This comes up again and again, hence why I am writing about it once again.
Raw should never be converted to log before recording if you want any benefit from the raw. You may as well just record the 10 bit log that most cameras are capable of internally. Or take log and output it via the cameras 10 bit output (if it has one) and record that directly on the ProRes recorder. It doesn’t matter how you do it but if you convert between different recording types you will always reduce the image quality and this is as bad a way to do it as you can get. This mainly relates to cameras like the PXW-FS7. The FS5 is different because it’s internal UHD recordings are only 8 bit, so even though the raw is still compromised by converting it to ProRes log, this can still be better than the internal 8 bit log.
S-Log like any other log is a compromise recording format. Log was developed to squash a big dynamic range into the same sized recording bucket as would normally be used for conventional low dynamic range gammas. It does this by discarding a lot of tonal and textural information from everything brighter than 1 stop above middle grey, instead of the amount of data doubling for each stop up you go in exposure, it’s held at a constant amount. Normally this is largely transparent as human vision is less acute in the highlight range, but it is still a compromise.
The idea behind Linear raw is that it should give nothing away, each stop SHOULD contain double the data as the one below. But if you only have 12 bit data that would only allow you to record 11 stops of dynamic range as you would quickly run out of code values. So Sony have to use floating point math or something very similar to reduce the size of each stop by diving down the number of code values each stop has. This has almost no impact on highlights where you start off with 100’s or 1000’s values but in the shadows where a stop may only have 8 or 16 values dividing by 4 means you now only have 2 or 4 tonal levels. So once again this is a compromise recording format. To record a big dynamic range using linear what you really need is 16 bit data.
In summary so far:
S-Log reduces the number of highlight tonal values to fit it a big DR in a normal sized bucket.
Sony’s FSRaw, 12 Bit Linear reduces the number of tonal Values across the entire range to fit it in a compact 12 bit recording bucket, but the assumption is that the recording will be at least 12 bit. The greatest impact of the reduction is in the shadows.
Convert 12 bit linear to 10 bit S-Log and now you are compromising both the highlight range and the shadow range. You have the worst of both, you have 10 bit S-Log but with much less shadow data than the S-log straight from the camera. It’s really not a good thing to do and the internally generated S-Log won’t have shadows compromised in the same way.
If you have even the tiniest bit of under exposure or you attempt to lift the shadows in any way this will accentuate the reduced shadow data and banding is highly likely as the values become stretched even further apart as you bring them up the output gamma range.
If you expose brightly and then reduce the shadows this has the effect of compressing the values closer together or pushing them further down the output curve, closing them together as they go down the output gamma range, this reduces banding. This is one of the reasons why exposing more brightly can often help both log and raw recordings. So a bit of over exposure might help, but any under exposure is really, really going to hurt. Again, you would probably be better off using the internally generated S-Log.
To make matters worse there is also often an issue with S-Log in a ProRes file.
If all that is not enough there is also a big problem in the way ProRes files record S-Log. S-Log should always be recorded as full range data. When you record an internal XAVC file the metadata in the clips tells the edit or grading software that the file is full range. Then when you apply a LUT or do your grading the correct transforms occur and all shadow textures are preserved. But ProRes files are by default treated as legal range files. So when you record full range S-Log inside a ProRes file there is a high likelihood that your edit or grading software will handle the data in the clip incorrectly and this too can lead to problems in the shadows including truncated data, clipping and banding, even though the actual recorded data may be OK. This is purely a metadata issue, grading software such as DaVinci resolve can be forced to treat the ProRes files as full range.
This topic comes up a lot. Whenever I have been in discussion with those that should know within Sony they have made it clear that the FS-Raw system is designed around S-Log2 for monitoring and post production etc. This stems from the fact that FS-Raw, the 12 bit linear raw from the FS700, FS7 and FS5 was first developed for the FS700 and that camera only had SGamut and S-Log2. S-Log3 didn’t come until a little later.
The idea is that if the camera is set to SGamut + S-Log2 it is optimised for the best possible performance. The raw signal is then passed to the raw recorder where it will be recorded. For a raw recorder that is going to convert the raw to ProRes or DNxHD the recorder converts the raw to S-SGamut + Log2 so that it will match any internal recordings.
Finally in post the grading software would take the FS-Raw and convert it to SGamut + S-Log2 for further grading. By keeping everything as SGamut and S-Log2 throughout the workflow your brightness levels, the look of the image and any LUT’s that you might use will be the same. Internal and external recordings will look the same. And this has been my experience. Use PP7 with SGamut and S-Log2 and the workflow works as expected.
