35mm photo film emulation - LUT design

[hyalinejim]

6 hours ago, MrSMW said:

OK, had a play with your LUT plus a couple of FilmConvert alternatives:

Thanks for doing this, it's interesting to see.

6 hours ago, MrSMW said:

The order is:

1 = The Portra Lut as provided, no tweaks

This looks very green to me which suggests two factors, where one or both might be at play here:

- S1H VLog is not as prone to magenta skintones as the GH5 so a LUT built for the GH5 will give greenish skin when used on S1H. How does my LUT appear for skintones when used on other shots?

- There was an abundance of green light reflected from foliage

6 hours ago, MrSMW said:

albeit, I brought the highlights down too far

The thing I don't like about grading in Premiere is that Lumetri is not a good tool for working directly on log footage. It was designed for Rec709. If you load up this gradient, which simulates V-Log L values and check the scopes you'll see how Lumetri's sliders interfere with the log signal. If you raise exposure, for example, then the linear part of the curve bends. For small adjustments it's OK and in fact I use it all the time. But, if you want to save a shot that was recorded with the completely wrong colour temperature, to give another example, then it's not particularly accurate as this adjustment affects the highlights more than the shadows.

7 hours ago, MrSMW said:

Nitrate's version of Portra based on S1H log settings no tweaks. Err, a bit purple innit?!

It's certainly true that different LUTs purporting to emulate the same film will look different depending on the age of the film, how it was exposed, how it was developed, how it was scanned and how it was emulated. But I agree with you here this does look very magenta and that shouldn't be how Portra looks in my experience.

One thing that's obvious to me when looking at the above is that even though the first screengrab with my LUT needs a bit of colour balancing towards magenta the luminance of the colour is very pretty in comparison to the other offerings. Look particularly at the ivy leaves on the tree trunk and the orange spots on the cardigan. They're very bright with my LUT. This seems to be a characteristic of the Noritsu look (the scanner that I used).

Thanks again for making the comparison. If you have a time maybe check out my LUT on some other scenes 🙂

[MrSMW]

I’ll have another play later, but deliberately chose a tricky bit of footage!

I like Nitrate, but their Portra 400 is not even slightly close to film with S1H v log.

Neither is the Velvia which is a vibrant film stock. The Nitrate version is very flat and yellowy/green.

It could be they are all miles off with the S1H camera pack…

On the other hand, the way you can adjust the log footage is reckoned to be better than the equivalent Lumetri controls and my (limited) experience is that is true.

I will play with both later, your @hyalinejim LUT and Nitrate in general…

[MrSMW]

Nope, sorry, can't get it to work on my S1H Vlog.

I only have limited footage to work with and all +2 stops over-exposed...and it needs pulling back too far to get anything remotely like a decent exposure...but then it's not designed to work on the S1 is it?

Ah well I tried and it's not your 'fault'. 

Back to my Nitrate games. Nearly there...

[Dustin]

Interesting stuff! Recently watched the film “The Conversation” shot back in 1974, really enjoyed the film look. Bringing this back to the thread, good work so far! I shoot 35mm film from time to time and use a Lightroom plugin called Negative Lab Pro, to convert my scans (using a plustek 8100). On the video side of things, I find the Fuji XT3, to have similar filmic looks enough for me but if I had a GH5, I would def try these out!

[Benjamin Hilton]

I played around with it a bit on some S1 and GH5 footage, liking the colors a whole lot so far. I think the contrast could use some more work though, to take advantage of the dynamic range a bit more. Tends to clip pretty fast

[hyalinejim]

10 hours ago, MrSMW said:

+2 stops over-exposed...and it needs pulling back too far to get anything remotely like a decent exposure...but then it's not designed to work on the S1 is it

You're right that it's not designed specifically for the S1 but the difference lies in the colour response rather than the contrast response. If the footage is overexposed by 2 stops then yes it would need to be brought down by 2 stops in order to give a usable image.

