Sorry, another Dragon optimal ISO thread...

[Dominik Bauch]

I'd been rating my Epic dragon at ISO 800, with the skin tone OLPF, on a few past projects.
I know that many people rate the above at 320 but I'd also heard of many big features / TV shows that rate at 800.

As I understand it, an 800 ISO gives you more highlight retention, a smoother roll off.
The other day I did some semi scientific tests where I shot a few scenes at both 800 and 320 ISO, stopping down to compensate so that the results are very identical exposure wise.
I shot outside on a bright day.
I found that the ISO 320 was quite a bit cleaner vs the standard ISO 800 shots, with DEB turned on this difference shrinks a lot.
Highlight retention and roll off is smoother on the 800 footage but this is only noticeable when literally dropping the exposure down to -6.

Granted I didn't test in a Studio lit situation with human skin tones but from what I can tell there is very little difference and so for the best of both worlds I might start trying ISO 500 or 640.
Anyone else have any thoughts? I hear that a lot of people swear by ISO 320 so want to make sure I'm not missing a trick here.

Thanks,

Dominik

 

[Phil Holland]

I shoot mainly between 250-1280. Weapon allows for a bit more, especially with the Low Light Optimized or Standard OLPF.

There's no trick. The key is mostly all about rating your ISO and nailing your exposure. Higher ISO Rating = more noise with more dynamic range allocated to highlight information. Lower ISO Rating = less noise with more dynamic range allocated to shadow information.

Here's some of my tests/notes:
http://www.reduser.net/forum/showthread.php?92999-Red-Dragon-Info-and-Data-Sheets


The "Base ISO", where you have approximately equal usable stops above and below Middle Gray is typically ISO 800, though up to ISO 1280 makes sense as well especially with the LLO and Standard OLPFs.

 

[Terry VerHaar]

OK - this question comes up a lot and it always mystifies me. With Phil weighing in, this may be my chance to put my confusion to bed. LOL

Since ISO is metadata, I am a bit perplexed about the obsession with nailing the exact right ISO while shooting. I understand the shift of the middle greys and the idea of allocating more stops to shadows or highlights and reducing noise or clipping, BUT (!), is it not true that this can be tweaked to your heart's desire in post AS LONG AS YOU DON'T LET YOUR SELECTION OF ISO UNDULY INFLUENCE YOUR OVERALL EXPOSURE?

No matter what, you get a shot at recording approximately 15-16 stops of dynamic range. So, as long as you keep your goal posts from lighting up, you are capturing everything there is to capture. If you let any red in the left post creep up, you are likely to have some noise. If you let the red in the right post creep up, you are threatened with clipping. This is true no matter what ISO you choose, no? Yes, your monitor path will look different as you shift ISO, and your histogram, too. But your goal posts (and stoplights) ignore ISO.

So, I ask: do you gain anything technical at all by futzing with ISO when shooting, assuming you are exposing properly? Have I missed something?

Note: I have been shooting RED since RED One M sensor, then MX, then my Scarlet and then Scarlet Dragon. I have also read everything I could get my hands on from the RED site (and Phil). So, here's hoping to be enlightened, finally.

 

[Alberto Guglielmi]

Terry, my understanding with the ISO as a raw/metadata info is that you can change in post, but it is essential to set it correctly for the proper exposure you mention.
Let's say the light is good for ISO320 and F1.8. That is equivalent to ISO800 and F2.8 and ISO1600 and F4.0 and ISO 3200 and F5.6 (if my calculations are wrong, please forgive me). Your exposure will always be correct and within the range. But if you shoot at ISO3200 and F5.6 you are actually starving the camera from light and the result will be noise even if the ISO is not baked and you can bring the ISO down to 800. In a similar way, if you go the opposite way, you might be burning highlights by flooding light on the sensor.

I am using the LLO on a Epic dragon, and I have been setting my base exposures using ISO 800, then moving it if I need tweaks. I am not sure it is the proper way, but so far I am pleased with results.

