Panasonic S1H review / hands-on - a true 6K full frame cinema camera

[helium]

21 minutes ago, Jonathan Bergqvist said:

You will of course know when the "Right" in ETTR will damage the final image. ETTR is a common way for me using GH5 and C200, each with its own zebra-levels that make me not over ETTR - so that colors wont get skewed when bringing the exposure down. If you know what you are doing, ETTR won't affect highlight roll-off.

Or, when using a camera like the C200 or BMPCC 4K, or any other capable 10 bit 4:2:2 camera, you could just expose using the traditional approach, as do most cinematographers.  It's ironic, because ETTR was suggested above as a way to avoid getting the bad results of youtube posters.  And yet ETTR, at youtube levels of competence, is more likely to destroy images than "proper" traditional exposure techniques.

This preoccupation with noise is also a bit strange, for a so-called cinema camera.  Cinema, after all, is full of noise, commonly known as grain.  

[deezid]

6 minutes ago, helium said:

Or, when using a camera like the C200 or BMPCC 4K, or any other capable 10 bit 4:2:2 camera, you could just expose using the traditional approach, as do most cinematographers.  It's ironic, because ETTR was suggested above as a way to avoid getting the bad results of youtube posters.  And yet ETTR, at youtube levels of competence, is more likely to destroy images than "proper" traditional exposure techniques.

It also will limit the tonality you would have using proper exposure.
But on the other hand it will help keeping temporal noise reduction artifacts out of the image on cameras such as the A7sII/A73, S1H/S1, X-T3, E2 etc...

[EthanAlexander]

There seems to be some confusion about what's happening with ETTR so hopefully this clears it up.

2 things that will never change no matter how you're exposing V Log are the "shape" of the curve and the dynamic range (For this post we will assume 14 stops). V Log will always be V Log will always be V Log. So, technically, the highlight rolloff will always be the same. @Jonathan Bergqvist and @Mmmbeats are right in this regard.

BUT - the useable part of the dynamic range (of the scene) is definitely shifting and it is definitely destructive, as @helium is pointing out. Once you lose something to overexposure, you'll never get it back. 

If a scene only has, for instance, 7 stops of dynamic range, then you could easily argue that ETTR will offer a better image because of the high SNR which will lead to low noise and a cleaner image. We're fitting 7 stops into.a 14 stop container so it's easy to make sure everything is captured. You could probably argue that when shooting raw or super low compression, anything less than the full dynamic range should be ETTR by the amount of stops in the scene fewer than the maximum allowed by the log curve.

The complication comes in when dealing with scenes with high dynamic range. This is when you have to decide what to put into the 14 stop "container." When using ETTR, you're making a compromise - higher SNR for less useable dynamic range in the highlights. This is definitely a "destructive" choice in the sense that this can't be undone in post. You'll never get back those stops in the highlights that you chose to sacrifice. For many people and scenes, this is an acceptable tradeoff. You're also getting more stops dedicated to your shadows, which can be useful. 

If your scene has a lot of stops above middle grey, then ETTR will definitely limit the amount of useable recorded dynamic range, and exposing according to "manufacturer guidelines" (or even lower) will indeed give you the better result if you're trying not to lose anything to overexposure.

This image shows how the captured range of 14 stops never changes but the amount of stops above and below middle grey definitely do change. I think it is for the Alexa but the concept applies to literally any curve, so just ignore the exact numbers of stops above and below 18% and look at how it always adds up to 14. These changes are 100% baked in no matter how you record (raw or highly compressed).

 

 

There can be color shifts as well, but that's a totally different topic to dive into...

 

[Otago]

7 minutes ago, EthanAlexander said:

If a scene only has, for instance, 7 stops of dynamic range, then you could easily argue that ETTR will offer a better image because of the high SNR which will lead to low noise and a cleaner image. We're fitting 7 stops into.a 14 stop container so it's easy to make sure everything is captured. You could probably argue that when shooting raw or super low compression, anything less than the full dynamic range should be ETTR by the amount of stops in the scene fewer than the maximum allowed by the log curve.

I think this is true if it is a linear 14 bit file but not if it is log 10 bit ( assuming each bit corresponds to an extra stop of dynamic range ) If you ETTR and put lots of information in the curve of the log then in the brightest values they will be sharing bits. Depending on what curve your camera uses you could end up with, say, your 2 brightest stops being compressed into one bit in the codec and so only have 512 values representing each stop rather than 1024 - whether that is noticeable is another question! If you exposure "correctly" then most of the values will fall in the linear part of the log curve and you'll get all 1024 values for each stop. 