3 hours ago, Benjamin Hilton said:

liking the colors a whole lot so far. I think the contrast could use some more work though, to take advantage of the dynamic range a bit more. Tends to clip pretty fast

Thanks for trying it. It's the contrast level that I'm most interested in getting feedback on, so thank you for that! The reason that it's so contrasty is because it's emulating how a mini lab scans negative film.... and this is quite contrasty because the intended output medium originally was to make prints on photographic paper. What's very interesting to me is that I have found while making profiles for Lightroom that a level of contrast that's acceptable to me there and is similar to Adobe's defaults appears to be far too contrasty when applied to video. I have a few ideas why this might be so, and it might be a combination of factors

- you can get away with more contrast in a photo because it's static, but with video it's likely that light levels are changing within a shot or sequence of shots, people move from light to shade etc. so lower contrast is preferable because these changes aren't as noticeable

- perhaps photos have always been more contrasty because their traditional output medium (paper, reflected light) is less contrasty than that of video (tv or screen, projected light)

- there is a recent (and in my view, regrettable) trend of doing very low contrast looks. This is either because people's brains have been fooled by looking at log footage for too long and so they don't add "enough" contrast when grading, or because consumer TVs at default settings have their contrast cranked up so much that this is the best way a colourist can get a TV show to look normal lol!

I've already reduced the contrast level in this LUT from the scanner's default level. But perhaps it's not enough. On the one hand I want to keep as much contrast as I can built into the LUT, because this is part of the look that I'm trying to emulate. On the other hand I do recognise that feeling of dismay when you see that the detail in the highlights or shadows is no longer clearly visible in your shot. Perhaps, for video, people prefer to hang onto the dynamic range for as long as possible before making the decision to crush the highlights or shadows. Nevertheless, if the average scene brightness range is around 7 stops (which it is often assumed to be) then a LUT that preserves around seven stops of brightness should be sufficient... on average!

Let's take a look at the contrast levels in this LUT in more detail, as an illustration of what I'm talking about.

Here's a digitally generated 12 stop step wedge for V-Log L. Note the position of middle grey. There are 4 and a bit stops above it (the uppermost stop is the clipping point. Full V-Log would go beyond this). There are 8 stops shown below middle grey. (in reality the eleventh and twelfth bottom-most stops would dissolve into the noise floor)

 

Next, here is the same step wedge with the curve from my LUT applied. Anthing above +2 is pushed into the upper highlights and anything below -2 is pushed into the shadows. The difference between stops is very much expanded around middle grey and compressed in the highlights and shadows But there are still around 7 stops preserved and the transition to highlights and shadows is smooth.

 

However, if the footage is overexposed or underexposed, these exposure "errors" will be amplified by the relatively strong contrast. From left to right: -2 under, correct, +2 over:

 

And here's how that would look on a real world image, in the same order from top to bottom:

 

So with this level of contrast being employed by the LUT, the trick in your colour correcting pipeline is to get that signal within range before applying the LUT, as there's no hope of getting the shadows and highlights back afterwards 🙂

Incidentally, this image provides a good example of my current feelings about contrast (that we shouldn't be afraid of it). If you look at the underexposed image it's clear that the camera captured most of the detail outside the window. A low contrast curve would preserve this detail. But do you need it? I would say no, in this case. So I'm happy for that info to bleed out into white as it does in the "properly" exposed shot, as it's not the subject of the image - the interior decor and the seated figures are.

An analogy that might help in understanding how I feel about contrast is to think of motion picture film. Here, the acquisition stock captures up 15 stops of dynamic range. But the print film stock that it's transferred to holds a fraction of that (and it was up to the "film timer" to decide how long to expose the negative to the print film).

There's something similar going on here. If you can get the brightness of the log signal right then the results should look good.... most of the time. If it's a genuinely high contrast scene, though, like a white cat in sunlight and a black cat in shade, then the contrast of the log signal will need to be reduced.

And actually, this is how I've graded my videos for the last few years, with only three controls (brightness, contrast, colour balance) directly on the log footage and then fed into a contrasty lut.

However, having said all that I think I'm going to go and make lower contrast versions of this LUT. I'd like to see if people thing a lower contrast version is better in how it looks and/or easier to work with.

Thanks for all the feedback so far!

