Underexposure as a Tool

[AJ Young]

Apologies if I've placed this topic in the wrong message board; I couldn't decide if this fit in a specific category.

 

Last summer I shot a feature that took place in the 1960's. Originally, the production wanted to shoot on film, but budget prohibited it and we decided to shoot it on the Alexa. We wanted film because, naturally, we wanted the image to look and feel like a 1960's movie; we were interested in the idea of degrading the organic image optically to make the movie look like an old film reel you would find in a grandparent's attic.

 

Needless to say, we still wanted to emulate old film with the digital camera. Through my research, I discovered film's propensity to have excellent highlight detail while digital naturally has more shadow detail. Colleagues and mentors recommended I look into shifting the dynamic range of the camera.

 

Sources of inspiration for shifting the dynamic range came from Martha Marcy May Marlene and Birth. I dug deep into the archives to find this old thread talking about Birth's process. LINK

 

Long story short, I decided to mimic the methods by Lipes and Savides to shift my dynamic range into the highlights utilizing ISO and underexposure (ISO essentially meaning re-interpreting the data). During testing, I ran multiple exposures at various ISO's and landed on 1600 ISO, with a further two stops underexposed. I detailed the testing here.

 

1600 ISO -2 stops effectively underexposes the image by 3 stops from the Alexa's base ISO of 800. My idea was to place my exposures in the two of the curve save as much highlight detail as possible while "milking" my shadows.

 

I shot the entire feature 3 stops under exposed, mostly in ProRes 4444 XQ and the occasional shot in ARRI Raw (sunsets, tricky color balance situations, etc). I mostly did incident metering with my meter rated at 6400 ISO.

 

Here are the results as JPG images:

 

What are everyone's thoughts on my process? Is the theory sound? Has anyone done this as well on a digital camera?

 

I can share more photos if needed.

[Miguel Angel]

Beautiful images!

 

Would love to watch a trailer or the finished feature!

 

As for myself, I just finished a feature at the beginning of January and I shot it entirely at 1600 ASA on an Alexa Plus (444) because I liked the texture better than when rated at 800 ASA.

 

In fact, all the projects I have shot last year were rated either at 1600 or 3200 ASA (Dragon and Alexa)

I shot a short-film in October all at night at 1600 ASA and the odd 3200 ASA and it looked good! a couple of commercials at 1600 ASA and another short-film at 1600 / 3200 ASA.

 

I shot a commercial at 400 ASA because the client wanted it to be shot at 400 ASA and I couldn't believe how clean it looked after all the 1600 ASA jobs I did, too clean in fact!

 

I usually underexpose a lot because that's the way I like the images I create and I think that the Alexa looks way better at 1600 ASA than at 800 ASA.

 

Again, when shooting in daylight, rating the Alexa at 1600 ASA gives you a bit more of detail in the highlights, which is always welcome.

 

Have a good day!

[Robin R Probyn]

Miguel

 

Sorry if I have this wrong..Im not shooting narrative drama,s.. but are you saying that when you rate the Alexa at 1600 instead of the "native" 800.. you are intentionally over exposing by 1stop.. to be corrected in post.. and giving up 1stop of highlight in the process ..

 

As an aside I was just last night shooting with the new -ish Panasonic varicam LT.. which has this mysterious double native ISO setting adjustment .. 800 and 5000.. I shot day time ext of course 800.. and night which was pitch black to begin with and then illuminated with about 2,000 burning torches..! (festival in Japan).. and I was shooting slo mo 50 fps. 180 shutter. stopped down a bit.. ! and it seems to look good.. how they do this magical 2 native ISO,s I have no idea.. but the camera takes a very low time to get up to speed about 15-20 seconds depending how full the card is.. and eats batteries ..

[Stuart Brereton]

Underexposing by 3 stops from 800 ISO will certainly protect your highlights very well, but it will also crush your shadow detail way down into the noise floor, and dark areas will be largely unretrieveable. There are some cameras which benefit from underexposure because their knee curve allows highlights to clip abruptly, but the Alexa is not one of them.

