Emulating the texture of film

[kye]

For some time I've been thinking about the texture of film.  I've also been thinking about the texture of RAW images, both 4K and also 1080p.  And I've been thinking of the texture of low-bitrate cheap digital camera images, and how much I don't like it.

Last night I watched Knives Out, which was very entertaining, but of note was that it was shot by Steve Yedlin ASC, and that it was shot in 2.8K RAW and mastered in 2K.

For those that aren't aware, Steve Yedlin is basically a genius, and his website takes on all the good topics like sensor size, colour science, resolution, and others, and does so with A/B testing, logic and actual math.  If someone disagrees with Steve, their work is cut out in convincing me that they know something Steve doesn't!

This inspired me to do some tests on processing images with the goal being to create a nice timeless texture.  Film has a nice analog but very imperfect feel with grain (both the random noise grain but also grain size of the film itself which controls resolution).  Highly-compressed images from cheap cameras have a cheap and nasty texture, often called digititis, and is to be avoided where possible.  RAW images don't feel analog, but they don't feel digital in digititis way either.  They're somewhere in-between, but in a super clean direction rather than having distortions, with film having film grain which isn't always viewed as a negative distortion, and highly-compressed digital having compression artefacts which are always viewed as a negative distortion.

Here's the first test, which is based on taking a few random still images from the net and adding various blur and grain to see what we can do to change the texture of them.  The images are 4-7K and offer varying levels of sharpness.  

The processing was a simple Gaussian Blur in Resolve, at 0.1 / 0.15 / 0.2 settings, and adding film grain to kind of match.

On the export file the 0.1 blur does basically nothing, the 0.15 blur is a little heavy handed, and the 0.2 looks like 8mm film, so very stylised!  

The video starts with each image zoomed in significantly, both so that you can see the original resolution in the file, but also so that you can get a sense of how having extra resolution (by including more of the source file in the frame) changes the aesthetic.  Interestingly, most of the images look quite analog when zoomed in a lot, which may be as much to do with the lens resolution and artefacts being exposed as it has to do with the resolution of the file itself.

My impression of the zooming test is that the images start looking very retro (at 5X all their flaws are exposed) but transition to a very clean and digital aesthetic.  The 0.15 blur seems to take that impression away, and with the film grain added it almost looks like an optical pull-out on film was shot of a printed photograph.  In a sense they start looking very analog and at some point the blur I'm applying becomes the limiting factor and so the image doesn't progress beyond a certain level of 'digitalness'.

In the sections where I faded between the processed and unprocessed image I found it interesting that the digitalness doesn't kick in until quite late in the fade, which shows the impact of blurring the image and putting it on top of the unprocessed image, which is an alternate approach to blurring the source image directly.  I think both are interesting strategies that can be used.

Now obviously I still need to do tests on footage I have shot, considering that I have footage across a range of cameras, including XC10 4K, GH5 4K, GH5 1080p, GoPro 1080p, iPhone 4K, and others.  That'll be a future test, but I've played in this space before, trying to blur away sharpening/compression artefacts.  

There are limits to what you can do to 'clean up' a compressed file, but depending on how much you are willing to degrade the IQ, much is possible.  

For example, here are the graded and ungraded versions of the film I shot for the EOSHD cheap camera challenge 18 months ago.  These were shot on the mighty Fujifilm J20 in glorious 640x480, or as I prefer to call it 0.6K....

IIRC someone even commented on the nice highlight rolloff that the video had.  All credit goes to the Fuji colour science  😂😂😂

Obviously I pulled out all the stops on that one, but it shows what is possible, and adding blur and grain was a huge part of what improved the image from what is certain to be several orders of magnitude worse than what anyone is working with these days, unless you're making a film using 90s security camera footage or something.

[androidlad]

https://videovillage.co/filmbox/

[josdr]

Fuji colour science is great and since I recently got the Emotive colour luts, I concur that a Gaussian blur between 0.18-0.20 seems best. The camera's colour science and sensor has to help you of course.  Fuji  has very organic footage . People talk about BM raw and its highlight retention etc but I find it plasticky next to Fuji colour science and motion cadence . Then again youtubers like sliders and everything bathed in bright light......

 

PS Try some night footage at 3200 iso... 

[dreamplayhouse]

I haven't seen any footage from a hybrid camera that can match the filmic look of BM. Yes Fuji's have nice color but BM just has the most organic look. Plasticky?? nope.

