CINESITE: A Kodak Company

[M Most]

Mike,

whatever but one or the goals is to get RED1MX/Epic/Scarlet image look as closer as possible to many of Kodak Color Negative Film and Print Film looks.

You're just not going to give up with this nonsense, but I am. Think whatever you want.

 

[Francesco Bonomo]

Film look is not so much about a grain

I guess I've used too many words...I just like the right amount of film grain, that's all. Sounds better?

 

[Sanjin Jukic]

You're just not going to give up with this nonsense, but I am. Think whatever you want.

A main part of that "nonsense" is to give an idea about eventual future cooperation between RED, Kodak and Fuji

in a way to find best option that RED Epic/Scarlet footage look closer as possible to the famous film stocks from both film stock maker companies.

 

[Julio Quintana]

Film look is not so much about a grain and is more about the specific look that film has an ability to

reproduce artwork with the look and feel of an original with accurate color reproduction.

I don't actually agree that film reproduces images more accurately than digital. In fact, I don't think our love for film has as much to do with accuracy as it does with familiarity. We don't see film grain in real life, and we don't see 24p motion blur in real life. If it were about accuracy, we would all love IMAX 60p footage, but tests show that most people don't. Even though it's arguably the highest quality image reproduction available right now, it feels like "video" because we associate the clarity of motion with sports or news.
My point is that there will come a time when we associate the ultra clear look of digital with our favorite movies, and 35mm film will look dated (as in, originating at a specific time from the past).

 

[Francesco Bonomo]

My point is that there will come a time when we associate the ultra clear look of digital with our favorite movies, and 35mm film will look dated (as in, originating at a specific time from the past).

Julio, I think I understand what you're saying, but aren't you leaving something like a hundred years of terrific films out of the equation? I have no problem believing that the future of filmmaking, whenever that will be, will be 90% or even 100% digital, and while I do agree with you that every movie is, so to speak, "son of its times" (meaning it reflects the styles, approaches, and even technologies that existed at the time of its making), I doubt we'll ever associate a specific "look" (i.e. ultra-clean digital or grainy film) to the reason we like one movie better than another. After all, and again I'm speaking based on my personal preferences, some of my favourite movies were made 30, 40 and 50 years ago. Cameras, sensors, film stocks, lenses, they're all tools...most masterpieces in filmmaking history were shot on film, I'm sure 100 years from now we'll have digital masterpieces as well, but they will be considered so because of how the filmmakers will use the tools available to them to tell a great story, not because the images look "grainy" or "clean".

Just my 2€c

 

[Sanjin Jukic]

RED already has got with EPIC the resolution and lightweight form factor digital cine camera wars.

But the battle for image look is still ongoing between the main digital cine acquisition systems...

...(it's about a look that should be closer to the famous film stocks as possible).

 

[Gunleik Groven]

I don't actually agree that film reproduces images more accurately than digital. In fact, I don't think our love for film has as much to do with accuracy as it does with familiarity.

This is true, but there are a couple of more issues in there...

Analogue media (like film for images and tape+ air in a recording situation) compresses the dynamic range a lot "on the fly", while digital mapping of level is fixed by nature.

This gives analogue media the ability to roll of peaks in the top, while still accentuating the lower signals bu nature, while in a digital world the low end suffers earlier on a SN problem in the low end, and peakes and spikes just overdrive/clip when you reach max level.

This is not very unlike how our eyes and ears work,so it is more within the same idiom, for lack of better words. Linearity is not how we see and hear.

In addition comes that we have more sensitivity (are more able to discern small nuances) in the lower/low intensity ranges than in the high, while in digital the precission is better at the top of the curves. Which is why we are so sensitive to noise/grain in the darks...

This doesn't mean that I by any means think digital is a bad idea or cannot give "as good" or "better" results than analogue. I am in fact a very digitally tilted person, but this is good to know.

The solution is to get digital formats with enough headroom in both ends for them not to break on spikes of high intensity or in areas with low intensity.

To achieve that, you need high bit depths... Or as RED has started to work out with the Epic/Scarlet: Multiple exposures/samples of the same frame

But, to take the "high bit depth" approach first.

The main advantage with this, is that you can throw away low-precission areas of the file, to get better precission in the lowlights.

Let's use our beloved 12 bit RED RAW as an example.

With each stop being available from the sensor being mapped to a new bit, that'd give us a theoretical 12 stops of dynamics.

The only issue is that the lowest 8 stops would lack a lot in precission (max out at 256 level pr channel) while the top 4 stops would have most of the precission (3840 levels per channel).

This is a bit unfair to the lowlights and skintones compared to the highlights and sky. This is also the background for the "expose to the right under low contrast" idiom. You "save" your image to a part of the file with higher precission...

There are many ways to solve this, which Graeme probably could write books about (and as likely allready has, in his notes), but the main thing is that "more bits" are good, because you can throw out more lower bits when one map the signal in the AD process (a sensor is NOT digital...) and still get better precission and map more dynamics into the file over all.