 

[Terry VerHaar]

My point is this - you don't change your exposure based on ISO... period. You set your exposure based on your aperture and shutter angle (and frame rate) and the fact that you are neither "in the noise" nor clipping the highlights. Your histogram will help guide you as to whether you have a high key or low key scene.

 

[Michael Tiemann]

OK - this question comes up a lot and it always mystifies me. With Phil weighing in, this may be my chance to put my confusion to bed. LOL

Since ISO is metadata, I am a bit perplexed about the obsession with nailing the exact right ISO while shooting. I understand the shift of the middle greys and the idea of allocating more stops to shadows or highlights and reducing noise or clipping, BUT (!), is it not true that this can be tweaked to your heart's desire in post AS LONG AS YOU DON'T LET YOUR SELECTION OF ISO UNDULY INFLUENCE YOUR OVERALL EXPOSURE?

No matter what, you get a shot at recording approximately 15-16 stops of dynamic range. So, as long as you keep your goal posts from lighting up, you are capturing everything there is to capture. If you let any red in the left post creep up, you are likely to have some noise. If you let the red in the right post creep up, you are threatened with clipping. This is true no matter what ISO you choose, no? Yes, your monitor path will look different as you shift ISO, and your histogram, too. But your goal posts (and stoplights) ignore ISO.

So, I ask: do you gain anything technical at all by futzing with ISO when shooting, assuming you are exposing properly? Have I missed something?

Yes, you are missing a few things. The RED sensor captures linear light data and it has some level of intrinsic noise. The goal posts report pixels that read as noise (on the left) or clipped (on the right). But just because a pixel graduates from "noise" to signal doesn't mean that it's suddenly reading out a perfect signal. It just means it's not reading out as perfect noise. If you've ever looked at a seriously, seriously underexposed image, like 12 stops underexposed, you will see the sea of noise and the occasional bits of signal that make it through. Those bits of signal are by no means perfect, but they are distinguishable as signal from noise. If you shot a test pattern with a random collection of white circles on a black background, you will find that there is an exposure level that lets you accurately count the circles, as ugly as those dots are.

So, down at the bottom range of the exposure, you have signal beginning to escape noise, but just barely. That's the first stop of dynamic range. It may be that you can perceive a signal that's 1% organized in a field of 99% noise. It may be that you can only perceive a signal that's 10% noise in a field of 90% noise. For sake of argument, let's say we both can perceive a signal that's 2.5% in a sea of 97.5% noise. That's life at the bottom of the 16 stop range, and it's not pretty. Four stops up (stop #5 of 16) you have a signal/noise ratio of 40% in a sea of 60% noise. The next stop up (stop #6 of 16) our noise factor is cut in half, meaning our signal increases to 70% vs. 30%. That's getting good. Another four stops up, our noise floor drops four stops, down to 1.125%. Our signal quality at stop #10 of 16 is now really good. And it keeps getting better, right until we clip. So yes, there are 16 stops of dynamic range, as measured by the ability to distinguish signal from noise, but the signal to noise ratio of the first 5 stops are nothing like the signal to noise ratio of the next 10 stops.

Making matters worse, consider what happens when, for whatever reason, a pixel reads (or misses) a few thousand photons in a given read cycle. On a percentage basis that might cause a read error of a stop or two down at the low end, or it might be a fraction of a percent (which is less than 1/70th of a stop) up toward the higher end of the range. So not only is intrinsic noise a beast, but tolerance for error at the dark end of the stop range is also very seriously compromised.

For these reasons, one should never simply say "any 10 stops across this 16 stop range is as good as any other 10 stops". It's just not mathematically true. And that's not RED's fault. It's not even anything they can fix. It's math.

 

[Daniel Reed]

...one should never simply say "any 10 stops across this 16 stop range is as good as any other 10 stops". It's just not mathematically true. And that's not RED's fault. It's not even anything they can fix. It's math.

