If shooting in low light don’t use Log!

One common question I am often asked is how to use Log or CineEI in low light situations. After all in the Sony CineEI mode or with a log camera such as the Alexa that use EI the actual recording gain is fixed, so often the recorded pictures are very dark. So what should you do?

My answer in many cases is simply not to use Log or EI. It’ really important to remember that the primary reason why log recording was developed was to make it possible to record a very large dynamic range using existing recording technologies. In order to do this lots of compromises are made. The main one being allocating less recording data to each stop of dynamic range. If you’re recording using 10 bit you typically have about 970 useable code values or shades. Use a gamma curve with 6 stop range and you have about 160 shades per stop, record using a log or other extended range gamma with a 14 stop range and you have just 70 shades per stop.

Lets think about a couple of different scenes for a moment. Scene one is a daytime scene that’s nice and bright with an 8 stop dynamic range. Scene 2 is a night time scene that is fairly dark and only has a 5 stop dynamic range. What happens if we shoot using log?

With the daytime scene things are pretty straight forward. You just expose as per the manufacturers recommendation. Assuming your camera is set to log and capable of a 14 stop dynamic range, if exposed at the base, recommended exposure levels you will be using a little under 60% of your available recording range, so a fair bit of data will be going to waste. But as the scene is bright you can always open up the aperture a bit and deliberately over expose by a stop or two so that now your using around 75% of the available data. You don’t normally want to over expose too much as grading becomes trickier, but at least when you have a bright scene you can open up the aperture to expand the recording range a bit to make better use of the data.

In low light however it get’s trickier as your aperture may already be wide open. In the second dark scene it’s quite possible that the 5 stops of light coming back from the scene aren’t bright enough to allow you to expose at the recommended levels. So instead of filling the zero to 5 stop recording range you might only be filling the zero to 4 stop range. Even if you can expose correctly with a 14 stop recording bucket like log you are only using a little over 35% of the available data and that’s a terrible waste. You won’t have much data to help you separate out any noise from the desired picture information in post production and because your recorded signal is small any compression noise will be relatively big in comparison to the picture information.

This is when it’s time to abandon log and go back to a conventional gamma curve. You don’t need log when the scene only has a limited dynamic range. If you use Rec-709, which has a 6 stop range (without any knee) instead of log, at the same ISO, then now instead of recording using only 35% of the available data you will be using almost 85% of the available data and that’s going to give you much more real picture information to work with in post production. You will get a much better end result by not using log.

If you need a bit more than a 6 stop range then there are plenty of other gamma curves to choose from that sit between 709 and log. Gamma curves such as Sony’s cinegammas or hypergammas which have dynamic ranges between 10 and 11 stops.

So remember, when you don’t need a 14 stop recording range, then consider using an alternate gamma curve. It’s not just in low light but also in other controlled lighting situations such as green screen work.

460x150_xdcam_150dpi If shooting in low light don't use Log!

40 thoughts on “If shooting in low light don’t use Log!”

  1. Alister, your posts so often help me to understand this stuff at a deeper level. Thank you so much for taking the time to do this! It’s much appreciated.

  2. What about burning in a LUT in low light.. or are they all fairly wide DR compared to REC 709.. and so still wasting data..

    And thanks for the continued informative postings .. I was getting worried for a while there !

  3. Yes, you could burn in a narrow range LUT such as 709(800). You could make a LUT specifically for low light applications.

  4. Great post!

    I have a question regarding the differences between CINE EI mode and CUSTOM MODE. What is the bit-depth of the SLOG-3 CINE EI image before applying the luts and recording? 10 bit? 12bit?
    And in CUSTOM mode, how do I match an image to the SLOG-3.cine color space, so I know that my primaries are in the same place. I have had some bad experiences, that required a bit of grading, so I’m curious if there are cinegammas or hypergammas that match SLOG-3.cine

    1. FS7 and F5/F55 are 10 bit if recording XAVC in either mode. The colorspace is completely different in Custom and CineEI. In custom it’s basically 709 colorspace, in Cine EI you can choose between S-Gamut, S-Gamut3 and S-Gamut3.cine.

  5. Had this conversation quite a few times now and I had already decided that Slog3 is not for low light. This article explains it very well.

    Like Flemming’s comment above, I get nervous about matching shots, if using Slog3 EI and then switching to custom. If you had a project you had shot 90% in Slog3 S-Gamut.cine and you had to get some low light stuff, would you still go into custom?

    1. Most likely I would go to custom. I’d rather have a small color missmatch than a noisy, grainy, hard to grade image. The average viewer is going to notice the noise and grain more than a small color shift. Generally in low light there isn’t going to be a lot of color anyway.