[Robin R Probyn]

Sorry Ive got it the wrong way round ! your under exposing not over !

[AlejandroGomez]

really interesting post guys!

[Brett Allbritton]

I’m absolutely fascinated by this, but completely lost! Hahaha. I’d love to ask a few questions if you don’t mind, as I think these shots look great!

 

What I didn’t know before reading this is that changing the ISO isn’t actually making the camera more sensitive to light, but rather shifting how much of the details above and below middle grey that the camera records. More details in the shadows are recorded when decreasing the ISO and more details in the highlights are recorded when increasing the ISO.

 

I guess this is what initially made it confusing for me, as it seems so contrary to how I normally think about it: lower ISOs for bright, usually daylight scenes, and higher ISOs for dark, usually nighttime scenes. It seems ironic that the scenes which might really benefit from a high ISO are already really bright.

 

The part that really confused me though is that you set the ISO on the camera to 1600, but referred to that as underexposing 1 stop from 800. I’m sure there’s some conceptual thinking at work here that I just don’t understand yet, but that really threw me off as I would’ve taken this to mean a one stop increase from ISO 800.

 

To help clarify myself, suppose your incident meter gives a reading of f/8 at 800 ISO, which would be equivalent to f/11 at 1600 ISO. Wouldn’t 1600 ISO minus 2 stops be f/5.6, which is equivalent to the meter reading for 800 ISO minus one stop? You also mentioned that your incident meter was set to 6400 ISO, which would give you a reading of f/22 in this scenario. If the camera is at 1600, wouldn’t this reading be just two stops under exposed, not three? Sorry if I’m rambling, I’m just very curious!

Edited February 7, 2017 by Brett Allbritton

[Miguel Angel]

Miguel

 

Sorry if I have this wrong..Im not shooting narrative drama,s.. but are you saying that when you rate the Alexa at 1600 instead of the "native" 800.. you are intentionally over exposing by 1stop.. to be corrected in post.. and giving up 1stop of highlight in the process ..

 

As an aside I was just last night shooting with the new -ish Panasonic varicam LT.. which has this mysterious double native ISO setting adjustment .. 800 and 5000.. I shot day time ext of course 800.. and night which was pitch black to begin with and then illuminated with about 2,000 burning torches..! (festival in Japan).. and I was shooting slo mo 50 fps. 180 shutter. stopped down a bit.. ! and it seems to look good.. how they do this magical 2 native ISO,s I have no idea.. but the camera takes a very low time to get up to speed about 15-20 seconds depending how full the card is.. and eats batteries ..

 

 

Hi Robin,

 

I am not intentionally over or under exposing anything when setting up the ASA, it is just that I like the texture at 1600 ASA way more than the one at 800 ASA. The over / under exposure is the consequence of such change but it is not relevant to me at all.

 

Then, once I have set an ASA and I light, I under expose, but I always try to have a bright spot in the frame so it looks darker rather than "underexposed"

 

When changing the ASA to 1600 you shift your middle grey towards the highlights and you get around 1 extra stop in the highlights but lose it in the shadows.

 

 

In daylight is an advantage that is very welcome because you are not usually very worried about the shadows.

At night I like it too but is more challenging because you have to make sure that you light the shadows a bit or else there will be nothing there and it will be pitch black :D!

 

Rating the Alexa at 1600 ASA at night is also a problem in cities because the streetlights give so much light, especially if you shoot at T1.3, that you end up blocking the light from the streetlights rather than lighting!

 

I would love to try the Varicam, "The OA" and "Bad Santa 2" look really good!

 

Have a good day!