I think log curves are used because it is assumed that the very brightest and darkest information will be lower in information and importance, and work similarly to film so it was easier to switch over. 

[Otago]

1 hour ago, Otago said:

I think this is true if it is a linear 14 bit file but not if it is log 10 bit ( assuming each bit corresponds to an extra stop of dynamic range ) If you ETTR and put lots of information in the curve of the log then in the brightest values they will be sharing bits. Depending on what curve your camera uses you could end up with, say, your 2 brightest stops being compressed into one bit in the codec and so only have 512 values representing each stop rather than 1024 - whether that is noticeable is another question! If you exposure "correctly" then most of the values will fall in the linear part of the log curve and you'll get all 1024 values for each stop. 

I think log curves are used because it is assumed that the very brightest and darkest information will be lower in information and importance, and work similarly to film so it was easier to switch over. 

Just realised some of this is incorrect. There are only 1024 values for the whole dynamic range, rather than each stop so the numbers about should be 50 values representing each stop rather than 100 - the concept is the same but the values are / were wrong. 

[EthanAlexander]

1 hour ago, Otago said:

I think this is true if it is a linear 14 bit file but not if it is log 10 bit ( assuming each bit corresponds to an extra stop of dynamic range ) There are only 1024 values for the whole dynamic range, rather than each stop so the numbers about should be 50 values representing each stop rather than 100 - the concept is the same but the values are / were wrong.

I appreciate your desire to dig into this further. In any linear recording, each stop brighter is actually getting twice as many bit values as the previous darker stop, starting with 1 for the darkest and 512 for the brightest in 10 bit. That means the brightest stop is actually the "top half" of all the bit values (so in 10 bit 513-1024 would actually be reserved for just one stop of light.) If you want each stop of light to be represented by an equal amount of values (for instance, ~100 as you are suggesting) It requires a log curve to map the input values to that. (How many and which values get used for the different stops is what makes the differences between different log curves like SLog2 and 3, V Log, N Log, etc.)

 

2 hours ago, Otago said:

If you ETTR and put lots of information in the curve of the log then in the brightest values they will be sharing bits. Depending on what curve your camera uses you could end up with, say, your 2 brightest stops being compressed into one bit in the codec and so only have 512 values representing each stop rather than 1024 - whether that is noticeable is another question!

They won't be sharing bits before compression whether it's linear or log, but to your point, you're right and this is a big reason why shooting log on a high compression camera is troublesome - the codec has to throw away information and that means that values that are close together will likely be compressed into one. This is why I said several times that highly compressed vs raw recording is a big factor. But if we're talking raw recording with lossless or no compression, or even ProRes HQ frankly, then a 10 or 12 bit file mapped with a log curve will look practically the same as a linear 14 bit recording. Either way you still have to decide where you want middle grey to land, which means you're deciding how many stops above and below you're going to get.

[Otago]

52 minutes ago, EthanAlexander said:

I appreciate your desire to dig into this further. In any linear recording, each stop brighter is actually getting twice as many bit values as the previous darker stop, starting with 1 for the darkest and 512 for the brightest in 10 bit. That means the brightest stop is actually the "top half" of all the bit values (so in 10 bit 513-1024 would actually be reserved for just one stop of light.) If you want each stop of light to be represented by an equal amount of values (for instance, ~100 as you are suggesting) It requires a log curve to map the input values to that. (How many and which values get used for the different stops is what makes the differences between different log curves like SLog2 and 3, V Log, N Log, etc.)

 

They won't be sharing bits before compression whether it's linear or log, but to your point, you're right and this is a big reason why shooting log on a high compression camera is troublesome - the codec has to throw away information and that means that values that are close together will likely be compressed into one. This is why I said several times that highly compressed vs raw recording is a big factor. But if we're talking raw recording with lossless or no compression, or even ProRes HQ frankly, then a 10 or 12 bit file mapped with a log curve will look practically the same as a linear 14 bit recording. Either way you still have to decide where you want middle grey to land, which means you're deciding how many stops above and below you're going to get.

