This comes up a lot. People shoot in log, take it in to the grading suite or worse still the edit suite, try to grade it and are less than happy with the end result. Some people really struggle to make log look good.
Why is this? Well we normally view our footage on a TV, monitor or computer screen that uses a gamma curve that follows what is known as a “power law” curve. While this isn’t actually a true linear type of curve, it most certainly is not a log curve. Rec-709 is a “power law” curve.
The key thing about this when trying to understand why log can be tricky to grade is that in the real world, the world that we see, as you go up in brightness for each stop brighter you go, there is twice as much light. A power law gamma such as 709 follows this fairly closely as each brighter stop recorded uses a lot more data than the previous. But log is quite different, to save space, log uses more or less the same amount of data for each stop, with the exception of the darkest stops that have very little data anyway. So conventional gamma = much more data per brighter stop, log gamma = same data for each stop.
So time to sit down somewhere quiet before trying to follow this crude explanation. It’s not totally scientifically accurate, but I hope you will get the point and I hope you will see why trying to grade Log in a conventional edit suite might not be the best thing to try to do.
Lets consider a scene where the brightness might be represented by some values and we record this scene with a convention gamma curve. The values recorded might go something like this, each additional stop being double the previous:
CONVENTIONAL RECORDING: 1 – 2 – 4 – 8 – 16
Then in post production we decide it’s a bit dark so we increase the gain by a factor of two to make the image brighter, the output becomes:
CONVENTIONAL AFTER 2x GAIN: 2 – 4 – 8 – 16 – 32
Notice that the number sequence uses the same numbers but they get bigger, doubling for each stop. In an image this would equate to a brighter picture with the same contrast.
Now lets consider recording in log. Log uses the same amount of data per stop, so the recorded levels for exactly the same scene would be something like this:
LOG RECORDING (“2” for each stop): 1 – 2 – 4 – 6 – 8.
If in post production if we add a factor of two gain adjustment we will get the same brightness as our uncorrected conventional recording, both reach 16, but look at the middle numbers they are different, the CONTRAST will be different.
LOG AFTER 2x GAIN: 2 – 4 – 8 – 12 – 16.
It gets even worse if we want to make the log footage as bright as the corrected conventional footage. To make the log image equally bright to the corrected conventional footage we have to use 4x gain. Then we get:
LOG AFTER 4x GAIN: 4 – 8 – 16 – 24 – 32
So now we have the same final brightness for both the corrected conventional and corrected log footage but the contrast is very different. The darks and mids from the log have become brighter than they should be, compare this to the conventional after 2x gain. The contrast has changed. This is the problem with log. Applying simple gain adjustments to log footage results in both a contrast and brightness change.
So when grading log footage you will typically need to make separate corrections to the low middle and high range. You want a lift control to adjust the blacks and deep shadows, a mid point level shift for the mid range and a high end level shift. You don’t want to use gain as not only will it make the picture brighter and darker but it will also make it more or less contrasty.
One way to grade log is to use a curve tool to alter the shape of the gamma curve, pulling down the blacks while stretching out the whites. In DaVinci Resolve you have a set of log grading color wheels as well as the conventional primary color wheels. Another way to grade log is to apply a LUT to it and then grade in more conventional 709 space, although arguably any grading is best done prior to the LUT.
Possibly the best way is to use ACES. The Academy of Motion Pictures workflow takes your footage, whether log or linear and converts it to true linear within the software. Then all corrections take place in linear space where it is much more intuitive before finally be output from ACES with a film curve applied.
Most tools with ACES grading, including Baselight and Catalyst Prepare, use ACESproxy color space for grading, which is log and not linear. Colorists familiar with industry-standard Lift/Gamma/Gain controls with experience in video and log graphing found ACES too unwieldy to grade in linear, hence the addition of ACESproxy for grading. It gives it a more familiar feel.
Hi Dennis. Everything I’ve seen points to mainstream ACES working primarily in 16 bit half float linear RGB is this incorrect? I thought this was done to preserve the full highlight range of material throughout the grading process, even when levels are taken well beyond the original input range. This is one of the key goals of ACES and can’t be achieved with log encoding. ACES is Light linear is it not? Obviously Log ACES is Log, but the open EXR files normally used for ACES are 16 bit half float light linear. If you apply a gain control in ACES you get an exposure shift, not just a range expansion and this indicates to me linear corrections not log. And why is grading in linear more unwieldy than grading in log?
ACESproxy provides smaller 10 bit log files for restricted systems and storage but is not suitable for end to end production, you still have to go back to full ACES files or at a push ACES log for the final output. When using ACES proxy my understanding is that an S curve is applied to the proxies internally within ACES so the grading controls have the same function as if the material is ACES light linear.