[AJ Young]

To help clarify myself, suppose your incident meter gives a reading of f/8 at 800 ISO, which would be equivalent to f/11 at 1600 ISO. Wouldn’t 1600 ISO minus 2 stops be f/5.6, which is equivalent to the meter reading for 800 ISO minus one stop? You also mentioned that your incident meter was set to 6400 ISO, which would give you a reading of f/22 in this scenario. If the camera is at 1600, wouldn’t this reading be just two stops under exposed, not three? Sorry if I’m rambling, I’m just very curious!

 

Not rambling at all!

 

You've got your math going the wrong direction. If the meter reads f/8 at 800 ISO, then the equivalent at 1600 ISO will be f/11. When I under exposed two further stops, the meter would read f/22 at 1600 ISO.

 

I decided to stick with changing my iso and further under exposing because of the way the ProRes codec compressed the information. The RAW shots were essentially 6400 ISO shots (three stops under from 800), but the ProRes shots were done through ISO and underexposure because of the way the codec recorded the image. My light meter was constantly reading exposures at 6400 ISO.

 

In hindsight, I probably could've just shot the entire project at 6400 ISO, but the testing and screening of the ProRes files lead me to decide on 1600 ISO minus two further stops.

[Stuart Brereton]

In hindsight, I probably could've just shot the entire project at 6400 ISO, but the testing and screening of the ProRes files lead me to decide on 1600 ISO minus two further stops.

I don't understand this statement. How is exposing at 6400 ISO any different to exposing at '1600 ISO minus two further stops'?

 

When you change the ISO, all you are doing is changing the distribution above and below mid gray. Setting the camera ISO at 1600 and then under-exposing by two stops, gives you 3 stops 'underexposure' from a base of 800. That's the same as setting the camera to 6400ISO. You are pushing the shadows way down into the toe of the curve either way.

[Brett Allbritton]

You've got your math going the wrong direction. If the meter reads f/8 at 800 ISO, then the equivalent at 1600 ISO will be f/11. When I under exposed two further stops, the meter would read f/22 at 1600 ISO.

 

D'oh! *facepalms*

 

Suddenly everything makes sense again, hahaha. I think in my head for whatever reason I was imagining the scene itself getting darker, hence my hypothetical meter reading numbers getting higher, rather than stopping down in camera to underexpose... Yeah, I'll use that excuse...

 

Thanks, AJ! I'd love to see more of the film sometime. Coincidentally I'm going to shoot a short film later this year that is also set in the 60s, and your footage is quite inspiring!

Edited February 7, 2017 by Brett Allbritton

[Miguel Angel]

I don't understand this statement. How is exposing at 6400 ISO any different to exposing at '1600 ISO minus two further stops'?

 

When you change the ISO, all you are doing is changing the distribution above and below mid gray. Setting the camera ISO at 1600 and then under-exposing by two stops, gives you 3 stops 'underexposure' from a base of 800. That's the same as setting the camera to 6400ISO. You are pushing the shadows way down into the toe of the curve either way.

 

 

I'm not a very technical person but if you could rate the Alexa at 6400 ASA (which you can't as far as I remember) the images that you would get from the Alexa at 6400 ASA and the Alexa at 1600 ASA + underexpose would be technically different in the way that you would lose around 2 stops of shadows when shooting at 6400 ASA because you would have shifted your middle grey values +-2 stops towards the highlights and lost those +-2 stops in the shadow details, as you stated.

 

Hence, "technically", if you wanted to grade both resultant images (6400 ASA and 1600 ASA) and wanted to lift the shadows, you wouldn't be able to do as much with the 6400 ASA image because there would be less detail in the shadows than in the 1600 ASA + underexpose image and you would get noisier shadows with the 6400 ASA image than with the 1600 ASA + underexpose.

 

Maybe that's what he discovered?

 

Have a good day!

[AJ Young]

That it's exactly it, Miquel.

 

I should've been a little more clear on why I landed on 1600ISO - 2.