We like that.

And as long as the recording doesn'tclip/go black - it is quite rudimentary to re-map the recorded image into even an 8-bit 422 image for low-contrast displays like TV's or computerscreens. The difference between max and min light intensity on these screens are so narrow that it still looks "good"... as long it isn't broken in the first place...

Another approach which has been discussed a bit here, and I'd like to do some tests on the comming year, is to use analogue filtering more emulating "filmresponse" (in addition to the OLPF). I am not sure really how this will work, but it sounds reasonable that one could extend the DR a bit that way, as you can analoguly roll off the peaks a bit. But I have not tested it on the RED yet, and I do see some catches with this approach. It wouldn't be for all situations, but could - if it works - be a nice addition to the toolset of handling dynamics in an image...

Cheers!

G

 

[Elsie N]

This is true, but there are a couple of more issues in there...

Analogue media (like film for images and tape+ air in a recording situation) compresses the dynamic range a lot "on the fly", while digital mapping of level is fixed by nature.

This gives analogue media the ability to roll of peaks in the top, while still accentuating the lower signals bu nature, while in a digital world the low end suffers earlier on a SN problem in the low end, and peakes and spikes just overdrive/clip when you reach max level.

This is not very unlike how our eyes and ears work,so it is more within the same idiom, for lack of better words. Linearity is not how we see and hear.

In addition comes that we have more sensitivity (are more able to discern small nuances) in the lower/low intensity ranges than in the high, while in digital the precission is better at the top of the curves. Which is why we are so sensitive to noise/grain in the darks...

This doesn't mean that I by any means think digital is a bad idea or cannot give "as good" or "better" results than analogue. I am in fact a very digitally tilted person, but this is good to know.

The solution is to get digital formats with enough headroom in both ends for them not to break on spikes of high intensity or in areas with low intensity.

To achieve that, you need high bit depths... Or as RED has started to work out with the Epic/Scarlet: Multiple exposures/samples of the same frame

But, to take the "high bit depth" approach first.

The main advantage with this, is that you can throw away low-precission areas of the file, to get better precission in the lowlights.

Let's use our beloved 12 bit RED RAW as an example.

With each stop being available from the sensor being mapped to a new bit, that'd give us a theoretical 12 stops of dynamics.

The only issue is that the lowest 8 stops would lack a lot in precission (max out at 256 level pr channel) while the top 4 stops would have most of the precission (3840 levels per channel).

This is a bit unfair to the lowlights and skintones compared to the highlights and sky. This is also the background for the "expose to the right under low contrast" idiom. You "save" your image to a part of the file with higher precission...

There are many ways to solve this, which Graeme probably could write books about (and as likely allready has, in his notes), but the main thing is that "more bits" are good, because you can throw out more lower bits when one map the signal in the AD process (a sensor is NOT digital...) and still get better precission and map more dynamics into the file over all.

We like that.

And as long as the recording doesn'tclip/go black - it is quite rudimentary to re-map the recorded image into even an 8-bit 422 image for low-contrast displays like TV's or computerscreens. The difference between max and min light intensity on these screens are so narrow that it still looks "good"... as long it isn't broken in the first place...

Another approach which has been discussed a bit here, and I'd like to do some tests on the comming year, is to use analogue filtering more emulating "filmresponse" (in addition to the OLPF). I am not sure really how this will work, but it sounds reasonable that one could extend the DR a bit that way, as you can analoguly roll off the peaks a bit. But I have not tested it on the RED yet, and I do see some catches with this approach. It wouldn't be for all situations, but could - if it works - be a nice addition to the toolset of handling dynamics in an image...

Cheers!

G

Interesting post.

Gunleik, along with some others, you are a favorite poster of mine. You are among the few here with the communication skills that can put a subject into language that can be understood by the layman as well as the professional.

Thanks for your time.

 

[M Most]

Gunleik:

I didn't want to quote your entire post, but as you probably know already, log formats exist specifically to handle the very situation you're describing. In fact, Redlog is specifically designed to provide full precision for the lower 8 stops and reduced precision for the upper 4. This allows the original 12 bits to be represented comfortably in a 10 bit container. You and I both know that many - if not most - people here don't understand the actual purpose of log coding or how to use it. But it's primary purpose is to handle exactly what you're talking about.

 

[Gunleik Groven]

I know Mike!

But as you and I also know, the mapping of light before encoding is linear.

Graeme has tried to explain a couple of times that one cannot "record log", but encode log from a linear file...

I have just tended to accept that on face value.

But lets say we want a 16 stop recording with linear light mapping before log-encoding, we'd still need more bits.

As a BTW, I know ways to get "around" the linear limit, but that's another story and would involve stuff RED has declared that they despise! )

Cheers and thanks for the clarification!