What about the other Gamuts?
However: The FS5 also has SGamut3, SGamut3.cine and S-Log3 available in the picture profiles. When shooting Log many people prefer S-Log3 and SGamut3.cine. Some people find it easier to grade S-Log3 and there are more LUT’s available for S-Log3/SGamut3.cine than for SGamut and S-Log2. So there are many people that like to use PP8 or PP9 for internal S-Log.
However, switching the FS5’s gamma from S-log2 to S-log3 makes no difference to the raw output. And it won’t make your recorder convert the raw to ProRes/DNxHD as S-Log3 if that’s what you are hoping for. But changing the gamut does have an effect on the colors in the image.
But shouldn’t raw be just raw sensor data?
For me this is interesting, because if the camera is recording the raw sensor output, changing the Gamut shouldn’t really change what’s in the raw recording. So the fact that the image changes when you change the Gamut tells me that the camera is doing some form of processing or gain/gamma adjustment to the signal coming from the sensor. So to try and figure out what is happening and whether you should still always stick to SGamut I decided to do a little bit of testing. The testing was only done on an FS5 so the results are only applicable to the FS5. I can’t recall seeing these same changes with the FS7.
DSC Labs Chroma Tru Test Chart.
For the tests I used a DSC Labs Chroma-tru chart as this allows you to see how the colors and contrast in what you record changes both visually and with a vectorscope/waveform. As well as the chart that you shoot, you download a matching reference overlay file that you can superimpose over the clip in post to visually see any differences between the reference overlay and the way the shot has been captured and decoded. It is also possible to place another small reference chart directly in front of the monitor screen if you need to evaluate the monitor or any other aspects of your full end to end production system. It’s a very clever system and I like it because as well as being able to measure differences with scopes you can also see any differences quite clearly without any sophisticated measuring equipment.
The chart was illuminated with a mix of mostly real daylight and a bit of 5600K daylight balanced light from a Stella LED lamp. I wanted a lot of real daylight to minimise any errors that could creep in from the spectrum of the LED light (The Stellas are very good but you can’t beat real daylight). The camera was set to 2000 ISO. The raw signal was passed from the camera to an Atomos Shogun Inferno where the clips were recorded as both ProRes Raw and also by using the recorders built in conversion to S-Log2 for internal recording as ProRes HQ. I did one pass of correctly exposed clips and a second pass where the clips were under exposed by 1 stop to asses noise levels. The lens was the 18-105mm kit lens, which without the cameras built in lens compensation does show a fair bit of barrel distortion as you will see!
The ProRes clips were evaluated in DaVinci Resolve using the DaVinci Color Managed workflow with the input colorspace set to S-Log2/Sgamut for every clip and output colorspace set to 709. I also had to set the input range of the ProRes clips to Full Range as this is what S-Log2 files always are. If I didn’t change the input range to Full Range the clips exhibited clipped and crushed black after conversion to 709, this confirms that the clips recorded by the Shogun were Full Range – which follows the S-Log specifications.
I did also take a look at the clips in Adobe Premiere and saw very similar results to Resolve.
I will do a separate report on my findings with the ProRes Raw in FCP as soon as I get time to check out the ProRes raw files properly.
So, what did I find?
In the images below the reference file has been overlaid on the very center of the clip. It can be a little hard to see. In a perfect system it would be impossible to see. But you can never capture the full contrast of the chart 1:1 and all cameras exhibit some color response imperfections. But the closer the center overlay is to the captured chart, the more accurate the system is. Note you can click on any of the capture examples to view a larger version.
Below is Picture Profile 6 (PP6) SGamut with S-Log2. It’s pretty good match. The camera didn’t quite capture the full contrast of the chart and that’s to be expected, reflections etc make it very difficult to get perfect blacks and shadow areas. But color wise it looks quite reasonable although the light blue’s are a little weak/pink.
Below is Picture Profile 7 (PP7) SGamut3 with S-Log3. Straight away we can see that even though the camera was set to S-log3, the contrast is the same in the S-Log2 color managed workflow proving that the gamma of the recording is actually ProRes recording from the Shogun is S-log2, confirming what we already know which is that changing the log curve in the camera makes no difference to the raw recording and no difference to the raw to ProRes conversion in the recorder.