 

We were pushing the image back up in the color grade to "normal" through a combination of lift and gain (barely touched gamma). The way the ProRes codec interpreted the data gave us the image we wanted. We made the RAW footage match the ProRes results.

 

Miquel also made an excellent point that the Alexa can't go to 6400 either (I completely forgot. Maybe this is why hindset isn't always the best?). The only way I could choose 6400 ISO was through reinterpreting the RAW data in post. However, we shot the tests and film in ProRes 4x4, so we decided to do the previously mentioned method.

[Robin R Probyn]

 

 

Hi Robin,

 

I am not intentionally over or under exposing anything when setting up the ASA, it is just that I like the texture at 1600 ASA way more than the one at 800 ASA. The over / under exposure is the consequence of such change but it is not relevant to me at all.

 

Then, once I have set an ASA and I light, I under expose, but I always try to have a bright spot in the frame so it looks darker rather than "underexposed"

 

When changing the ASA to 1600 you shift your middle grey towards the highlights and you get around 1 extra stop in the highlights but lose it in the shadows.

 

 

In daylight is an advantage that is very welcome because you are not usually very worried about the shadows.

At night I like it too but is more challenging because you have to make sure that you light the shadows a bit or else there will be nothing there and it will be pitch black :D!

 

Rating the Alexa at 1600 ASA at night is also a problem in cities because the streetlights give so much light, especially if you shoot at T1.3, that you end up blocking the light from the streetlights rather than lighting!

 

I would love to try the Varicam, "The OA" and "Bad Santa 2" look really good!

 

Have a good day!

 

Yes sorry I realized I had it the wrong way round.. for the Sony F5/55 in Slog it seems now standard practice to intentionally "over expose" by at last a stop to help with noise .. I guess actually the opposite of what you are doing.. the LT varicam is the smaller version of the full sized varicam no doubt used on the features.. but the insides are the same I believe.. just has this funny control panel off the back of the camera.. as no room on the actual body.. but still quite chunky I thought.. but its claim to fame seems to be this dual native ISO thing they have..

[Miguel Angel]

I've never shot with a Sony F5 / 55 but I would love too as I have seen so many beautiful things shot with those ones that at some stage I will have to use one :), either one of those or the F65!

 

If you happen to have the piece you shot with the Varicam, feel free to post it, I'd love to see it! :)

Have a good day!

[Robin R Probyn]

It was for the BBC and I doubt I will get a copy anytime soon.. well at least not till it has aired..and then I probably wouldn't be allowed to put it on a public site.. we had a total of 17 camera,s covering the main event.. 4 X Varicams and a load of Sony a7,s with Canon primes.. and some go pro,s.. but quite specular .. about 2000 people with fire torches coming down a pitch black mountain ..all dressed in white with straw shoes.. well some even running down 500 odd steps..its a race they have been doing for a very long time.. thats why they went with the varicams for the 5000 ISO setting.. Im pretty sure Panasonic will want to use it for promotion purposes somewhere down the line though..

 

Theres much talk in Sony land of a new F65.. so that might be coming sooner than later.. some sort of combined F65 with the smaller form of the F55..

[Miguel Angel]

Thank you Robin! sounds like you had a great time there!

 

If that new F65 you are talking about pops out I will be very happy as the F65 is such a premium camera in both, Ireland and Spain, that if you want to rent it is just impossible so a new F65 will mean that the old F65 will be cheaper to rent! :D

 

Have a lovely day!

[Robin R Probyn]

Yes they could plummet in price like the F35..! good luck with your next feature sir ..

[Stuart Brereton]

the images that you would get from the Alexa at 6400 ASA and the Alexa at 1600 ASA + underexpose would be technically different in the way that you would lose around 2 stops of shadows when shooting at 6400 ASA because you would have shifted your middle grey values +-2 stops towards the highlights and lost those +-2 stops in the shadow details, as you stated.