  6. I thought that S-Log3 was developed to increase the number of steps available in the shadows.
    The way it was explained in an article I read some time ago, full white uses close to all the steps. One stop down, you halve the steps available. 2 stops down, you halve it again, leaving 1/4 the steps. S-Log redistributes the steps so that each stop gets a fairly equal allocation, so it was explained in the article.
    There’s probably some of that redistribution going on with normal gamma curves of REC709, but much more modestly.
    What do you think about this?

    1. S-Log3 only uses 3.5IRE to 92IRE, so compared to most modern gammas that use 0 to 109IRE you already have a restricted data range, then you’re cramming 14 stops into that small range compared to 6 stops in an expanded range with 709. Compared to S-Log2, S-log 3 has a little more data in the shadows, but it’s still a fraction of what a 6 stop gamma like 709 would have.

      Normal gamma curves like 709 use an increasing amount of data for each brighter stop of exposure. This is a good match for the real world where each successive stop has twice the brightness range of the previous one. The first stop in 709 uses about the same amount of data as the first stop in log. But while 709 is going to quickly increase the amount of data in each stop, the increase is much slower with log and once you get to about the 5th stop the amount of data stops increasing.

  7. Hi Alister, it’s your friendly proof reader again! Second paragraph … 970 useable data bits … must have been too late at night, or was it that Raffles beer? It should have read 970 useable levels…. the mistake then runs on into the following paragraphs with things like 70 bits per stop. Not sure what 2 to the power of 70 is, but it is a big number! Once again, thanks for all the work, it is very, very useful, and to quote my late Dad, the man who never made a mistake never made anything. Bob.

    1. That para is correct and that’s the point of log. The total available recording range is 970 data bits (bit’s 0 to 970 for S-Log3). Log uses more or less the same amount of data for each stop whereas conventional gamma’s use almost twice as much data per additional stop.

  8. Hi Alister, I think we are talking about different things. To me, and anyone else with a computing background the term ‘data bit’ means binary data. A ‘bit’ is by definition one binary digit, either a zero or a one. A word length of 10 binary bits gives a range from 0 to 1023 in decimal, that is, 1024 different code values, of which 970 are used. I found your nomenclature somewhat confusing, although obviously others have not. I think my original use of the term ‘levels’ did not help me get my point across. My fault. Not sure if I have explained myself any better this time, but I hope you can see the point I am trying to get make. Bob.

  9. OK, yes point taken. Levels or shades would be more accurate. Problem is than many people with no concept of digital technology have no clue as to what a code value is and when you start to talk about having 970 levels then people question how can you have 970 levels in something that only goes from 0 to 109IRE.

    But yes code values is less confusing and the correct term.

  10. Hi Alistair,

    You confirm my experience concerning Slog3 in low light. But in practice, when you are in documentary shooting, with real life and very very little time to adjust camera parameters (I mean going into the menus) how do you switch from an outside shot in Slog3 to an inside scene in Custom mode? The FS7 takes a while just to switch from one mode to the other.

  11. Hi Alister, the first time I came across the term CV I thought it stood for ‘Component Video’ and rather foolishly decided that the value given afterwards must be the signal voltage in milli volts! It took me some time to realise it was a Code Value. You walk a very tricky and technical path with great aplomb. Bob.

  12. Well I don’t know for the Sony cameras. But for the Alexa what you say is false I think.
    – The 709 image is created from and already Log encoding, like an “s-curve” LUT.
    – Low light is often very high contrast due to light sources being in the frame.
    – If you have details of your scene low, close to noise level, you’ll have your “only” 70 shades per stop (although log also have slope) in Log, but even less in Rec 709 or any good looking curves, cause, if not completely crushed, you’ll be in the slope of the curve, where some stops only have 10 values or less.
    So to be short burning in a lut with crush details that you could otherwise record with a Log format.

    1. What you are saying would only be correct with a LUT applied externally to the camera. The internal processing in a video camera is done at a much greater bit depth than 10 bit. Don’t forget that the Alexa is capable of 12 bit log prores, so at the very minimum the log data must be 12 bit before encoding, so there is more data there. The Sony F5/F55/FS7 cameras process the LUT’d images directly from the sensor data. The LUT’s are not added on top of the log, the images are processed as log+LUT in one single mathematical process so there is no artificial 10 bit log data restriction. Alexa is almost certainly the same as don’t forget you can also have raw + LUT. Internally the image processing is at least 16 bit (the minimum needed for 14 stops of DR) so there will be a lot more data in the shadows available prior to the signal being passed to the codec. Baking in a LUT internally is exactly the same processing process as using any other alternative gamma curve.

