UPDATE: Following much debate and discussion in the comments section and on my Facebook feed I think one thing that has become clear is an important factor in this subject is the required end contrast. If you take S-Log3 which has a raised shadow range and shoot with it in low light you will gain a low contrast image. If you choose to keep the image low contrast then there is no accentuation of the recorded noise in post and this can bring an acceptable and useable result. However if you need to grade the S-log3 to gain the same contrast as a dedicated high contrast gamma such as 709, then the lack of recorded data can make the image become coarser than it would be if recorded by a narrow range gamma. Furthermore many other factors come into play such as how noisy the camera is, the codec used, bit depth etc. So at the end of the day my recommendation is to not assume log will be better, but to test both log and standard gammas in similar conditions to those you will be shooting in.
Log gamma curves are designed for one thing and one thing only, to extend the dynamic range that can be recorded. In order to be able to record that greater dynamic range all kinds of compromises are being made.
Lets look at a few facts.
Amount of picture information: The amount of picture information that you can record, i.e. the amount of image samples, shades or data points is not determined by the gamma curve. It is determined by the recording format or recording codec. For example a 10 bit codec can store up to 1023 shades or code values while an 8 bit codec can record up to 255 shades or code values (in practice this is a maximum of 235 shades as 16 are used for sync). It doesn’t matter which gamma curve you use, the 10 bit codec will contain more usable picture information than the 8 bit codec. The 10 bit picture will have over 1000 shades while the 8 bit one less than 255. For low light more “bits” is always going to be better than less as noise can be recorded more faithfully. If noise is recorded with only a few shades or code values it will look coarse and ugly compared to noise recorded with lots more levels which will look smoother.
Bottom line though is that no matter what gamma curve, the maximum amount of picture information is determined by the codec or recording format. It’s a bit of a myth that log gives you more data for post, it does not, it gives you a broader range.
Log extends the dynamic range: This is the one thing that log is best know for. Extending the dynamic range, but this does not mean we have more picture information, all it means is we have a broader range. So instead of say a 6 or 7 stop range we have a 14 stop range. That range increase is not just an increase in highlight range but also a corresponding increase in shadow range. A typical rec-709 camera can “see” about 3 or 4 stops below middle grey before the image is deemed to be too noisy and any shades or tone blend into one. An S-log2 or S-log3 camera can see about 8 stops below middle grey before there is nothing else to see but noise. However the lower 2 or 3 stops of this extended range really are very noisy and it’s questionable as to how useful they really are.
Imagine you are shooting a row of buildings (each building representing a few stops of dynamic range). Think of standard gammas as a standard 50mm lens. It will give you a great image but it won’t be very wide, you might only get one or two buildings into the shot, but you will have a ton of detail of those buildings.
Think of a wide dynamic range gamma such as S-log as a wide angle lens. It will give you a much wider image taking in several buildings and assuming the lens is of similar quality to the 50mm lens, the captured pictures will appear to be of similar quality. But although you have a wider view the level of detail for each building will be reduced. You have a wider range, but each individual building has less detail
But what if in your final scene you are only going to show one or two buildings and they need to fill the frame? If you shoot with the wide lens you will need to blow the image up in post to the show just the buildings you want. Blowing an image up like this results in a lower quality image. The standard lens image however won’t need to be blown up, so it will look better. Log is just the same. While you do start off with a wider range (which may indeed be highly beneficial) each element or range of shades within that range has less data than if we had shot with a narrower gamma.
Using log in low light is the equivalent of using a wide angle lens to shoot a row of buildings where you can actually only see a few of the buildings, the others being invisible and then blowing up that image to fill the frame. The reality is you would be better off using the standard lens and filing the frame with the few visible building, thus saving the need to blow up the image.
S-Log2/3 has a higher base ISO: On a Sony camera this higher ISO value is actually very miss-leading because the camera isn’t actually any more sensitive in log. The camera is still at 0dB gain, even though it is being rated at a higher ISO. The higher ISO rating is there to offset an external light meter to give you the darker recording levels normally used for log. Remember a white card is recorded at 90% with standard gammas, but only 60% with log. When you change the ISO setting upwards on a light meter it will tell you to close down the aperture on the camera, that then results in the correct darker log exposure.
S-Log3 may appear at first brighter than standard gammas when you switch to it. This is because it raises the very bottom of the log curve and puts more data into the shadows. But the brighter parts of the image will be no brighter than with a camera with standard gammas at 0db gain. This extra shadow data may be beneficial for some low light situations, so if you are going to use log in low light S-Log3 is superior to S-Log2.
If you can’t get the correct exposure with log, don’t use it! Basically if you can’t get the correct exposure without adding gain or increasing the ISO don’t use log. If you can’t get your midrange up where it’s supposed to be then you are wasting data. You are not filling your codec or recording format so a lot of data available for picture information is being wasted. Also consider that because each stop is recorded with less data with log not only is the picture information a bit coarser but so too is any noise. If you really are struggling for light, your image is likely to be a bit dark and thus have a lot of noisy and coarse noise is not nice. Log has very little data allocated to the shadows in order to free up data for the highlights because one of the key features of log is the excellent way it handles highlights as a result an under exposed log image is going to lack even more data. So never under expose log.
Think of log as the opposite of standard gammas. With standard gammas you always try never to over expose and often being very slightly under exposed is good. But log must never be under exposed, there is not enough data in the shadows to cope with under exposure. Meanwhile log has more data in the highlights, so is very happy to be a little over exposed.
My rule of thumb is quite simple. If I can’t fully expose log at the base sensitivity I don’t use it. I will drop down to a cinegamma or hypergamma. If I can’t correctly expose the hypergamma or cinegamma then I drop down to standard gamma, rec-709.