 

The noise floor doesn't change when you change the ISO. It's a constant. It really doesn't matter whether you underexpose 1600ISO by 2 stops, or set the camera at 6400ISO (assuming this was possible), the net result is a 3 stop underexposure relative to 800ISO. Detail that is pushed down into the noise floor will be noisy, regardless of the rationale you used to put it there.

[AJ Young]

I think the key was how I reinterpreted the data in the grade through the ProRes file. I pushed and pulled parts of the new exposure to shape the dynamic range because the image was underexposed in the ProRes file. If it were RAW, rating it to 6400ISO wouldn't give me the same results in comparison to a combination of gain/gamma/lift to shadow drop off and highlight roll off.

[Albion Hockney]

AJ - I think the images look nice - but they don't seem underexposed by two stops. In all the daylight shots you have highlights creeping up atleast 80/90 IRE - so maybe your saying your under exposing the key light two stops on faces? but that is not the whole picture. For example in the third still you couldn't even really go much brighter without clipping highlights so I would look at like pretty much a normal exposure. Did you lift them back up in post? if so then yes its the same as shooting 6400ISO - the only difference is you did the lift in post with more flexibility but you could have shot 6400ISO in camera and gotten the same result - just a slight different process.

 

I personally have never understood this whole idea of underexposing a film by a number of stops - I came up on digital though and am a waveform/monitor person. My question when people say that is always. underexposing compared to what? this idea assumes that the normal way to shoot a movie is have every face you shoot registering the same f stop as the camera with a incident reading and pretty much no one shoots like that.

Edited February 9, 2017 by Albion Hockney

[AJ Young]

AJ - I think the images look nice - but they don't seem underexposed by two stops. In all the daylight shots you have highlights creeping up atleast 80/90 IRE - so maybe your saying your under exposing the key light two stops on faces? but that is not the whole picture. For example in the third still you couldn't even really go much brighter without clipping highlights so I would look at like pretty much a normal exposure. Did you lift them back up in post? if so then yes its the same as shooting 6400ISO - the only difference is you did the lift in post with more flexibility but you could have shot 6400ISO in camera and gotten the same result - just a slight different process.

 

I did lift the highlights back up when I corrected the entire exposure in the grade, but changed how they rolled off to 100%. We soft clipped the highlights and essentially had a lot of information stored in the highlight range that we fiddled with on a shot by shot basis.

 

Unfortunately, the JPG images I've shared may not represent the full picture, no pun intended. The difference is more apparent when screened on a DCP.

 

However I see what you're saying, the difference between 800ISO and my re-rated ISO in terms of highlights is pretty subtle. Maybe too subtle to warrant this method I did?

[Miguel Angel]

The noise floor doesn't change when you change the ISO. It's a constant. It really doesn't matter whether you underexpose 1600ISO by 2 stops, or set the camera at 6400ISO (assuming this was possible), the net result is a 3 stop underexposure relative to 800ISO. Detail that is pushed down into the noise floor will be noisy, regardless of the rationale you used to put it there.

 

 

That's true!

I stand corrected!!! :)

[Daniel Meier]

Great looking shots, AJ.

I would like to compare some of your test shots in Resolve. Could you provide a download link to those clips that you have on Youtube?

 

And two general questions:

Does that grayscale mapping theory apply to all digital cameras or just the ones that are able to shoot RAW?

EDIT: And does it apply to film as well? For example Kodak 200T giving me more shadow detail than Kodak 500T?

 

I'd like to understand that theory a little better.

One example using the BMPCC in RAW mode. Let's say I shoot an exterior scenery at bright sunlight rating the camera at 1600 ISO and underexposing one stop. Then bringing the shot into Resolve and choosing ISO 800 instead of ISO 1600. Does that change the 18%-mapping (along with the highlight range)? Or does it always need to be done in-camera?

 

My approach is that doing it in-camera is the only way. Because Resolve would just pull exposure by one stop, leaving the DR untouched.