  13. hey I love your videos and tips! I have now tested your different HG settings (dowloaded from web) and tested Doug Jensen’s Universal 1 setting, which is based on S-log3 plus some other adjustments he has made in Custom mode,( you have to buy access on vimeo in order to download the image profile). This Universal – 1 becomes the brightest setting I have managed to find. I have also tried out all of the STD with the different options available , all HG versions and S- log2/3 . But none will be as bright as Doug Jensen’s Universal 1 which is based on S- log3. It is perhaps not as much information in the material, but I do not understand witch other picture profile setting that allows the camera to be brighter.

    So Alister please explain if there is something I have not understood…..and pleas share the exact picture profile you use when its very dark outside.

    1. It’s brighter because it’s adding gain.

      Gamma = gain. Gamma is a gain curve.

      Nothing in the camera can make it more sensitive, only changing the sensor will make a camera more sensitive, so anything else is just a gain adjustment whether it’s gamma or gain. So, making the picture brighter through gamma adds noise just the same as adding gain.

      Raising the gamma gain will create a low contrast image with data spread thinly. This will often lead to banding and other artefacts in the image.

      S-Log3 cannot record beyond 92%, so any profile or settings based on S-Log3 will have a constricted recording range. For low light if I want low noise and a good picture I use a fast lens to get as much light as possible on to the sensor and if that fails add light. Anything else will add grain and noise.

  14. Hi Allister
    Thank you so much for your great articles, they help to understand the sony fs7 so much better!
    I’ve done a series of testshootings where you can see that low light with SLOG3 (SLOG2 is a little better) is producing tons of noise.

    https://www.youtube.com/watch?v=RGBk7DPN3g8
    However in this testshootings you can see even noise on a late afternoon scene
    > Noise in SLOG3 (fence area), cineEI: https://youtu.be/RGBk7DPN3g8?t=364
    The aperture was wide open and i expected to have nice shadows although in the abandoned rabbit house one can see a lot of colored noise!
    So is SLOG even a problem on cloudy days with reduced light also? The thing is if you are shooting in 4K an want to enlarge a certain portion of the screen it’s getting really noisy in this cases right?
    Thanks Andy

    1. S-Log3 is not producing more noise, the signal to noise ratio of both S-Log2 and S-Log3 is the same. This is a very common missconception based on past experience with standard gammas where what you record is what the end viewer see. With log you have to stop thinking this way and understand that the footage is designed to be graded. After grading if the brightness levels are the same the noise will also be the same.

      See: https://www.xdcam-user.com/2016/07/understanding-the-all-important-signal-to-noise-ratio/

      AND

      https://www.xdcam-user.com/2014/03/understanding-sonys-slog3-it-isnt-really-noisy/

      Log is designed for one thing, a increased dynamic range in a limited recording space. You can’t fit more into a traditional recording space without other compromises.

  15. Thanks so much for posting this, and clarifying. I am currently challenged with a shoot that will be mostly all low light scenes: shooting at night on the sidewalk by a street, and in a night club. There will be practical lights, but it is not “lit” and the shots will all be mostly low light.

    We are using the Sony A7S II. The camera only outputs 4K at 8-bit, and will be recorded to pro-res in a Shogun Inferno.

    This footage will be graded, so log is an option, but i am concerned about noise.

    Q1: Does it make sense to stick to a cine gamma, to maximize the data per stop, rather than accommodate the dynamic range of Slog?

    Q2: Is there any reason to record ProRes HQ, over ProRes 4:2:2 if the camera is only sending 4:2:2?

    1. ProRes HQ has fewer compression artefacts than vanilla ProRes 422.

      I’d stick with a cinegamma as you will be seeing the final image while shooting, so will have an idea of what your final noise levels are.

  16. But can’t the SLOG 3 image be graded down to look like HG3 or HG7? Also, what if you are filming, let’s say a fire dancer, and though the majority of the scene is low light, wouldn’t that scene be considered to have a high dynamic range due to the bright fire and the pitch dark night? So wouldn’t SLOG 3 be better given a high dynamic range, low light scene? I’ve done some tests and HG 3 has the least noise, but is also very black/dark. SLOG 3 is the noisiest, but it is also the brightest and lets you see into the shadows. When I crush the blacks to bring down the noise, the image gets darker, but the noise dissipates too, so what you are left with is an image that has a dynamic range close to HG3, but maybe a little wider and also more nuanced colors than HG 3.

