8-bit REC709 is more flexible in post than you think

[IronFilm]

yeah, linear vs log

Plus ARRI can do 12bit prores too

[zlfan]

20 hours ago, PannySVHS said:

Well, well:)

1. 10bit Vlog on the S1 smashes Pannys S1 8bit 709 like Bruce Lee kicked Chuck "chesthair superchamp" Norris' butt in the fake Collosseum of Rome. No contest.

2. Most part of the curve of 10bit log if not all should hold much more values than linear 709.

3. Alister Chapman shown that 10bit Slog2 held no disadvantage on the FS7 compared to its linear 12bit Raw output. Always a great presenter and a good reason to grab some chocolate milk and some cookies and watch a 2.5 hour presentation.

I am applying kyes neat and effective form of presenting arguments. Learning from the best. 1, 2, 3... 🙂

alister chapman also has an article on the bit rate and bit depth. basically, 12 bit uncompressed raw beats 10 bit uncompressed log, and 10 bit uncompressed log beats 8 bit uncompressed rec709 or log. if compression is taken into consideration, then the codecs themselves need to be tested, things get very complicated. 

[zlfan]

15 hours ago, IronFilm said:

That was more a design flaw of the FS7 itself though

12bit from an ARRI for instance will be great. 

that is why i never consider sony's lower tier cams after 2014. fs7 fs5 sdi raw out is crippled. 

[zlfan]

13 hours ago, kye said:

Isn't ARRIRAW 12-bit LOG?  That would be broadly as good as 14-bit Linear, one 'level' better than 12-bit Linear or 10-bit log.

i think arriraw is 14 bit uncompressed. its log c is also 14 bit. probably at the step of recording shortened to 12 bit. 

[JulioD]

6 hours ago, zlfan said:

i think arriraw is 14 bit uncompressed. its log c is also 14 bit. probably at the step of recording shortened to 12 bit. 

No. 

12bit LOG uncompressed for more than 10 years. The most very recent model, the Alexa 35 does 13bit log uncompressed. 

[John Matthews]

Look how great 8bit is!

...10bit is better ...12bit is even better than 10bit ...I think I know where this conversation is going.

According to ChatGPT, to get the most out of 8bit:

1. Use a Flat Picture Profile: Any suggestions on Panasonic (without adjusting highlights and shadows)?
2. Expose Carefully: I use roughly -.3 to -.7 EV comp. Anyone else?
3. Control Contrast: that's part of the Picture Profile (on Panasonic)
4. Use a Lens Hood: good point in general, but it makes my setup bigger. With modern lenses, does it really matter so much?
5. White Balance: seems obvious... for those don't have multi-camera setups, do you just use auto WB?
6. Avoid Aggressive Color Grading: of course, it's 8bit, but how far can you go?
7. Shoot in the Best Quality: what's the minimum? 4k 100mbps -ish? What about on a tripod?
8. Use a Tripod or Stabilization: less movement = better image?
9. Control Lighting: obviously, but does 10bit even matter in such controlled situations?

[KnightsFan]

12 hours ago, John Matthews said:

...10bit is better ...12bit is even better than 10bit ...I think I know where this conversation is going.

Lol true. My point with 8 vs 10 was that the difference is readily apparent to the naked eye in most shooting conditions without any color grading (though again, it could just be my camera's implementation). From my experience shooting DNG vs ProRes on old blackmagic cameras, I can't say I ever saw a difference. So it's all about diminishing returns.

[kye]

Shot a bunch of test shots, this time with skintones.  I shot the usual shots gradually under-exposing, but instead of changing the WB in-camera manually to create the off-WB shots I bought a pack of flash gels and held them in front of the single LED light.

Also, as @John Matthews enquired about skintones, I have just done a first attempt at grading for the skintones rather than the chart.  I looked at the chart of course, but considering that people is what the audience is looking at in the image, skintones are vastly more important.  So I didn't try to get the greyscale on the test chart perfect, or correct the highlights or shadows perfectly, etc.