 

[Benjamin Hilton]

Yeah, totally agree with this. It seems to me that about 12 or so stops of dynamic range is needed if you want a high end look in bright environments. More is always nice, but not really necessary. Like you said, there isn't much of a point always being able to see out of the window. Less than 12 stops can definitely be doable, but seems to always be kind of frustrating, always just clipping out something I wanted to see. That was one of the huge things I noticed in the S1 coming from a GH5. I suddenly realized all most all of my documentary scenes weren't clipping any information I actually wanted. Anytime it clipped it was something I'd expect to be clipped, the sun etc. Pretty crazy stuff!

[hyalinejim]

Indeed! In the case of the GH5 there are only four (and a bit!) above middle grey, which is not a lot really. So even though the camera has maybe 11.5 stops in total, most of those are in the shadows. So if you expose normally for a wide dynamic range scene then highlight clipping becomes very apparent.

[kye]

1 hour ago, hyalinejim said:

You're right that it's not designed specifically for the S1 but the difference lies in the colour response rather than the contrast response. If the footage is overexposed by 2 stops then yes it would need to be brought down by 2 stops in order to give a usable image.

Thanks for trying it. It's the contrast level that I'm most interested in getting feedback on, so thank you for that! The reason that it's so contrasty is because it's emulating how a mini lab scans negative film.... and this is quite contrasty because the intended output medium originally was to make prints on photographic paper. What's very interesting to me is that I have found while making profiles for Lightroom that a level of contrast that's acceptable to me there and is similar to Adobe's defaults appears to be far too contrasty when applied to video. I have a few ideas why this might be so, and it might be a combination of factors

- you can get away with more contrast in a photo because it's static, but with video it's likely that light levels are changing within a shot or sequence of shots, people move from light to shade etc. so lower contrast is preferable because these changes aren't as noticeable

- perhaps photos have always been more contrasty because their traditional output medium (paper, reflected light) is less contrasty than that of video (tv or screen, projected light)

- there is a recent (and in my view, regrettable) trend of doing very low contrast looks. This is either because people's brains have been fooled by looking at log footage for too long and so they don't add "enough" contrast when grading, or because consumer TVs at default settings have their contrast cranked up so much that this is the best way a colourist can get a TV show to look normal lol!

I've already reduced the contrast level in this LUT from the scanner's default level. But perhaps it's not enough. On the one hand I want to keep as much contrast as I can built into the LUT, because this is part of the look that I'm trying to emulate. On the other hand I do recognise that feeling of dismay when you see that the detail in the highlights or shadows is no longer clearly visible in your shot. Perhaps, for video, people prefer to hang onto the dynamic range for as long as possible before making the decision to crush the highlights or shadows. Nevertheless, if the average scene brightness range is around 7 stops (which it is often assumed to be) then a LUT that preserves around seven stops of brightness should be sufficient... on average!

Let's take a look at the contrast levels in this LUT in more detail, as an illustration of what I'm talking about.

Here's a digitally generated 12 stop step wedge for V-Log L. Note the position of middle grey. There are 4 and a bit stops above it (the uppermost stop is the clipping point. Full V-Log would go beyond this). There are 8 stops shown below middle grey. (in reality the eleventh and twelfth bottom-most stops would dissolve into the noise floor)

 

Next, here is the same step wedge with the curve from my LUT applied. Anthing above +2 is pushed into the upper highlights and anything below -2 is pushed into the shadows. The difference between stops is very much expanded around middle grey and compressed in the highlights and shadows But there are still around 7 stops preserved and the transition to highlights and shadows is smooth.

 

However, if the footage is overexposed or underexposed, these exposure "errors" will be amplified by the relatively strong contrast. From left to right: -2 under, correct, +2 over:

 

And here's how that would look on a real world image, in the same order from top to bottom:

 

So with this level of contrast being employed by the LUT, the trick in your colour correcting pipeline is to get that signal within range before applying the LUT, as there's no hope of getting the shadows and highlights back afterwards 🙂

Incidentally, this image provides a good example of my current feelings about contrast (that we shouldn't be afraid of it). If you look at the underexposed image it's clear that the camera captured most of the detail outside the window. A low contrast curve would preserve this detail. But do you need it? I would say no, in this case. So I'm happy for that info to bleed out into white as it does in the "properly" exposed shot, as it's not the subject of the image - the interior decor and the seated figures are.