  17. Just did some low light tests comparing SLOG 3, HG 7, and HG 3. My conclusions are:
    1) The FS7 does not do well in low light (like the GH4!)
    2) All the clips had small amounts of noise that could not be graded out unless the image became unacceptably dark.
    3) Noise suppression did not help that much.
    4) Neat Video plug in was able to remove the noise from all the clips, but is a processor intensive process. Takes a long time!

    Bottom line: FS7 does not do well in low light; be prepared to deal with noise in post, so far, only Neat Video has completely solved the problem.

  18. Sorry, another conclusion was there is a very significant difference in color between SLOG 3 CINE EI and Custom Mode, even in low light. SGAMUT had deeper, richer colors while CUSTOM Matrix had bland, less saturated colors.

    1. S-Log is a trade-off. You are trying to fit a bigger range into the same amount of data as you use for normal shooting. Nothing comes for free, so when you squeeze that big range into a limited amount of data, each recorded stop will contain less picture information, fewer tonal values, fewer steps. As a result the noise will be coarser as it is recorded using less data. But it all depends on how much you can afford to crush the signal. Crush it down and the noise gets crushed too so it might go away depending on what you are shooting. Then there is what do you mean by low light? If what you are shooting is bright enough that you can afford to crush it then this may not really be low light but instead simply high contrast.

  19. Hi Alister
    When shooting a nighttime scene, say a street scene lit only by streetlights (max available exposure – everything wide open), but wanting to see as far as possible into the shadows, would a standard gamma be better? The dynamic range between objects in the light and those hiding in the shadows can be quite high… If you want to see into the shadows without burning out the rest of the scene… I am talking in an uncontrolled environment here for wildlife…

    1. It really depends on the individual scene and the light levels you have. A good middle ground tends to be the Hypergammas, but really the only way to know for sure is to test for the actual scene/scenario you are shooting. In other words, try it and see for yourself which works best for you and your workflow.

  20. We have an FS7 with raw adapter and a Odyssey 7q external recorder. What is the highest quality we can achieve in low light?
    Raw->Prores?
    Raw->DNGs?
    Is the raw signal in log colour space? It looks like it is.
    When I turn on Custom to get away from sLog the Raw is deactivated.
    Should I use the HDMI or SDI on the camera to send best signal to external recorder or should I now use internal ProresHQ recording?
    Or is raw dngs the best we can get, even for low light?

    1. Raw DNG is going to be the highest raw quality. Raw is in the full sensor colorspace and not tied to any particular gamut, however by default will be converted to SGamut by most grading software. But in low light 12bit linear may not be the best choice. In some cases the cameras internal 10 bit log will work better or perhaps a standard gamma or Hypergamma will work better.

  21. Hi Alister, Jumping on this rather old blog. Your statement about (not) using SLOG/CI in low light makes perfect sense. Also for green screen work. But what about shooting with the R7 in X-OCN? You get 16 bit on the one hand, but are stuck to SLOG on the other. When f.e. shooting green screen, or like me wolves at dawn, would you prefer the 16 bit X-OCN in SLOG or rather go for an 10 bit XAVC ‘standard’ gamma curve?

    1. X-OCN is linear, not log. The S-Log curve is just the reference gamma used when the X-OCN is decoded in post so you can easily work with the linear X-OCN material. X-OCN has far more data than S-Log or any of the standard gamma curves. But of course it won’t magically improve the low light performance, so if you are struggling with light and noise you will still struggle. X-OCN is amazing for green screen.

      1. Thanks Alister. I already guessed it was linear. The S-Log thing with it is very confusing for me, to be honest. Why should you want a Log-reference for linear footage? But never mind. Rather struggle with low light in linear X-OCN I suppose than with XAVC S-Log3? Or then still better take a ‘standard’ gamma in XAVC in your opinion? I guess the 16bit will bring the most benefit.
        And yes, low light is a struggle in wildlife filming and sometimes I am far better of with my consumer a7s than with my F55/R7 combo 🙂

        1. You can’t view linear in any sensible way, so you have to start from somewhere and that somewhere has to be able to deal with 14/15 stops of DR, so you need log. Plus there is an ocean of LUT’s that you can then use. ALthough personally I much prefer a colour managed workflow such as ACES that bypasses the log stage altogether.

          X-OCN is basically everything the sensor captures, so it’s as good as it gets.

  22. So this means in custom mode at Rec709(800) or with other gammas our gain will be in negative and not 0db (-6db approx) …and this is why noise will be less?

    Is this correct?

Leave a Reply to Robin Probyn Cancel reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.