Note the extra noise in the greens. The greens appear to have more color, but they also appear a little darker. If you reduce the brightness of a color without altering the saturation the color appears to be deeper and I think that is what is happening here, it is a lightness change rather than just a saturation change. There is also more noise in the darker bars, grey and black really are quite noisy. Light blues have the same weak/pink appearance and there is a distinct green tint to the white, grey and black bars.
Below is when the camera was set to SGamut3.cine with S-Log3. Again we can see that the recording gamma is obviously S-Log2. The greens are still a touch stronger looking but now there is less noise in the greens. Cyan and reds are slightly lighter than SGamut and yellows appear a bit darker. This is also a little more noisy overall than SGamut, but not as bad as SGamut3. When you play the 3 clips, overall SGamut has the least noise, SGamut3.cine is next and then SGamut3 is clearly the noisiest. As with SGamut there is a distinct green tint to the white, grey and black bars.
So that’s what the images look like, what do the scopes tell us. Again I will start with SGamut and we can see that the color response is pretty accurate. This suggests that Atomos do a good job of converting the raw to S-Log2/SGamut before it’s recorded and confirms what we already know which is this is that this is clearly how the system is designed to work. Note how the Red strip falls very close to the R box on the 2x vectorscope, yello almost in Y, green very close to G, Blue almost in B. Magenta isn’t so clever and this probably explains why the pinky blues at the top of the chart are not quite right. Do remember that all these test were done with the preset white balance so it’s not surprising to see some small offsets as the white balance won’t have been absolutely perfect. But that imperfection will be the same across all of my test examples.
Below is SGamut3. The first thing I noticed was all the extra noise on the right side of the waveform where the greens are. The waveform also shows the difference in lightness compared to SGamut with different colors being reproduced at different brightness levels. The greens are being reproduced at a slightly lower luma level and this is probably why the greens appear more saturated. Also notice how much more fuzzy the vectorscope is, this is due to some extra chroma noise. There is a bit more red and magenta is closer to it’s target box, but all the other key colors are further from their boxes. Yellow and Green and Cyan are all a long way from their target boxes. Overall the color is much less accurate than SGamut and there is more chroma noise.
And finally below is SGamut3.cine. There is less noise on the green side of the waveform than SGamut and SGamut3 but we still have a slightly lower luma level for green, making green appear more saturated. Again overall color accuracy is not as good as SGamut. But the vector scope is still quite fuzzy due to chroma noise.
I just want to show you a couple of under exposed examples. These have had the under exposure corrected in post. Below is SGamut and as you can see it is a bit noisy when under exposed. That shouldn’t be a surprise, under expose and you will get noisy pictures.
Below is SGamut3 and you can really see how much noisier this is than SGamut. I recommend clicking on the images to see a full screen version. You will see that as well as the noise in the greens there is more chroma noise in the blacks and greys. There also seems to be a stronger shift towards blue/green in the whites/greys in the under exposed SGamut3.
Clearly changing the gamut makes a difference to the raw output signal. In theory this shouldn’t really happen. Raw is supposed to be the unprocessed sensor output. But these test show that there is a fair bit of processing going on in the FS5 before the raw is output. It’s already known that the white balance is baked in. This is quite easy to do as changing the white balance is largely just a matter of changing the gain on the pixels that represent red and blue relative to green. This can be done before the image is converted to a color image.
What I believe I am seeing in this test is something more complex than that. I’m seeing changes in the luminance and gain levels of different colors relative to each other. So what I suspect is happening is that the camera is making some independent adjustments to the gamma of the Red, Blue and Green pixels before the raw signal is output. This is probably a hang over from adjustments that need to be made when recording S-Log2 and S-Log3 internally rather than something being done to deliberately adjust the raw output. But I didn’t design the camera so I can’t be sure that this is really the case. Only Sony would know the truth.
Does it matter?
Yes and no. If you have been using SGamut3.cine and have been getting the results you want, then, no it doesn’t really matter. I would probably avoid SGamut3. It really is very noisy in the greens and shadows compared to the other two. I would be a little concerned by the green tint in the parts of the image that should be colour free in both SGamut3 and SGamut3.cine. That would make grading a little tougher than it should be.
So my advice remains unchanged and continues to match Sony’s recommendation. This is that you should use PP7 with SGamut and S-Log2 when outputting raw. That doesn’t mean you can’t use the other Gamuts and your milage may vary, but these tests do for me at least confirm my reasons for sticking with PP7.
