FS700 gave us 4K RAW but it also gives the standard camera some new curves to record with, these are SLog2 and 800%REC709. Sonys reason for adding them is primarily to help with monitoring RAW recording on the LCD, the SLog2 adds a massive amount of detail in the highlights which gets around how to monitor the large dynamic range of RAW on the internal LCD.
Of course you can record with these curves with the stock FS700 and they offer some useful options. The biggest problem with them is that the output of the stock FS700 is only ever 8 bit, so you have 255 levels of brightness. Using curves that place more of those values in the highlights robs the shadows and mid tones. With a 10bit output this isn't so much of an issue - you have 4 times as many brightness values to manipulate. The F3 for example offers 10bit Slog output and you'd basically always record in this way because it makes best use of the range for all situations. 10bit was always considered the minimum for recording Log footage.
When i say manipulate i don't just mean grading but also simply turning the Slog footage into images that look correct. Slog typically looks milky and desaturated and most workflows start off by using a LUT to convert the image into what was really there for further manipulation. It's able to hold an enormous amount of subtle colour and brightness information which a standard curve would simply clip or not capture at all. So even the simple act of applying an Slog LUT to an 8 bit image can reveal noise and lack of detail in the shadows and mid tones.
What i've done here as a quick and dirty test is to record the ranges of some profiles using a transmission step wedge from stouffer. I'm not a massive one for testing and i've hopefully been as accurate as i can within reason. The general feeling is tha the SLog curve increases the dynamic range of the camera, so i wondered if this was the case. The set up is very simple. The step wedge has 41 values in 1/3 stop increments for a total dynamic range of 13.5 Stops. The first test was just to see the curve and range of Slog compared to a cine 4 profile. The wedge is indirectly back lit (to avoid hotspots which can happen if directly lit, especially with an LED panel) and the camera is pointed at the other side in darkness. Then i'd adjust exposure on the lens so that the brightest step just clips. This is the baseline and then let the shadows fall wherever they do.
When judging dynamic range it can be quite tricky because the issue is' at what point is it too noisy to see a difference from one step to the other?'. There's a vast difference between usable range and technical-i-can-just-about-see-it range. So the first comparison is a simple test between SLog and my Cine 4 profile. You can see quite clearly that it's a tough call when to call the bottom of the range, i tend to be conservative. The noise takes over the signal.
The above diagram shows from the brightest down, the green vertical lines mark stops, so there are at least 10 stops of real range in there, then it gets mushy. There's probably another stop in there for a total of 11 stops. The blacks of the SLog curve seem quite elevated (they don't reach zero, black never seems to get to black) and the Black Level profile setting has no effect. This seems a shame because we're losing vital range within our 8 bit container. But, as we'll come to later, the Black Gamma does have an effect and this can be quite important.
I'd also like to do some tests using noise reduction on these files, to see if temporal reduction helps and cleans the shadows, so i'll post those results when i can get to it.
This is the Cine 4 Profile that i used a lot, based on Alistar Chapmans profile, it uses a Black Gamma of Low set to +7 and you can see the effect o this on the bottom part of the curve. Again i count 10+ stops of real range plus perhaps, maybe, just about, in some circumstances another stop. It does appear that SLog is slightly better, but not that much.
The biggest observation is the difference in the curves, the 384 code value (Resolves charts only show 1024 values) for SLog appears 6 stops in compared to 5 stops in cine 4.
I really hasten to add, i'm being conservative here, i know that others have claimed 12+ stops but i think that's pushing it a bit. We know that the RAW output of the camera is 12 bits maximum so i feel it's unlikely anyone has ever really seen more than 12 stops.
However we can't use this to compare exposure directly because i tweaked it so that i exposed the top of the curve the same. This was to get an idea of dynamic range. And i would say SLog has slightly more, but it's quite clear that the detail is in the top of the range. The lower part is noisy and mushy. You can see that Cine4 distributes more values to the midtones whilst squeezing the top and bottom.
I'm also not entirely convinced this matches the SLog curves of the F3 or other Sony cameras, it seems actually a bit less 'loggy' which means it's a bit better for shooting with. I cannot compare directly though, i could be wrong here. One aspect that adds weight to this idea is that if i use the standard Sony LUT in Resolve then the 8 bit Slog files don't look right, they come over very overexposed and too contrasty and you need to gain them down in resolve, that just makes me think that the 'log' from the FS700 isn't quite what the LUT is expecting...
Secondly i then pegged 50% exposure (on my SmallHD) in the middle of the wedge, kept that fixed and rotated through the Gamma Curves to compare them all side by side with the same exposure. So we can see exposure over and under. Because exposure is fixed then it's easier to compare curves. However after i did this i realised that ISO is different between SLog and the Cine curves so these results can only be viewed relative to each other. I will try to compare cine4 and Slog by exposing for 18% grey and check the curves there - at a later stage.
We can see that SLog works well in capturing high lit scenes, i think it actually does better than it ought too. I'd love to compare an F3 and 700 side by side because i feel the 700 is less like a log curve, out of the camera it's a bit closer to reality (not as milky as other logs look). As for the others, i'm not sure SLog really gives more range, it might be scene dependent but it's all about the shadows and how mushy you find acceptable. I think because it defines the highlights better yet still has the blacks then the 'feeling' is of a larger range. Certainly profiles like REC709 should be avoided unless you're going for an old video camera look. The 800% REC709 version is simply for monitoring. It shows the top range, has the contrast and saturation of REC709 and shows the shadows but almost no detail in the shadows. You can monitor the highlights and the blackest black but i'd never record with it.
I'm very impressed with SLog, i'd written it off for special occasions but after shooting with it a bit and grading it has a nice distribution of values to work with. In daylight at least. In low light i suspect it's a different story. However the standard cine curves are still the best for most uses, especially if you don't have a post workflow in place.
We've now added a follow on article from this which tests the range in a different way: SLog2 Redux