The first image is the reference.  Those following that are each one-stop darker than the previous one.  Then comes the gels.  The gels are so strong they're essentially a stress test, not how anyone would / should / will be shooting, but it's nice to see where the limits are.

The first set is using only the basic grading controls and no colour management:

The results aren't perfect, but if you're shooting this horrifically then you don't deserve good images anyway!  What is interesting is that some images fell straight into line and I felt like I was fighting with the controls on others, but this had no correlation with how large a correction was required.  I suspect it's my inexperience and getting lucky on some shots and unlucky on others - sometimes I'd create a new version and reset and have another go and get significantly better (or worse) results than the previous one.  No single technique or approach seemed reliable between images.  

The next set had colour management involved and grading was done in a LOG space:

I also felt this set was very hit and miss in grading.

This next set was a different set of tools again, and I added the Kodak 2383 Print Film Emulation LUT at the end.  I did this because any serious colour grading work done in post is likely to be through a 'look' of some kind and the heavier the look is the more it obscures small differences in the source images being fed into it.

Next set was a different set of tools, but keeping the 2383 LUT in the grade:

...and the final set was adding a film-like grade in addition to the Kodak 2383 LUT.  I mean, no self-respecting colourist would be using a PFE LUT as the only element of their final look...  right?

All in all, I am pretty impressed with how much latitude is in the files, and although lots of the above results aren't the nicest images ever made, if you're shooting anywhere close to correctly then the artefacts after correction are going to be incredibly minor, and if you actually shot something that was 2 stops under and was lit by a single candle using a daylight WB then you'd be pretty freaking happy with the results because you would have been thinking that that your film was completely ruined.

This is all even more true if your shots were all shot badly but using the same wrong settings, as the artefacts from neutralising the images you apply will be consistent between shots and so will have a common look.

I'm also quite surprised at how the skintones seem to survive much greater abuse than the colour chart - lots of the images are missing one whole side of the colour wheel and yet the skintones just look like you decided on the bleach-bypass look for your film and you actually know what you're doing.  The fact that half the colour chart is missing likely won't matter that much because the patches on the colour chart are very strongly saturated and real-life scenes mostly don't contain anything even close to those colours.

I'm still working on this (and in fact am programming my own set of plugins) so there will be more on this to come, especially as now I have a good set of test images.  

I think I should find a series of real-world shots that only require a small adjustment so I can demo the small changes required on real projects.

[John Matthews]

On 4/9/2024 at 10:32 AM, kye said:

I'm still working on this (and in fact am programming my own set of plugins) so there will be more on this to come, especially as now I have a good set of test images.

Thank you so much for this hard work. I'm going to look further into this during my holiday.

Yes, some of the images with wrong WB and underexposed would be expected to be trash, but it's nice to know there are some editing techniques that save it a little. In all honesty, I'd probably just go for monochrome or tint if this were to happen to me.

[kye]

13 hours ago, John Matthews said:

Thank you so much for this hard work. I'm going to look further into this during my holiday.

Yes, some of the images with wrong WB and underexposed would be expected to be trash, but it's nice to know there are some editing techniques that save it a little. In all honesty, I'd probably just go for monochrome or tint if this were to happen to me.

As you were the one that asked for skin tones, was there a specific situation you were thinking of when you asked?

I thought your question might be related to your adventures with the GX850 and shooting out in the real-world?

[John Matthews]

5 hours ago, kye said:

As you were the one that asked for skin tones, was there a specific situation you were thinking of when you asked?

I thought your question might be related to your adventures with the GX850 and shooting out in the real-world?

Yes. The question I've asked myself many times has been: "If I make a mistake in WB, is it better to error on warm side or the cool side when considering skin tones in 8 bit?" From your results, I think it's more on the warm side. Do you concur?