Both Premiere and Resolve show the same behaviour. Next I want to take a look at what happens in FCP with the ProRes Raw clips. This could prove interesting as FCP decodes and converts the FS-Raw to S-Log3 and SGamut3.cine rather than S-Log2/Sgamut by default. Whether this will make any difference I don’t know. What I do know is that having a recorder that’s converting to S-Log2 for display and software that converts to S-Log3 is very confusing as you need different LUT’s for post and the recorder if you want to use LUT’s for your monitoring. But FCP will have to wait for another day. I have paying work to do first.
Even though I have written about these many times before the message still just doesn’t seem to be getting through to people.
Since the dawn of photography and video the only way to really change the signal to noise ratio and ultimately how noisy the pictures are is by changing how much light you put onto the sensor.
Gain, gamma, log, raw, etc etc only have a minimal effect on the signal to noise ratio. Modern cameras do admittedly employ a lot of noise reduction processes to help combat high noise levels, but these come at a price. Typically they soften the image or introduce artefacts such as banding, smear or edge tearing. So you always want to start off with the best possible image from the sensor with the least possible noise and the only way to achieve that is through good exposure – putting the optimum amount of light onto the sensor.
ISO is so confusing:
But just to confuse things the use of ISO to rate an electronic cameras sensitivity has become normal. But the problem is that most people have no clue about what this really means. On an electronic camera ISO is NOT a sensitivity measurement, it is nothing more than a number that you can put into an external light meter to allow you to use that light meter to obtain settings for the shutter speed and aperture that will give you the camera manufacturers suggest optimum exposure. That’s it – and that is very different to sensitivity.
Lets take Sony’s FS7 as an example (most other cameras behave in a very similar way).
If you set the FS7 up at 0dB gain, rec-709, it will have an exposure rating of 800 ISO. Use a light meter to expose with the meters ISO dial set to 800. Lets say the light meter says set the aperture to f8. When you do this the image is correctly exposed, looks good (well as good as 709 gets at least) and for most people has a perfectly acceptable amount of noise.
Now switch the camera to S-Log2 or S-Log3. With the camera still set to 0dB the ISO rating changes to 2000 which give the impression that the camera may have become more sensitive. But did we change the sensor? No. Have we added any more gain? No, we have not, the camera is still at 0dB. But if you now expose at the recommended levels, after you have done your grading and you grade to levels similar to 709 the pictures will look quite a lot noisier than pictures shot using Rec-709.
So what’s going on?
If you now go back to the light meter to expose the very same scene, you turn the ISO dial on the light meter from 800 to 2000 ISO and the light meter will tell you to now set the aperture to f13 (approx). So starting at the f8 you had for 800 ISO, you close the aperture on the camera by 1.3 stops to f13 and you will have the “correct” exposure.
BUT: now you are putting 1.3 stops less light on to the sensor so the signal coming from the sensor is reduced by 9dB and as a result the sensor noise that is always there and never really changes is much more noticeable. As a result compared to 709 the graded S-Log looks noisy and it looks noisier by the equivalent of 9dB. This is not because you have changed the cameras sensitivity or changed because you have changed the amount of camera gain but because compared to when you shoot in 709 the sensor is being under exposed and as a result it is outputting a signal 9dB lower. So in post production when you grade or add a LUT you have to add 9dB of gain to get the same brightness as the original direct rec-709 recording and as well as making the desirable image brighter it also makes the noise 9dB higher (unless you do some very fancy noise reduction work in post).
So what do you do?
It’s common simply to open the aperture back up again, typically by 1.5 stops so that after post production grading the S-log looks no more noisy than the 709 from the FS7 – Because in reality the FS7’s sensor works best for most people when rated at the equivalent of 800 ISO rather than 2000 – probably because it’s real sensitivity is 800 ISO.
When you think about it, when you shoot with Rec-709 or some other gamma that won’t be graded it’s important that it looks good right out of the camera. So the camera manufacturer will ensure that the rec-709 noise and grain v sensitivity settings are optimum – so this is probably the optimum ISO rating for the camera in terms of noise, grain and sensitivity.
So don’t be fooled into thinking that the FS7 is more sensitive when shooting with log, because it isn’t. The only reason the ISO rating goes up as it does is so that if you were using a light meter it would make you put less light onto the sensor which then allows the sensor to handle a brighter highlight range. But of course if you put less light onto the sensor the sensor won’t be able to see so far into the shadows and the picture may be noisy which limits still further the use of any shadow information. So it’s a trade-off, more highlights but less shadows and more noise. But the sensitivity is actually the same. Its’s an exposure change not a sensitivity change.