Dear Mr. Chapman,
Thank you for all your careful work and explaining the nature and use of slog2 and the A7s. It helped me understand a lot, however it has not helped me in practical terms. Maybe you can shed some light on this – after all, I can only humbly learn.
I tried to follow your instructions using the A7ii, seeing that it has slog2 built-in and should work just the same, if I understood correctly (only the base iso should be different and chosen lower to maximize the dynamic “spread” the values of the codec can cleanly be assigned to.)
Oddly the results using your luts (the plain vanilla 709 and plus one and plus two stops) were not good. Especially the first was really, really noisy. I used the “cheap” method with a white paper and zebra set to 70% to expose with just no zebra visible (with one or two stops over, I simply adjusted the shutter speed to twice and four times as long – it was a sunny day and at f16 I was in the 1/1000 region anyhow).
An uneducated guess of mine: Possibly the sony engineers forgot to fine tune for the a7ii sensor? After all the minimum ISO setting in picture profile 7 is 1600 which seems a bit high for a 24Mpix camera as base iso… What do you think?
Best regards and thank you for all the work
George
If you have exposed with white at 70% you are 1.5 stops over exposed. Opening the shutter still longer will take you even more overexposed. It’s not unusual to have a bit of noise.
May you can doublecheck your example explanation. From my understanding its unfortunately wrong.
You start log with “2 values per stop”. Now you add 2x gain. Result is that you have “4 values per stop” 4-8-12-16-20-24…. If you add 4xgain you will have “8 values per stop” 8-16-24-32-40-48-…. So the log has still the same amount of data per stop. No Contrast Change!
(It´s clear that contrast of log is different than contrast of conventional recording, you cannot compare the values. Log footage is always brighter in the mids than conventional footage)
If you grade log footage you normally use contrast and Pivot. Gain is nothing different than adjusting contrast with a Pivot at 0 (Zero).
For a start log footage is not always brighter in the mids than conventional gamma. That is simply incorrect. If anything they are normally darker, or to be more precise recorded at lower code values than conventional gammas.
What you are forgetting is that to avoid a reduction in contrast when you add gain each successively brighter stop needs to become not just twice as bright as the previous but must also have double the range of the previous.
If our scene is 6 stops then it could be represented thus, each brighter stop having double the range of the previous:
2-4/5-8/9-16/17-32/33-64/65-128/
Conventional gamma would record it with these hypothetical code values:
2-4/5-8/9-16/middle grey/17-32/33-64/65-128
Log gamma would record it thus:
2-4/5-8/9-16/middle grey/17-32/33-48/49-64 each stop above middle grey using 16 code values in this hypothetical example.
Add 2 x gain to these and you get:
Standard gamma- much brighter but retaining normal contrast, each brighter stop using twice as many code values as the previous.
4-8/10-16/18-32/34-64/66-128/130-256
Log gamma, above the scenes middle grey each stop now in a higher, brighter range, but still has the same number of code values as the previous stop.
4-8/10-16/18-32/34-64/66-96/98-128
The top two stops of the log with 2x gain have the same range (total range of 66-128) and thus on screen contrast as stop 4 of the standard gamma plus 2x gain recording, therefore the contrast has been in effect reduced as the image is being displayed brighter, but the log recordings brighter 2 stops now occupy half as much data as the should. In fact the log + 2x gain recordings top two stops would fit into the standard gamma without gains top stop range. Each time you double the gain with a log recording you halve the on screen contrast of anything above middle grey.
My example is absolutely correct.
For a human being viewing a scene or image, for contrast to remain constant as you make the scene brighter each successively brighter stop must have double the brightness range of the previous previous stop. So as you go up in brightness to keep the contrast constant, your brightness range expands much more in the upper range compared to the mids and shadows. As you add gain this relationship must be preserved otherwise the contrast will decrease as you make the image brighter. Highlights get much much brighter while shadows only increase by a tiny amount.
With log as each stop (above middle greater least) has the same amount of data and thus each stop has exactly the same brightness range as the previous, as you add gain the image gets brighter, but because the stops don’t expand out away from each other (they all get brighter by the same amount, the mids become brighter by the same amount as the highlights) the viewed contrast reduces.
First of all many thanks for the long and detailed investigation.
But I´m not sure if I really get your Point.
In Log the mids are brighter as the rest of the values coming from the log function. This is not in an absolute way but in a relative comparison to the other values. Absolut the mids are darker, thats right, with for example middle Grey log C sitting at 400, cineon at 470, Slog3 at 420 … Thats clear.
But how would you grade log footage? Means expanding the values while keeping the same amount of values per stop that the LUT at the end is doing the right Job? It is essential that you don´t get different amounts of values in different stops during expanding the log footage…