An analogy that might help in understanding how I feel about contrast is to think of motion picture film. Here, the acquisition stock captures up 15 stops of dynamic range. But the print film stock that it's transferred to holds a fraction of that (and it was up to the "film timer" to decide how long to expose the negative to the print film).

There's something similar going on here. If you can get the brightness of the log signal right then the results should look good.... most of the time. If it's a genuinely high contrast scene, though, like a white cat in sunlight and a black cat in shade, then the contrast of the log signal will need to be reduced.

And actually, this is how I've graded my videos for the last few years, with only three controls (brightness, contrast, colour balance) directly on the log footage and then fed into a contrasty lut.

However, having said all that I think I'm going to go and make lower contrast versions of this LUT. I'd like to see if people thing a lower contrast version is better in how it looks and/or easier to work with.

Thanks for all the feedback so far!

 

The film curves that I have seen typically have about 5-6 stops in the "linear" range, then hit the roll-offs pretty hard after that, so your curve seems to be right around what I would expect.

I recall a landscape photographer I watch on YT taking out a medium format camera with an older stock that had something like 5-6 stops in total.  After carefully metering the scene he took a shot of some clouds.  The shot was absolutely amazing - the moodiest clouds I have ever seen!  Just incredible contrast, but looked great because nothing was clipping and the image was thick as hell.

In colour grading I've found a consistent challenge is how to get the 10+ stops of camera DR into the 709 container and not make it look unnatural - the rolloffs on your LUT look pretty nice in that regard 🙂 

[Alexis Fontana]

On 2/13/2022 at 4:55 AM, hyalinejim said:

Let's take one aspect of the LUT which is possibly its most important: hue, saturation and lightness of colour and let's leave the contrast constant and compare the film emulation (bottom) to Panasonic's own V-Log to Rec709 colour transform (top). You might agree with me that one of these looks better than the other in terms of luminance and hue of skintones:

Portra 400 is famous for its attractive skintone reproduction, offering slightly desaturated, even porcelain-like skin which skews more towards tan than pink in terms of hue.

I think this is closer to a photochemical look. 

[Alexis Fontana]

On 2/13/2022 at 4:55 AM, hyalinejim said:

Let's take one aspect of the LUT which is possibly its most important: hue, saturation and lightness of colour and let's leave the contrast constant and compare the film emulation (bottom) to Panasonic's own V-Log to Rec709 colour transform (top). You might agree with me that one of these looks better than the other in terms of luminance and hue of skintones:

Portra 400 is famous for its attractive skintone reproduction, offering slightly desaturated, even porcelain-like skin which skews more towards tan than pink in terms of hue.

I think this is closer to a photochemical look. 

[Alexis Fontana]

On 2/13/2022 at 4:55 AM, hyalinejim said:

Let's take one aspect of the LUT which is possibly its most important: hue, saturation and lightness of colour and let's leave the contrast constant and compare the film emulation (bottom) to Panasonic's own V-Log to Rec709 colour transform (top). You might agree with me that one of these looks better than the other in terms of luminance and hue of skintones:

Portra 400 is famous for its attractive skintone reproduction, offering slightly desaturated, even porcelain-like skin which skews more towards tan than pink in terms of hue.

I think this is closer to a photochemical look. 

[Alexis Fontana]

Woooops. Plz delete copies🥲

[webrunner5]

To my old eyes the top one is the one I would pick. The other 2 look more like log files than final output. Maybe I am missing something here. Not hard to believe I guess.

[hyalinejim]

On 2/18/2022 at 7:43 AM, Alexis Fontana said:

I think this is closer to a photochemical look

Thanks for taking the time to post your edit. However, what I'm trying to achieved here is based on empirical research, rather than memory or imagination. What I mean by that is that instead of creating a look based on how I feel, imagine or remember film should look, I am creating a look by generating data on how film does look under specific circumstances. For this specific LUT the circumstances are:

Portra 400 exposed at +1 (for cleaner shadows)
Scanned on a Noritsu minilab scanner (which introduces its own colour magic)
However, contrast can be variable as there isn't a contrast level that can be considered inherent to a negative, without regard concerns about how it's reproduced (printed on photographic paper of varying contrast levels, or digitised at varying contrast levels)

Let me show how this looks in practice:

 

My question, for now, in this thread is something like "what specific contrast level is appropriate or desirable in a video editing workflow". I've mentioned how the Noritsu's default contrast level looks OK for photos but seems far too contrasty for video. My first version of the LUT is at contrast level -4 in the scanner's settings.