So then we get into the S-Log2 or S-Log3 debate.
First of all lets just be absolutely clear that both have exactly the same highlight and shadow ranges. Both go to +6 stops and -8 stops, there is no difference in that regard. Period.
And lets also be very clear that both have exactly the same signal to noise ratios. S-log3 is NOT noisier than S-log2. S-log 3 records some of the mid range using higher code values than S-Log2 and before you grade it that can sometimes make it appear like it’s noisier, but the reality is, it is not noisier. Just like the differing ISO ratings for different gamma curves, this isn’t a sensitivity change, it’s just different code values being used. See this article if you want the hard proof: http://www.xdcam-user.com/2014/03/understanding-sonys-slog3-it-isnt-really-noisy/
Don’t forget when you shoot with log you will be grading the image. So you will be adjusting the brightness of the image. If you grade S-Log2 and S-Log3 to the same brightness levels the cumulative gain (the gain added in camera and the gain added in post) ends up the same. So it doesn’t matter which you use in low light the final image, assuming a like for like grade will have the same amount of noise.
For 8 bit records S-Log2 has different benefits.
S-Log2 was designed from the outset for recording 14 stops with an electronic video camera. So it makes use of the cameras full recording range. S-Log3 is based on an old film log curve (cineon) designed to transfer 16 stops or more to a digital intermediate. So when the camera only has a 14 stop sensor you waste a large part of the available recording range. On a 10 bit camera this doesn’t make much difference. But on a 8 bit camera where you are already very limited with the number of tonal values you can record it isn’t ideal and as a result S-Log2 is often a better choice.
But if I shoot raw it’s all going to be so much better – isn’t it?
Yes, no, maybe…. For a start there are lot’s of different types of raw. There is linear raw, log raw, 10 bit log raw, 12 bit linear, 16 bit linear and they are all quite different.
But they are all limited by what the sensor can see and how noisy the sensor is. So raw won’t give you less noise (it might give different looking noise). Raw won’t give you a bigger dynamic range so it won’t allow you to capture deeper or brighter highlights.
But what raw does normally is to give you more data and normally less compression than the cameras internal recordings. In the case of Sony’s FS5 the internal UHD recordings are 8 bit and highly compressed while the raw output is 12 bit, that’s a 4 fold increase in the amount of tonal values. You can record the 12bit raw using uncompressed cDNG or Apples new ProResRaw codec which doesn’t introduce any appreciable compression artefacts and as a result the footage is much more flexible in post production. Go up to the Sony Venice, F5 or F55 cameras and you have 16 bit raw and X-OCN (which behaves exactly like raw) which has an absolutely incredible range of tonal values and is a real pleasure to work with in post production. But even with the Venice camera the raw does not have more dynamic range than the log. However because there are far more tonal values in the raw and X-OCN you can do more with it and it will hold up much better to aggressive grading.
It’s all about how you expose.
At the end of the day with all of these camera and formats how you expose is the limiting factor. A badly exposed Sony Venice probably won’t end up looking anywhere near as good as a well exposed FS7. A badly exposed FS7 won’t look as good as a well exposed FS5. No camera looks good when it isn’t exposed well.
Exposure isn’t brightness. You can add gain to make a picture brighter, you can also change the gamma curve to change how bright it is. But these are not exposure changes. Exposure is all about putting the optimum amount of light onto the sensor. Enough light to produce a signal from the sensor that will overcome the sensors noise. But also not so much light that the sensor overloads. That’s what good exposure is. Fiddling around with gamma curves and gain settings will only every make a relatively small difference to noise levels compared to good exposure. There’s just no substitute for faster lenses, reflectors or actually adding light if you want clean images.
And don’t be fooled by ISO ratings. They don’t tell you how noisy the picture is going to be, they don’t tell you what the sensitivity is or even if it’s actually changing. All it tells you is what to set a light meter to.
There are a lot of videos circulating on the web right now showing what appears to be some kind of magic trick where someone has shot over exposed, recorded the over exposed images using ProRes Raw and then as if by magic made some adjustments to the footage and it goes from being almost nothing but a white out of over exposure to a perfectly exposed image.
This isn’t magic, this isn’t raw suddenly giving you more over exposure range than you have with log, this is nothing more than a quirk of the way FCP-X handles ProRes Raw material.
