I started writing this as an explanation of why I often choose not to use log for low light. But instead it’s ended up as an experiment you can try for yourself if you have a waveform monitor that will hopefully allow you to better understand the differences between log and standard gamma. Get a waveform display hooked up to your log camera and try this for yourself.
S-Log and other log gammas are wonderful things, but they are not the be-all and end-all of video gammas. They are designed for one specific purpose and that is to give cameras using conventional YCbCr or RGB recording methods the ability to record the greatest possible dynamic range with a limited amount of data, as a result there are some compromises made when using log. Unlike conventional gammas with a knee or gammas such as hypergammas and cinegammas, log gammas do not normally have any highlight roll off, but do have a shadow roll-off. Once you get above middle grey log gammas normally record every stop with almost exactly the same amount of data, right up to the clipping point where they hard clip. Below middle grey there is a roll off of data per stop as you go down towards the black clip point (as there is naturally less information in the shadows this is expected). So in many respects log gammas are almost the reverse of standard gammas. The highlight roll off that you may believe that you see with log is often just the natural way that real world highlights roll off anyway, after all there isn’t an infinite amount of light floating around (thank goodness). Or that apparent roll off is simply a display or LUT limitation.
An experiment for you to try.
If you have a waveform display and a grey scale chart you can actually see this behaviour. If you don’t have a chart use the grey scale posted here full screen on your computer monitor. Start with a conventional gamma, preferably REC-709. Point the camera at the chart and gradually open up the aperture. With normal gammas as you open the aperture you will see the steps between each grey bar open up and the steps spread apart until you reach the knee point, typically at 90% (assuming the knee is ON which is the default for most cameras). Once you hit the knee all those steps rapidly squash back together again.
What you are seeing on the waveform is conventional gamma behaviour where for each stop you go up in exposure you almost double the amount of data recorded, thus capturing the real world very accurately (although only within a limited range). Once you hit the knee everything is compressed together to increase the dynamic range using only a very small recording range, leaving the shadows and all important mid range well recorded. It’s this highlight compression that gives video the “video look”, washed out highlights with no contrast that look electronic.
If you repeat the same exercise with a hypergamma or cinegamma once again in the lower and mid range you will see the steps stretch apart on the waveform as you increase the exposure. But once you get to about 65-70% they stop stretching apart and now start to squeeze together. This is the highlight roll off of the hypergamma/cinegamma doing it’s thing. Once again compressing the highlights to get a greater dynamic range but doing this in a progressive gradual manner that tends to look much nicer than the hard knee. Even though this does look better than 709 + Knee in the vast majority of cases, we are still compressing the highlights, still throwing away a lot of data or highlight picture information that can never be recovered in post production no matter what you do.
Conventional video = Protect Your Highlights.
So in the conventional video world we are taught as cameramen to “protect the highlights”. Never overexpose because it looks bad and even grading won’t help a lot. If anything we will often err on the side of caution and expose a little low to avoid highlight issues. If you are using a Hypergamma or Cinegamma you really need to be careful with skin tones to keep them below that 65-70% beginning of the highlight roll off.
Now repeat the same experiment with Slog2 or S-log3. S-log2 is best for the experiment as it shows what is going on most clearly. Before you do it though mark middle grey on your waveform display with a piece of tape or similar. Middle grey for S-log2 is 32% (41% for S-log3).
Now open up the aperture and watch those steps between the grey scale bars. Below middle grey, as with the standard gammas you will see the gap between each bar open up. But take careful note of what happens above middle grey. Once you get above middle grey and all the way to the clip point the gap between each step remains the same.
So what’s happening now?
Well this is the S-log curve recording each stop above middle grey with the same amount of data. In addition there is NO highlight roll off. Even the very brightest step just below clipping will be same size as the one just above middle grey. In practice what this means is that it doesn’t make a great deal of difference where you expose for example skin tones, provided they are above middle grey and below clipping. After grading it will look more or less the same. In addition it means that that very brightest stop contains a lot of great, useable picture information. Compare that to Rec-709 or the Cinegammas/Hypergammas where the brightest stops are all squashed together and contain almost no contrast or picture information.
Now add in to the equation what is going on in the shadows. Log has less data in the shadows than standard gammas because you are recording a greater overall dynamic range, so each stop is recorded with overall less data.
Standard Gammas = More shadow data per stop, much less highlight data = Need to protect highlights.
Log= Less shadow data per stop, much more highlight data = Need to protect shadows.