Interesting point Michael. In this theory, exposing an unclipped shot that GIO registers falling w/in a range from stop 6 to stop 16, will be cleaner then a shot exposed to land between a range of stop 3 and stop 13? Isn't this the basis of the ETTR philosophy?

 

[Michael Tiemann]

Interesting point Michael. In this theory, exposing an unclipped shot that GIO registers falling w/in a range from stop 6 to stop 16, will be cleaner then a shot exposed to land between a range of stop 3 and stop 13? Isn't this the basis of the ETTR philosophy?

Yes: https://en.wikipedia.org/wiki/Exposing_to_the_right

Now, there are some who like a little texture (aka noise) in their image. They know who they are. More importantly, they know what they are doing. And they can make damn good images, too.

 

[Dominik Bauch]

Thanks Phil but to someone like you who has worked with Dragon more than most (!) which ISO (when perfectly exposed) gets the most filmic image in your opinion?
Is it the cleaner (and possibly sharper) lower ISO ranges or the better highlight retention of ISO 800 with that bit of texture?

 

[Nick Gardner]

It's also about consistency. You expose the in the same range, at the same ISO, and all your shots have consistent noise characteristics. I don't mind jumping from 800 to 1600 in different scenes, but from shot to shot in the same scene, they should match.

Nick

 

[Phil Holland]

Thanks Phil but to someone like you who has worked with Dragon more than most (!) which ISO (when perfectly exposed) gets the most filmic image in your opinion?
Is it the cleaner (and possibly sharper) lower ISO ranges or the better highlight retention of ISO 800 with that bit of texture?

If you are trying to get something similar to the texture of film, say ASA 500 5219 Vision, you likely want to be shooting ISO 500-800 with the Skin Tone - Highlight OLPF. Maybe even a bit beyond ISO 800. Lower sensitivity stock is cleaner of course. It's not something there's truly one answer for.

Film itself has a maximum dynamic range of 13-15 stops depending on various factors. At ISO 250 you have approximately 7 stops of highlight latitude, but since film is a photochemical medium you might want to error on having more highlight retention at a higher ISO. On Epic Dragon I really enjoyed ISO 500 with the STH OLPF. Weapon I've been lately shooting with the Standard OLPF around ISO 800-1280, which I feel gives me the most flexible film-like image.

I'm more or less doing the same thing that DPs are doing on notable RED features when it comes down to it. It's more about decision making and staying the course. There's certainly times where you'll want a cleaner image. Many here have discovered what you can "get away with" in daylight conditions at higher ISOs as well because there's so much fill typically.

I mostly keep my ISO consistent across the project or at least scene by scene.

 

[Brett Harrison]

As others have said - from reading Graeme Nattress's comments on here, ISO is just a 'grade' of the material at the debayer level, and Graeme has said that F.L.U.T., the Floating point LUT* system is actually the basis of RED's colour science ISO and all) - i.e. he's literally spelling out (it's in his RU signature) that ISO is just the result of passing the linear sensor data through a LUT.

That LUT looks to map the gamut of REDs camera to the exposure choice you're indicating by selecting a FLUT, which is a floating point position between ISOs (FLUT is the basis of RED's ISO system, there's no difference). For low exposures it will desaturate, for high exposures it will saturate (not this simple but choose a low FLUT and then grade it up with LGG then see what it looks like compared so an overexposed FLUT choice graded down)

Graeme has also said that RLF is the first step in arriving at RG4, which is a gamma curve designed to transform the RLF curve - to my knowledge there's no gamut mapping (alteration of colour), it's just contrast/saturation.

So all this is to say that ISO is an estimation tool based on the old system of a characteristic curve, which film has by its nature (there's no way to shoot linear film) whereas CMOS sensors don't need to fuss with that concept, but rather need to talk about clipping and noise floor.

A lot of DPs have said that there's no right exposure, just a range of exposures that would work for a range of artistic intents. Each framing of a scene has a unique contrast range and unique distribution of reflectance information. Middle grey is statistically the 'middlest' but only statistically, not on a case by case basis. We don't need to rely on that statistical rule or spot meter becuse the camera is telling us all this - it's literally a light meter with a display.