[TomTheDP]

5 hours ago, dreamplayhouse said:

I haven't seen any footage from a hybrid camera that can match the filmic look of BM. Yes Fuji's have nice color but BM just has the most organic look. Plasticky?? nope.

The 12 bit RAW makes a difference. Color holds up better. Now 12 bit Prores raw from the S1 should be just as organic looking tho.

[BenEricson]

5 hours ago, TomTheDP said:

The 12 bit RAW makes a difference. Color holds up better. Now 12 bit Prores raw from the S1 should be just as organic looking tho.

I have to disagree. In my experience, even the ProRes HQ offers a very unique and pleasing unprocessed feeling image. IMO a lot of the magic is in the highlights.

[greenscreen]

ETTR.

[TomTheDP]

16 hours ago, BenEricson said:

I have to disagree. In my experience, even the ProRes HQ offers a very unique and pleasing unprocessed feeling image. IMO a lot of the magic is in the highlights.

Blackmagic probably has better highlight roll off then. Though highlight roll off can be controlled in post as long as the dynamic range is there. 

[kye]

6 hours ago, TomTheDP said:

Blackmagic probably has better highlight roll off then. Though highlight roll off can be controlled in post as long as the dynamic range is there. 

Every camera has great rolloff if the DR is there, just apply a curve.  If I can get a compliment on the highlight rolloff from a budget P&S camera from 2009 then there are no excuses!

It's also possible to get a nice rolloff on areas that are clipped, although you don't get any detail back from them obviously.  

The more I learn about colour grading, the more that I realise that grading is in the same category as production design or lighting, you can't expect your footage to look great if you don't do any set dressing, hair or makeup, or just shoot with whatever light happens to be there when you happen to show up, so why would it look great if you didn't do any colour grading either?

[BenEricson]

6 hours ago, kye said:

Every camera has great rolloff if the DR is there, just apply a curve.  If I can get a compliment on the highlight rolloff from a budget P&S camera from 2009 then there are no excuses!

It's also possible to get a nice rolloff on areas that are clipped, although you don't get any detail back from them obviously.  

The more I learn about colour grading, the more that I realise that grading is in the same category as production design or lighting, you can't expect your footage to look great if you don't do any set dressing, hair or makeup, or just shoot with whatever light happens to be there when you happen to show up, so why would it look great if you didn't do any colour grading either?

You are correct in your research.

Let me rephrase what I said. I believe that if you're after that overall softer / filmic look, the bmpcc will get you there easier and quicker. There's no sense shooting with a GH5 if you can get that look with less work. A colorist friend of mine often says, "the least amount of moves wins.

[kye]

4 hours ago, BenEricson said:

You are correct in your research.

Let me rephrase what I said. I believe that if you're after that overall softer / filmic look, the bmpcc will get you there easier and quicker. There's no sense shooting with a GH5 if you can get that look with less work. A colorist friend of mine often says, "the least amount of moves wins.

I definitely agree with the sentiment "the least moves wins", although I would refine that considerably.

I used to find myself basically fighting the footage, unable to get the footage to look even half-way decent.  This was because my skill level was basically zero at that point.  As I gradually learned, I started being able to make a few good adjustments before the 'tweaking' started doing more harm than good.  Now, I think of it as a skill-level thing - the more skill you have the more adjustments you can make before you're making it worse and not better.

My new workflow is now to apply a look (Kodak 2393 LUT with some blurring/texture as shown above is my current favourite but I'm adjusting and optimising over time) and then to WB, adjust levels, and do any localised corrections required, but all underneath the look.  The secondaries are normally Hue-v-Hue / Hue-v-Lum / Hue-v-Sat but I will also do local adjustments if required.  These are often to match shots rather than significantly push things around.  I'm often doing things like darkening / desaturating distracting objects in the background or doing a large and soft power-window to brighten the subject.  

I regard this as being quite minimal, considering that the 2393 is by far the largest adjustment in the grade and it was created by people far above my skill level, so I'm still responsible for the minority of the grade.  With my control surface I'm able to rip through an edit only spending a short amount of time per shot, by only making a few relatively repeatable adjustments.

[BjornT]

On 12/31/2020 at 2:10 PM, kye said:

I definitely agree with the sentiment "the least moves wins", although I would refine that considerably.