Before going any further – this isn’t a put-down of raw or ProRes raw. It’s really great to be able to take raw sensor data and record that with only minimal processing. There are a lot of benefits to shooting with raw (see my earlier post showing all the extra data that 12 bit raw can give). But a magic ability to let you over expose by seemingly crazy amounts isn’t something raw does any better than log.
Currently to work with ProRes Raw you have to go through FCP-X. FCP-X applies a default sequence of transforms to the Raw footage to get it from raw data to a viewable image. These all expect the footage to be exposed exactly as per the camera manufacturers recommendations, with no leeway. Inside FCP-X it’s either exposed exactly right, or it isn’t.
The default decode settings include a heavy highlight roll-off. Apple call it “Tone Mapping”. Fancy words used to make it sound special but it’s really no different to a LUT or the transforms and processes that take place in other raw decoders. Like a LUT it maps very specific values in the raw data to very specific output brightness values. So if you shoot just a bit bright – as you would often do with log to improve the signal to noise ratio – The ProRes raw appears to be heavily over exposed. This is because anything bright ends up crushed into nothing but flat white by the default highlight roll off that is applied by default.
In reality the material is probably only marginally over exposed, maybe just one to 2 stops which is something we have become used to doing with log. When you view brightly exposed log, the log itself doesn’t look over exposed, but if you apply a narrow high contrast 709 LUT to it, it then the footage looks over exposed until you grade it or add an exposure compensated LUT. This is what is happening by default inside FCP-X, a transform is being applied that makes brightly exposed footage look very bright and possibly over exposed – because thats the way it was shot!
This is why in FCP-X it is typical to change the color library to WCG (Wide Color Gamut) as this changes the way FCP-X processes the raw, changing the Tone Mapping and most importantly getting rid of the highlight roll off. With no roll-off, highlights and any even slight over exposure will still blow out as you can’t show 14 stops on a conventional 6 stop TV or monitor. Anything beyond the first 6 stops will be lost, the image will look over exposed until you grade or adjust the material to control the brighter parts of the image and bring them back into a viewable range. When you are in WCG mode in FCP-X the there is no longer a highlight roll off crushing the highlights and now because they are not crushed they can be recovered, but there isn’t any more highlight range than you would have if you shot with log on the same camera!
None of this is some kind of Raw over exposure magic trick as is often portrayed. It’s simply not really understanding how the workflow works and appreciating that if you shoot bright – well it’s going to look bright – until you normalise it in post. We do this all the time with log via LUT’s and grading too! It can be a little more straight forward to recover highlights from Linear Raw footage as comes form an FS5 or FS7 compared to log. That’s because of the way log maintains a constant data level in each highlight stop and often normal grading and colour correction tools don’t deal with this correctly. The highlight range is there, but it can be tricky to normalise the log without log grading tools such as the log controls in DaVinci Resolve.
Another problem is the common use of LUT’s on log footage. The vast majority of LUT’s add a highlight roll off, if you try to grade the highlights after adding a LUT with a highlight roll off it’s going to be next to impossible to recover the highlights. You must do the highlight recovery before the LUT is added or use a LUT that has compensation for any over exposure. All of these things can give the impression that log has less highlight range than the raw from the same camera. This is not normally the case, both will be the same as it’s the sensor that limits the range.
The difference in the highlight behaviour is in the workflows and very often both log and raw workflows are miss-understood. This can lead to owners and users of these cameras thinking that one process has more than the other, when in reality there is no difference, it’s appears to be different because the workflow works in a different way.
This is a much discussed topic right now, so as I promised in my last article about this, I have put together a video. Unfortunately YouTube’s compression masks many of the differences between the UHD XAVC and the ProRes Raw, but you can still see them, especially on the waveform scopes.
To really appreciate the difference you should watch the video on a large screen at at high quality, preferably 4K.
This just keeps coming up over and over. Almost all log gamma curves and the majority of raw recording formats don’t have a highlight roll-off. Any roll off that you might see is probably in the LUT’s that you are using.
The whole point of log and raw is to capture as much information about the scene that you are shooing as you can. Log normally achieves this by recording every stop above middle grey with a constant amount of data, so even the very brightest stop has the same amount of recording data as the ones below it – there is no roll off.
In conventional limited range recordings such as Rec-709, hypergamma, cinegamma etc, highlight roll-offs work by reducing the contrast in the highlights to make the amount of data needed to record the very brightest stops much smaller than used for the rest of the image. This allows 2 or 3 stops to be squeezed into a very small recording range, keeping most of the recording data available for a nice bright high contrast image. The reduction in contrast in the extreme highlights helps hide any highlight handling problems and makes it appear as though the sensors clipping point is reach in a more pleasing soft manner.