Here's a version that's at -7. Have a go on any V-Log footage and let me know what you think. I think this might be it. Highlight and shadow info is retained but it still has a usably contrasty look.

https://drive.google.com/file/d/1eL65p0Frx5MZFqbVLi41N5S7fXDqN-xU/view?usp=sharing

Firstly this is what we're aiming to emulate. A colour target shot at +1 on Portra 400 and scanned so that middle grey = middle grey:

 

Next, here's the same chart normally exposed on a GH5 in V-Log:

 

Let's add Panasonic's colour conversion and match the contrast. It's way off:

 

Now let's use the colour from my LUT. A very close match:

It's hard to judge these sequentially. But if you open them in new tabs and switch back and forth it's like night and day.

And on real footage the colours from the Portra emulation are so much prettier, as you might expect the colour of one of the most highly regarded colour negative films to be, in comparison to Panasonic's default colour conversion (which admittedly is a bit botched on the GH5).

So I have the colour part down, I feel, and I can use the same technique to emulate any film that I can get my hands on, and at various exposure levels. But I'm still in the process of figuring out how much contrast is "correct" for video while still staying true to the characteristics of the film stock.

What do you guys think about this lower contrast version? I think it might be almost there 🙂

[PannySVHS]

Thank you very much, just by comparing the two charts one can recognize so much! Great work! @hyalinejim Will try it out on my S1 footage.

[hyalinejim]

Thanks @PannySVHS, I'm looking forward to it 🙂 As mentioned, the colour hues won't be as accurate (skintones may be too green perhaps?) but the contrast level will be the same. And with this lower contrast version I think that the extra highlight detail in the S cameras should be preserved and not clipped.

In the meantime, and for those on mobile where it's hard to open tabs, here's a GIF that shows the colour conversion involved from Panasonic's V709 colour to a Kodak colour palette. It's dithered but the colour remains:

 

A few things to note here are how the hue and luminance of various colours are altered in comparison to the default colour:

- Skin becomes lighter and less magenta
- Skies become more azure and darker
- Some foliage becomes darker and less "electric" (Fuji 400H is the king of green foliage, IMO)
- Reds are brighter and warmer
- Yellows are bright and clear

[kye]

A key point that I feel is worth mentioning here is that you can't expect a LUT to grade your film for you.

What I mean by this is that you should be adjusting the image before putting it through the LUT (grading "under" the LUT) and potentially adjusting the overall levels after the LUT too, for more overall adjustments.

Grading under the LUT is simply adjusting the images exposure, WB, contrast, and other attributes, which is the same as adjusting the lighting ratios on set, the exposure of the camera and WB of the scene while shooting on film.

If you don't like the amount of contrast in a LUT, lower the contrast after you apply the LUT and adjust to taste prior to the LUT.  If you want a stronger look then increase the contrast before the LUT and lower it afterwards and you'll get a more pronounced look, or do the opposite.

These are creative tools, not one-click grading machines 🙂 

[kye]

In addition to the above, one thing I saw recently (can't remember where though!) was that someone was shooting film and wanted to have much lower saturation skies instead of bright-blue skies, so they overexposed the film and then developed it darker to normalise the exposure.

This gives much less saturated skies as it pushed all the highlights up into the highlight rolloff where there's much less difference between the R G and B levels, and thus less saturation.  The skies were a pale blue rather than saturated and it gave a kind of look to the skies vaguely reminiscent of the bleach-bypass look.

That's also the kind of thing you can do with a Film Emulation LUT - raise the levels beforehand and then lower them afterwards and see the effects.

[TrueIndigo]

Mostly shooting HLG on the GH5, but I found this test shot I made in VLOG-L some time ago and used the -7 version of your LUT on it. Skin tones look quite good. First of all here is the unchanged image with its RGB parade.