When I shoot, I'd really prefer to just choose ONE WB for the entire time. If I need to, I'd like to adjust just a little in post. I think the most important is to get the first shot right.

[kye]

1 hour ago, John Matthews said:

Yes. The question I've asked myself many times has been: "If I make a mistake in WB, is it better to error on warm side or the cool side when considering skin tones in 8 bit?" From your results, I think it's more on the warm side. Do you concur?

When I shoot, I'd really prefer to just choose ONE WB for the entire time. If I need to, I'd like to adjust just a little in post. I think the most important is to get the first shot right.

Good question.

I think the fundamental challenge of making corrections in post is having the tool operate in a colour space that matches the footage as closely as possible.  

For example, if your footage is in Linear, and you have a node in Linear, and you adjust the Gain wheel (which literally applies gain by doing a simple multiplication) then they match exactly and the result will be a perfect exposure or WB change, just like it was done in camera.  If you get your colour management pipeline correct then you can get this practically perfect adjustment for LOG footage too.

The challenge comes when the camera records in 709.  This is mostly because cameras don't just do a CST from Linear to 709, they apply all sorts of "make it look lovely" sort of small tweaks.  When we record in the wrong exposure or WB then these tweaks get applied wrongly.  For example, the profile might compress the skintones, and do so by expanding the reds and yellows on either side.  If you shoot a clip where the skintones are too yellow then your skintones might get expanded rather than compressed.  No CST will un-do all these small tweaks, so you're left with an image that's curved in all the wrong places rather than all the right ones.

So, what happens in practice is it comes down to the individual profile you choose (which will have its own unique set of tiny curves that make that look) and your own ability to manipulate it using the right combination of tools to get the most pleasing result.  My results vary mostly based on the luck that I had when correcting each individual test image - your results will likely suffer the same variance unless you're a far better colourist than I am.

I'd suggest you do your own tests.  Either find a spot in the shade on a sunny day, or even better is to do it on a cloudy day.  Do a manual WB against a grey card (or piece of white copy paper if you don't have a grey card), then just shoot a clip of yourself (or a volunteer model if you can get one 🙂 ).  Then shoot a range of test clips setting the Colour Temp manually.  Then just pull them all into post and see which tools seem to work the best for you, and which gives you the more pleasing looks.

One thing I did notice was that I had trouble getting the blacks and shadows to be right when the skintones were dialled in, with them tending to be the opposite of the original tint on the image (ie, if the image was warm then the correction ended up with cooler shadows) so with everything else being equal that might be a reason to go warmer so you get a bit of colour separation in the final images.

[kye]

2 hours ago, John Matthews said:

If I need to, I'd like to adjust just a little in post. I think the most important is to get the first shot right.

Thinking about this more, I think there are three different approaches.

The first is to shoot manually and get it perfect every time.  Not even the pros do this with completely controlled sets.  Colourists say that they're always making small changes to WB on a shot-by-shot basis, even on big budget productions, so this is only mentioned to make sure we understand that we will be dealing with small changes in post.

The second is to shoot on a manual WB.
This will mean that you're going to get errors in the WB, potentially being quite noticeable, but they're likely to mostly be in the warm/cool Temp direction.

The third is to shoot on auto-WB.
I've found that, on my Panasonic cameras at least, the WB errors are pretty minor, and the WB is pretty close - even if the lighting is quite variable and I'm taking shots from different angles and in different locations etc.  
This means that you'll be making only very small corrections, but they could be in the magenta/green Tint axis as well as the warm/cool Temp direction.  We're quite sensitive to Tint errors, so this means that adjusting these is a bit more fiddly, and can take some practice, but is perfectly possible.

I know that when I shoot I am very likely to completely forget a manually set WB, and will end up shooting a whole evening at 6500K and it'll be so warm it'll look like I shot it through a jar of honey, so I shoot auto-WB and therefore inevitably have to make minor corrections in post but never have to make large ones.