Hopefully now you can see that with S-log we need to flip the way we shoot from protecting highlights to protecting shadows. When you shoot with conventional gammas most people expose so the mid range is OK, then take a look at the highlights to make sure they are not too bright and largely ignore whats going on in the shadows. With Log you need to do the opposite. Expose the mid range and then check the shadows to make sure they are not too dark. You can ignore the highlights.
Yes, thats’ right, when shooting log: IGNORE the highlights!
For a start you monitor or viewfinder isn’t going to be able to accurately reproduce the highlights as bright as they are . So typically they will look a lot more over exposed than they really are. In addition there is a ton of data in those highlights that you will be able to extract in the grade. But most importantly if you do underexpose your mid range will suffer, it will get noisy and your shadows will look terrible because there will be no data to work with.
When I shoot with log I always over expose by at least 1 stop above the manufacturer recommended levels. If you are using S-log2 or S-log3 that can be achieved by setting zebras to 70% and then checking that you are JUST starting to see zebras on something white in your shot such as a white shirt or piece of paper. If your camera has CineEI use an EI that is half of the cameras native ISO (I use 1000 or 800 EI for my FS7 or F5).
I hope these experiments with a grey scale and waveform help you understand what is going on with you gamma curves. One thing I will add is that while controlled over exposure is beneficial it can lead to some issues with grading. That’s because most LUT’s are designed for “correct” exposure so will typically look over exposed. Another issue is that if you simply reduce the gain level in post to compensate than the graded footage looks flat and washed out. This is because you are applying a linear correction to log footage. Fo a long tome I struggled to get pleasing results from over exposed log footage. The secret is to either use LUT’s that are offset to compensate for over exposure or to de-log the footage prior to grading using an S-Curve. I’ll cover both of these in a later article.
What about shooting in low light?
OK, now lets imagine we are shooting a dark or low light scene. It’s dark enough that even if we open the aperture all the way the brightest parts of the scene (ignoring things like street lights) do not reach clipping (92% with S-Log3 or 109% with S-Log2). This means two things. 1: The scene has a dynamic range less than 14 stops and 2: We are not utilising all of the recording data available to us. We are wasting data.
Log exposed so that the scene fills the entire curve puts around 100 code values (or luma shades) per stop above middle grey for S-log2 and 75 code values for S-Log3 with a 10 bit codec. If your codec is only 8 bit then that becomes 25 for S-log2 and 19 code values for S-Log3. And that’s ONLY if you are recording a signal that fills the full range from black clip to white clip.
3 stops below middle grey there is very little data, about thirty 10 bit code values for S-Log2 and about 45 for S-log3. Once again if the codec is 8 bit you have much less, about 7 for S-Log2 and about 11 for S-log2. As a result the darker parts of your recorded scene will be recorded with very little data and very few shades. This impacts how much you can grade the image in post as there is very little picture information in the darker parts of the shot and noise tends to look quite coarse as it is only recorded with a limited number of steps or levels (this is particularly true of 8 bit codecs and an area where 8 bit recordings can be problematic).
So what happens if we use a standard gamma curve?
Lets say we now shoot the same scene with a standard gamma curve, perhaps REC-709. One point to note with Sony cameras like the FS5, FS7, F5/F55 etc is that the standard gammas normally have a native ISO one to two stops lower than S-Log. That’s because the standard gammas ignore the darkest couple of stops that are recorded when in log. After all there is very little really useable picture information down there in all the noise.
Now our limited dynamic range scene will be filling much more of our recording range. So straight away we have more data per stop because we are utilising a bigger portion of the recording range. In addition because our recorded levels will be higher in our recording range there will be more data per stop, typically double the data especially in the darker parts of the recorded image. This means than any noise is recorded more accurately which results in smoother looking noise. It also means there is more data available for any post production manipulation.
But what about those dark scenes with problem highlights such as street lights?
This an area where Cinegammas or Hypergammas are very useful. The problem highlights like strret lights normally only make up a very small part of your your overall scene. So unless you are shooting for HDR display it’s a huge waste to use S-log just to bring some highlights into range as you make big compromises to the rest of the image and you’ll never be able to show them accurately in the finished image anyway as they will exceed the dynamic range of the TV display. Instead for these situations a Hypergamma or Cinegamma works well because below about 70% exposure Hypergammas and cinegammas are very similar to Rec-709 so you will have lots of data in the shadows and mid range where you really need it. The highlights will be up in the highlight roll off area where the data levels or number of recorded shades are rolled off. So the highlights still get recorded, perhaps without clipping, but you are only giving away a small amount of data to do this. The highlights possibly won’t look quite as nice as if recorded with log, but they are typically only a small part of the scene and the rest of the scene especially the shadows and mid tones will end up looking much better as the noise will be smoother and there will be more data in that all important mid-range.