So there's no such thing as an ISO rating you can give to a CMOS unless you're willing to stick to a given characteristic curve (contrast curve, ISO rating) for all time. If you always shoot at ISO 800 you can think of that curve as the film characteristic curve that you're stuck with, but then when you get to the DI you will perhaps decide on another exposure choice and realise that a different amount of light hitting the sensor would have been better. That's why as others have said above, it's better to mind the stop lights, goal posts, take a look at various ISO ratings of your scene to get an idea of how you might be able to work with the footage in the grade.

*This isn't an actual LUT to my knowledge, but actually a mathematical formula working in floating point.

 

[David Battistella]

You need to look at ISO as a rating or number, FROM WHICH you base exposure decisions.

Two things about what is going on with you findings.

You say the images are cleaner when rating at 320 iso, well, simply put when you rate at 320 you open up and allow MORE LIGHT to hit the sensor, this is always going to give you a cleaner look. Tradeoff=less highlight lattitude and rolloff.

Much of this depens on the DR of your scene and what you want to pull out of it.

Filmic is not one ISO setting. It sorta means replicating DR in a wide range of scenes. That is the work of a cinematographer, matching that filmic DR across various scenes and exposure situations. Thats why in digital there is a working range (like phil describes) and that is the best strategy to have. You need to start looking at scenes and determining the best ISO within the range to get the DR you want out of that scene, so there is no one ISO that is going to do the trick.

But you can use ISO as a "basis" for an exposure decision.

With a 500ISO filmstock, other things like filtration and lighting are needed to accomplish this. You wouldnt just rate the 500ISO film at 1600 iso without special instructions and without understanding the tradeoffs and you would be using a lot of fill in some scenes or ND in others to flatten out the DR to create acceptable images.

With Digital sensors the penalty is no where near as high with ISO rating.

Phil is spot on with describing a working range. It's your job to begin to dive in and understand when and how to shift your rating.

David

Just as an example. This was "rated" at 1000ISO, why? to see into the shadows better. if this was rated at 320 ISO I might not be able to see into the shadows and preseve the bits of blue sky and noise (even in a complex foliage scene) is minimal or non existent.

Now this fits the profile of exterior, Day, low ISO etc, but this shot called for a higher rating (or put another way underexposing) for the desire result.

Here is another clear example of where you can easily rate the STH in daylight at 1000ISO

 

[Ulf Krentz]

Terry, I think you are spot on. With RED ISO is just metadata and no matter where it is set to it only depends on the amount of light hitting the sensor. I am not entirely sure you´re recording straight linear data though. It has been my understanding that the tone / data mapping of the OLPF settings comes into play somehow. That said I try to rate the Skintone at 400-500ASA and the LowLight at 800ASA as I feel this is the most pleasing image for me. Coming from reversal film all this discussion about dynamic range seems strange, RED is so forgiving

Cheers
Ulf

 

[David Battistella]

Coming from reversal film all this discussion about dynamic range seems strange, RED is so forgiving

Cheers
Ulf

Bingo. Not much room for error in reversal film at all. Really have to nail exposure.

David

 

[Hrvoje Simic]

I shoot mainly between 250-1280. Weapon allows for a bit more, especially with the Low Light Optimized or Standard OLPF.

There's no trick. The key is mostly all about rating your ISO and nailing your exposure. Higher ISO Rating = more noise with more dynamic range allocated to highlight information. Lower ISO Rating = less noise with more dynamic range allocated to shadow information.

Here's some of my tests/notes:
http://www.reduser.net/forum/showthread.php?92999-Red-Dragon-Info-and-Data-Sheets


The "Base ISO", where you have approximately equal usable stops above and below Middle Gray is typically ISO 800, though up to ISO 1280 makes sense as well especially with the LLO and Standard OLPFs.