During shooting I already baked that tonal range into the BMPCC's sensor, by deliberatly underxposing and therefore getting more highlight detail.

Hence this is different from e.g. White Balance, which can be adjusted in post, without any quality loss.

Edited February 11, 2017 by Daniel Meier

[AJ Young]

Great looking shots, AJ.

I would like to compare some of your test shots in Resolve. Could you provide a download link to those clips that you have on Youtube?

 

And two general questions:

Does that grayscale mapping theory apply to all digital cameras or just the ones that are able to shoot RAW?

EDIT: And does it apply to film as well? For example Kodak 200T giving me more shadow detail than Kodak 500T?

 

I'd like to understand that theory a little better.

One example using the BMPCC in RAW mode. Let's say I shoot an exterior scenery at bright sunlight rating the camera at 1600 ISO and underexposing one stop. Then bringing the shot into Resolve and choosing ISO 800 instead of ISO 1600. Does that change the 18%-mapping (along with the highlight range)? Or does it always need to be done in-camera?

 

My approach is that doing it in-camera is the only way. Because Resolve would just pull exposure by one stop, leaving the DR untouched.

During shooting I already baked that tonal range into the BMPCC's sensor, by deliberatly underxposing and therefore getting more highlight detail.

Hence this is different from e.g. White Balance, which can be adjusted in post, without any quality loss.

 

I unfortunately can't share the original test footage as download from internet; the production cleared me to upload them but I am not at liberty to allow them to be downloaded without the production's permission.

 

In theory, the "grayscale mapping" is applicable to all digital cameras.

 

As far as I know, film works a tad differently and the ISO of film is actually dependent on the grain structure of the emulsion (smaller silver halide vs. larger silver halide) and doesn't result in a dramatic change in dynamic range between the shadows/highlights of different film ISO's. However, "re-rating" the ISO of a stock (under/overexposing) will yield the same results.

 

About your question on the theory, the underexposure would need to be done in camera. When shooting RAW, it's less about changing the ISO electronically on the camera, and more physically about changing the ISO with a light meter. When not shooting RAW, like ProRes/XVAC/etc, it then becomes a combination of the ISO electronically on the camera and changing the ISO on the light meter.

 

For RAW, ISO is just another variable for interpreting the data from the sensor. It's metadata just like White Balance. The key to redistributing the dynamic range is through the amount of light you give the sensor in RAW. For example, the camera's native ISO is 800 (in this case means the most equal distribution of the dynamic range from shadows to highlights). When shooting, however, I "re-rate" my light meter to 3200 ISO which will physically give the sensor 2 stops less of light, regardless to what the camera's ISO is set to electronically. To correct the RAW image in post, you would either change the ISO of the RAW file in DaVinci to match your light meter or you use a combination of gain/gamma/lift to "fix" the image to proper exposure.

 

The feature I did used gain/gamma/lift and ISO changes within camera because we weren't shooting in RAW, but LogC in ProRes. Unlike RAW, log images do have an ISO baked into them because they are interpreted data from the sensor, in contrast RAW files are uninterpreted data from the sensor. Log profiles of course try to preserve the dynamic range and color values, but ISO and White Balance aren't easy variables to adjust at that point in comparison to RAW. I liked how the LogC profile responded to 1600 ISO 2 stops under and decided to shoot the project at those specs. I set the camera to 1600 ISO (1 stop under exposed from 800 ISO), and my meter to 6400 ISO (3 stops under exposed from 800 ISO).

 

In either case, the key factor is getting an underexposed image in camera. A digital sensor is just like film; it can only see so much and the ISO really doesn't change, with the assumption that ISO in general is the equal distribution of dynamic range between shadows/highlights. Therefore, changing the ISO of a digital sensor means you're giving the sensor less/more light and re-interpreting the data to compensate. Film is the same; you would re-rate the ISO of the stock by giving the emulsion less/more light and correcting it through push/pulling the stock.