But you don’t want this in a log or raw recording as it makes grading much harder as the footage will contain different contrast ranges, each needing it’s own grading adjustments. Also by reducing contrast in the highlights you are reducing the data. It would be very difficult to un-pick a highlight roll off and if you did want to expand the data back out you will get issues such as banding.
S-Log2 and S-Log3 like almost all log gammas have no highlight roll-off. The only roll off is from middle grey and down. So if you underexpose you will start to roll away the data in your scenes mid range and that’s not good. Expose for the mid range, this is the most important part of any image. If your highlights are a bit clipped don’t worry about this too much. In post production you can add a roll off in the grade that will make any clipped highlights roll away gently. Adding a bit of highlight diffusion in post will also nicely mask any clipped highlights and make them look natural.
There has been a lot of discussion recently and few videos posted that perhaps give the impression that if you shoot with S-Log2 on an FS5 and compare it to raw shot on the FS5 there is very little difference.
Many of the points raised in the videos are correct. ProRes raw won’t give you any more dynamic range. It won’t improve the cameras low light performance. There are features such as automatic lens aberration correction applied when shooting internally which isn’t applied when shooting raw. Plus it’s true that shooting ProRes raw requires an external recorder that makes the diminutive little FS5 much more bulky.
So why in that case shoot ProRes Raw?
Frankly, if all you are doing is producing videos that will be compressed to within an inch of their life for YouTube, S-Log2 can do an excellent job when exposed well, it can be graded and can produce a nice image.
But if you are trying to produce the highest quality images possible then well shot ProRes raw will give you more data to work with in post production with fewer compression artefacts than the internal 8 bit UHD XAVC.
I was looking at some shots that I did in preparation for my recent webinar on ProRes raw earlier today and at first glance there isn’t perhaps much difference between the UHD 8 bit XAVC S-Log2 files and the ProRes raw files that were shot within seconds of each other. But look more closely and there are some important differences, especially if skin tones are important too you.
Skin tones sit half way between middle grey and white and typically span around 2 to 3 stops. So with S-Log 2 and an 8 bit recording a face would span around 24 to 34 IRE and have a somewhere between 24 and 35 code values – Yes, that’s right, maybe as few as 24 shades in each of the R, G and B channels. If you apply a basic LUT to this and then view it on a typical 8 bit monitor it will probably look OK.
But compare that to 12 bit linear raw recording and the same face with 2 to 3 stops across it will have anywhere up to 10 to 20 times as many code values ( somewhere around 250 – 500 code values depending on exactly how it’s exposed) . Apply the same LUT as for the S-Log2 and on the surface it looks pretty much the same – or does it?
Look closely and you will see more texture in the 12 bit raw. If you are lucky enough to have a 10 bit monitor the differences are even more apparent. Sure, it isn’t an in-your-face night and day difference but the 12 bit skin tones look less like plastic and much more real, they just look nicer, especially if it’s someone with a good complexion.
In addition looking at my test material I am also seeing some mpeg compression artefacts on the skin tones in the 8 bit XAVC that has a smoothing effect on the skin tones, reducing some of the subtle textures and adding to the slightly un-real, video look.
The other deal with a lack of code values and H624 compression is banding. Take 8 bit S-Log2 and start boosting the contrast in a sky scene, perhaps to bring out some cloud details and you will start to see banding and stair stepping if you are not very careful. You will also see it across wall and other textureless surfaces. You can even see this on your grading suite waveform scopes in many cases. You won’t get this with well exposed 12 bit linear raw (for any normal grading at least).
None of these are huge deals perhaps. But what is it that makes a great picture? Take Sony’s Venice or the Arri Alexa as examples. We know these to be great cameras that produce excellent images. But what is it that makes the images so good? The short answer is that it is a combination of a wide range of factors, each done as well as is possible. Good DR, good colour, good skin tones etc. So what you want to record is whatever the sensor in your camera can deliver as well as you can. 8 bit UHD compressed to only 100Mb/s is not really that great. 12 bit raw will give you more textures in the mid range and highlights. It does have some limitations in the shadows, but that is easily overcome with a nice bright exposure and printing down in post.
And it’s not just about image quality.