The most relevant part is nicely summed up here by Phil and most can be understood by starting from this and doing own tests in different light scenarios which affect camera noise characteristics, night exteriors especially.

 

[Alberto Guglielmi]

My point is this - you don't change your exposure based on ISO... period. You set your exposure based on your aperture and shutter angle (and frame rate) and the fact that you are neither "in the noise" nor clipping the highlights. Your histogram will help guide you as to whether you have a high key or low key scene.

The ISO is actually a variable of the exposure in the digital world. It was also in the film world, but you would not change film as easily. I use to have two cameras with different films in, and pick one based on the lighting we had available.

Your exposure is a function of the ISO you select at the beginning of the project/day/scene/take. Whether that was because you had a mechanical camera with film rated at a certain ISO, or a digital camera with changeable OLPF to give you too many options, that stays the same. So ISO is an essential part of calculating exposure as much as scene available lighting. Both are variables you can (or cannot) intervene on depending on the situation. Since digital came in, ISO became a variable we can change easily, but it is not always proper to do so.
Even if ISO is a metadata, the limitation of pushing it still stay, as very well explained by Michael: "one should never simply say "any 10 stops across this 16 stop range is as good as any other 10 stops"

PS I do not recommend changing ISO every scene or take...

 

[Tarquin Cardona]

I think a key take away from a lot of the technical expertise in these comments can be summed up in the realization that these cameras are here to do your bidding. Not the other way around. IMO since we moved away from film a lot of us have become much more timid about bending the notes so to speak, and trying to find some 'sweet spot'. You can go from silky smooth to Man On Fire punch by manipulating how you want to play with this tool. So there is no single film look answer.

Most often I believe the ISO question is really asking a simpler question about how far can I push the camera before I get noisy pictures. This is probably most critical to people who work in situations with less luxury to control their surroundings. I do a lot of work in pediatric documentary. And I can't be in medical critical situations asking people to please stop what they are doing so I can stick a light in the OR. So in that case, knowing the ISO capability and things like mixed light handling are good questions to ask. I can say from my experience that the DSMC2 with Standard goes up to 1280 with zero issue. ZERO. That is pretty cool from where we've been. With fast glass there is almost no situation that can't be managed - and the results are great and 'filmic' because Red color is so good.

 

[Ulf Krentz]

The ISO is actually a variable of the exposure in the digital world. It was also in the film world, but you would not change film as easily. I use to have two cameras with different films in, and pick one based on the lighting we had available.

Your exposure is a function of the ISO you select at the beginning of the project/day/scene/take. Whether that was because you had a mechanical camera with film rated at a certain ISO, or a digital camera with changeable OLPF to give you too many options, that stays the same. So ISO is an essential part of calculating exposure as much as scene available lighting. Both are variables you can (or cannot) intervene on depending on the situation. Since digital came in, ISO became a variable we can change easily, but it is not always proper to do so.
Even if ISO is a metadata, the limitation of pushing it still stay, as very well explained by Michael: "one should never simply say "any 10 stops across this 16 stop range is as good as any other 10 stops"

PS I do not recommend changing ISO every scene or take...

Simply no, NO! Please do your research.

ISO on RED is in no way effecting the recorded data. If you try a comparative in the analog world you would have two cameras with the same film stock and one cameras film will be push developed in the lab, and guess what, it´s going to result in more grain ;-) Hell, I´m wondering why this is so hard to understand.

Cheers
Ulf

 

[Gerrit Piechowski]

I think it definitely depends on your OLPF... for the Dragon+Skintone for example 320ISO should be best. I've had a shoot a while ago where we were able to try out all the settings since we didn't have enough light to cover the whole range of our 60m long set.

On the first day we shot most of it on ISO800. Later that day I played around with the footage in DaVinci and it wasn't as clean as I expected it to be. So the next day we changed our "strategy" and shot everything on Native ISO 320. Even though it looked too dark on the screen we were able to pump up the ISO in post without problems up to 1600 - clean image.

I'm also with Terry on this, never expose your image with ISO.