Ah, I have misunderstood the terms then! I assumed by linear it was meant that the total values had been remapped to make better use of the graduations rather than the absolute light levels - but that is what log is - I assumed that this was done on all sensor ADC's as a matter of course ( the data coming off was log) and then a further log conversion was done to fit that in a 10-bit container while prioritising mid tones but I suppose it's not necessary if the sensor bit depth is close to the final bit depth, I forgot this wasn't a 24 bit instrumentation ADC where you have loads of data to throw away.

Thanks! 

[Tyler Hansen]

I got sick of these terrible test videos so I decided to go out and make a quick video. It's an amazing camera. 

 

[thebrothersthre3]

2 hours ago, Tyler Hansen said:

I got sick of these terrible test videos so I decided to go out and make a quick video. It's an amazing camera. 

 

Looks very nice! Hows the auto focus for you, if you've used it?

[crevice]

Pretty cool fight scene shot on the S1H. Found it on Instagram via searching for s1h hashtag. The creator confirmed it was shot on the s1h and then sent me a direct link to the full YouTube video. Some weird stuff happening with the lens or something else here and there, but overall colors look good

 

[thebrothersthre3]

1 minute ago, crevice said:

Pretty cool fight scene shot on the S1H. Found it on Instagram via searching for s1h hashtag. The creator confirmed it was shot on the s1h and then sent me a direct link to the full YouTube video. Some weird little gimbal shakes here and there, but overall colors look good

 

Really good fight, maybe the best indie fight scene I've seen. I think those shakes may have been intentional. 

[Lars Steenhoff]

6 hours ago, thebrothersthre3 said:

Really good fight, maybe the best indie fight scene I've seen. I think those shakes may have been intentional. 

Looks like some post warp stabiliser was used, that created that kind of distortions in the background

[AaronChicago]

I love the look/color/DR of this camera. The one thing that bugs me is the brittle looking codec. It's very noticeable in that BMPCC4K vs SH1 test. I would imagine recording to ProRes is going to save the day.

[thebrothersthre3]

2 hours ago, AaronChicago said:

I love the look/color/DR of this camera. The one thing that bugs me is the brittle looking codec. It's very noticeable in that BMPCC4K vs SH1 test. I would imagine recording to ProRes is going to save the day.

If the look/color/ and dynamic range are good, what part of it makes it brittle? 

[deezid]

27 minutes ago, thebrothersthre3 said:

If the look/color/ and dynamic range are good, what part of it makes it brittle? 

Premiere maybe?
No issues in Resolve

[mercer]

3 hours ago, thebrothersthre3 said:

If the look/color/ and dynamic range are good, what part of it makes it brittle? 

The lack of weight to the image. 

I’ve noticed it in some of the footage as well.

[thebrothersthre3]

9 minutes ago, mercer said:

The lack of weight to the image. 

I’ve noticed it in some of the footage as well.

What does that mean? An image is all looks, so if you say it looks good, has good color and dynamic range, what is left? 

[wildrym]

30 minutes ago, mercer said:

The lack of weight to the image. 

I’ve noticed it in some of the footage as well.

Yes lack of "thickness", and still the "blotchiness" on skin tones, like on GH5 Vlog.

Here's some BMPCC footage that i find to have more "weight"

Hopefully external Prores or prores RAW would improve this, but from reviews i've seen on EVA 1, 10 bit prores RAW does not improve quality that much over internal recording 

Maybe 12 bits vs 10 bit is the key...

 

 

[deezid]

44 minutes ago, mercer said:

The lack of weight to the image. 

I’ve noticed it in some of the footage as well.

What I could imagine what you mean is the strong temporal noise reduction which makes the image sometimes look artificially clean and thin and even creates smearing and ghosting in some situations.

Unless we get an option to turn it off, the image won't look as thick as Varicam footage, even recorded externally which basically looks identical (tried recording onto an Odyssey 7Q+ in ProRes HQ). 

[mercer]

5 minutes ago, deezid said:

What I could imagine what you mean is the strong temporal noise reduction which makes the image sometimes look artificially clean and thin and even creates smearing and ghosting in some situations.

Unless we get an option to turn it off, the image won't look as thick as Varicam footage, even recorded externally which basically looks identical (tried recording onto an Odyssey 7Q+ in ProRes HQ). 

I noticed it more in lowlight shots, dimly lit interiors and ext. night so I guess noise reduction could be kicking in more aggressively at higher ISOs?