I used to find myself basically fighting the footage, unable to get the footage to look even half-way decent.  This was because my skill level was basically zero at that point.  As I gradually learned, I started being able to make a few good adjustments before the 'tweaking' started doing more harm than good.  Now, I think of it as a skill-level thing - the more skill you have the more adjustments you can make before you're making it worse and not better.

My new workflow is now to apply a look (Kodak 2393 LUT with some blurring/texture as shown above is my current favourite but I'm adjusting and optimising over time) and then to WB, adjust levels, and do any localised corrections required, but all underneath the look.  The secondaries are normally Hue-v-Hue / Hue-v-Lum / Hue-v-Sat but I will also do local adjustments if required.  These are often to match shots rather than significantly push things around.  I'm often doing things like darkening / desaturating distracting objects in the background or doing a large and soft power-window to brighten the subject.  

I regard this as being quite minimal, considering that the 2393 is by far the largest adjustment in the grade and it was created by people far above my skill level, so I'm still responsible for the minority of the grade.  With my control surface I'm able to rip through an edit only spending a short amount of time per shot, by only making a few relatively repeatable adjustments.

I love the texture of film and always try to get some aspect of into all of my projects. I think the camera matters less these days as almost all the dslrs out there can give you a filmic image if one bothers to learn resolve and a bit of color grading. I used the GH5 for years and one can get beautiful images from that camera. Adding an OLPF effect in post with some blur helps getting that digital sharpness down that many DSLR sensors have. There are also so many good resources out there now for getting a filmic look without much work. Juan Melara's Kodak 2393 FM Powergrade is great for getting filmic colors and if you want halation, grain and colors that are as close to film as possible I think Dehancer has that down, very impressive plugin. Also Tom Bolles is releasing Cineprint 16 powergrade for sale soon (video below was shot on GH5). and VFX artist Miguel Santana has Spectra coming out next year and they both look great. Learning some color grading and buying the powergrades/plugins of those that truly understand film emulation will give you better results than buying a blackmagic pocket 4K and slapping their standard lut on. Once you tailor your grade a bit to your camera just save it as a powergrade to use on future projects and then it is as easy as adding a lut to get a good starting point on any project.

)

[Mr. Freeze]

This Emulation really looks good. A bit too much saturation for my taste, but I like the blooming diffusion and the grain in the example clips.

[Caleb Genheimer]

I’ve been working out my own methods on this stuff for quite some time now, and as someone who also shoots some actual film here and there, some of the extreme minutiae are maddening to mail down, and prompted me to move into Resolve’s Fusion tab for many operations. It has been like learning to ride a bike all over again, and I’m not there yet, but Fusion is absolutely  powerful enough to get me where I want to go eventually. There is complex spatial and temporal interplay (especially with overscanned footage) that simply can’t be done easily with power grades and edit page compositing. I guess do 99% of people even notice? Probably not, but we do, right?

Im interested to see how filmbox is implemented, as well as Tom Bolles power grade (which for that approach looks to provide a lot of nuanced manipulations resulting in the most convincing Color tab results I’ve seen so far). I will probably grab Cineprint 16 just to see if I can replicate the various nodes in Fusion alongside my own spatial machinations.

Digging deep into this stuff at all kills render times like crazy, but that’s maybe half the fun, right?

I still think halation could be brought 80-90% of the way there in the physical world with a bespoke filter from some expert like Tiffen or Schneider. Film has a set tonal dynamic range, which honestly a number of digital cameras get close enough to these days as to not make much difference, but spatially especially with regards to halation, the dynamic range in film’s reaction is much higher. Basically, something can be blown out, but two blown out sources of different luminance will still show a difference in the amount and reach of halation, and that reaction is still way outside the dynamic ability of digital capture... but a physical filter I believe could be constructed to react with that nuance and that could be subsequently baked into digital capture. Tiffen’s glimmerglass (of which there is a “bronze” variant already) looks to my eyes to be a prime candidate. If that could be tweaked to be red (or add red), it would to my eyes go a long way to achieving that nuance, and a bit of digital halation would be the added tweak to polish it off to taste.

It’s super fun to build film effects, but at the end of the day, it’s also not all that difficult to just shoot film on certain projects, and in that case, the look is just... there. I’d encourage anyone who is spending significant time on emulation to also put some solid time into researching 16mm cameras, and doing a bit of hunting until you find a kit that fits your use-case. SR3s and Aatons are nice cameras for sure, but there are plenty of other very capable cameras out there, and you can still find good deals if you are patient.