Don’t forget that ProRes Raw makes shooting at 4K DCI possible. If you hope to ever release your work for cinema display, perhaps on the festival circuit, you are going to be much better off shooting in the cinema DCI 4K standard rather than the UHD TV standard. It also allows you to shoot 60fps in 4K (I’m in the middle of a very big 4K 60p project right now). Want to shoot even faster – well with ProRes Raw you can, you can shoot at up to 120fps in 4K. So there are many other benefits to the raw option on the FS5 and recording to ProRes raw on a Shogun Inferno.
There is also the acceptability of 8 bit UHD. No broadcaster that I know of will ever consider 8 bit UHD unless there is absolutely no other way to get the material. You are far more likely to be able to get them to accept 12 bit raw.
Future proofing is another consideration. I am sure that ProRes raw decoders will improve and support in other applications will eventually come. By going back to your raw sensor data with better software you may be able to gain better image quality from your footage in the future. With Log you are already somewhat limited as the bulk of the image processing has already been done and is baked into the footage.
It’s late on Friday afternoon here in the UK and I’ve promised to spend some time with the family this evening. So no videos today. But next week I’ll post some of the examples I’ve been looking at so that you can see where ProRes raw elevates the image quality possible from the FS5.
Both webinars will feature a Q&A session where you will be able to ask questions online. You will find the full details about both webinars by following the links above, including how to register. The webinars are free but registration is required to obtain the login details for the events.
This seems to be a source of frustration for many people shooting raw or using S-Log2 or S-Log3 on a Sony camera. When shooting log and raw you should also be using a matching S-Gamut colour gamut if you want to get the best from the camera and this ties you into one of 3 preset white balances.
With a PXW-FS7, PMW-F5 or F55 it is possible to use custom mode to select a different colour space to mix with S-Log2 or S-Log3 and then have a variable white balance. With the Alpha cameras, PXW cameras such as the FS5 you can choose any Gamut you want in the picture profiles, but I don’t recommend this. For a start, if you don’t use one of the S-Gamuts you will be limited to Rec-709 Gamut, so you won’t be recording the cameras full colour range. Also in custom mode there are some other things like noise reduction that you really don’t want when shooting S-log2/3 (it can cause banding).
So why is the S-Gamut white balance fixed to the 3 presets for daylight, fluorescent and tungsten? The main reason is to ensure you get the cameras full dynamic range in each colour. White balance is a gain function, it adjusts the gain of the red, green and blue channels so that white objects appear white under differing light sources. So if the light source lacks blue light – making the pictures look excessively warm – you add extra gain to the blue channel to compensate.
But the problem with this is that gain affects dynamic range. When shooting log (or raw) the camera needs to operate the sensor at the optimum gain level to squeeze the highest possible dynamic range from the it. Changing the gain in just one colour channel to shift the white balance could result in a reduction of dynamic range in the channel. This could manifest itself as colours in one channel that clip sooner than the others. This can be really hard to deal with in post production and can show up as things like bright clouds with a colour cast that isn’t in the rest of the picture.
Another potential issue is that because of the way silicon sensors work the blue channel is almost always noisier than the red and green. So you want to keep the gain in the blue channel as low as possible to prevent the pictures getting too noisy. This is particularly important when shooting log as you won’t see your end result until after the images have been graded. So manually shifting the gain of the blue channel in camera to correct the white balance could lead to footage that ends up noisier than you would expect.
So – Sony chose to fix the white balance to 3 carefully tuned presets designed to avoid this situation and maximise the dynamic range. After all, when shooting log or raw it is expected that the footage will be graded anyway, so the white balance will normally be adjusted as part of the post production process.
There are some people that advocate adjusting the FS5’s white balance via the picture profile settings, personally I don’t recommend this or feel that it’s necessary. But yes, you can do this, but just keep a very close eye on your highlights and if you can use monitor with RGB parade to make sure you have equal recording levels for your whites without one colour channel clipping ahead of the others. Also apply a LUT in the monitor that is close to your desired output so that you can keep an eye on the noise levels.
In summary – the white balance is preset to ensure you don’t encounter problems later on. You should be able to fully adjust and fine tune your white balance in post production to a far greater degree than is possible in camera anyway, so don’t worry if the WB is a touch off when shooting.
The only exception to this is the new Sony Venice. Venice has enough dynamic range and enough internal processing power to allow you to make a wide range of white balance adjustments in camera. Hopefully we will see some of this flexibility trickle down to the next generations of lower cost Sony digital cinema cameras.
Camera setup, reviews, tutorials and information for pro camcorder users from Alister Chapman.