Going back to the minor curves that are part of the Look, and how we can't un-do in post because we don't have a complete profile of that camera/look combination, shooting on auto-WB will mean that these get applied to the footage in a place that will likely only be a very small distance from where they should have been if the shot had perfect WB.  

Obviously this still depends on your camera, the profile, your colour management pipeline, the tools in your software, your skill in applying them, and the weather and position of the stars etc...  so this is also something that you would be best testing for yourself too.

[John Matthews]

3 hours ago, kye said:

The third is to shoot on auto-WB.
I've found that, on my Panasonic cameras at least, the WB errors are pretty minor, and the WB is pretty close - even if the lighting is quite variable and I'm taking shots from different angles and in different locations etc.  
This means that you'll be making only very small corrections, but they could be in the magenta/green Tint axis as well as the warm/cool Temp direction.  We're quite sensitive to Tint errors, so this means that adjusting these is a bit more fiddly, and can take some practice, but is perfectly possible.

I like the idea of Auto-WB and have done it many times. The only issue I have with the idea is when the camera decides to alter the WB mid shot; then I need to go through the footage and try to correct. It always seems to look strange. Panasonic, in general, isn't so bad at this, but my camcorder (VX980/81) is more agressive with this.

If I shoot in daylight WB and just make sure I don't turn on any lights (or add daylight lighting), the shots will look generally great. If I'm off a little, I can correct the whole thing at once.

[kye]

I'm still lost down this rabbit hole, but these are an interesting reference.

This is what happens if you put the GX85 through a "look".  I put the GX85 test image through a bunch of output LUTs to see which (if any) I liked the flavour of.  In order to compare them equally, I adjusted after the LUT to match the black and white points, exposure, and contrast.  This way we're not just comparing the different contrast curves, but the other aspects of the LUTs like colour rendering etc.

The node structure was this:

• slightly lower gain to bring GX85 image into range (it records super-whites)
• CST from 709/2.4 to Davinci Intermediate (my preferred working colour space)
•     (my grade would go here, but for this test no adjustments were made)
• CST to whatever colour space the LUT expects
• The LUT (these all convert to 709/2.4)
• A curve to adjust the overall levels from the LUT to roughly approximate the GX85 image

The round-trip from 709/2.4 to DWG to 709/2.4 is almost transparent, if you compensate for the gamut and saturation compression in the conversion at the end, so I didn't bother to grab it.

Results:

The famous ARRI K1S1 LUT (the ARRI factory LUT):

One of the 5000 BMD LUTs that come with Resolve, which I tried just for fun:

The Kodak 2383 PFE (Print Film Emulation) LUT.  The D55 one seemed the closest match to the WB of the image for some reason, but everyone always uses the D65 ones, so I've included both here for comparison.

The D65 one:

The Kodak 2393 PFE.  It doesn't come with Resolve but it's free online from a bunch of places.  I like it because it doesn't tint the shadows as blue, so the image isn't as muddy / drab.   

The FujiFilm 3513 PFE:

I find the ARRI LUT a bit weak - it helps but not as much as I'd like.  The comparison above is flattering to the LUT because it has a bit more contrast compared to the SOOC so looks a bit better.  The skintones are a little more flattering on it though, which might be enough if you want a more neutral look.

All the PFE looks are very strong, and aren't really meant to be used on their own.  The film manufacturers designed the colour science to look good when used with a negative film like Kodak 250D or 500T stocks, so it's "wrong" to use it unless you're grading a film scan, I think people use it like this anyway 🙂

Some time ago I purchased a power-grade that emulated both the 250D and 2393 PFE from Juan Melara, which looks like this:

To me it looks much more normal than just the 2393 PFE on its own, but it's definitely a stronger look.  The powergrade is split into nodes that emulate the 250D separately to the 2393, and the 2393 nodes are almost indistinguishable from the LUT, so I'd imagine this is probably a good emulation.

Anyway, lots of flexibility in these 8-bit files!