Overexposing for cleaner footage?

Jarlath McKernan · November 1

I shot this low light test yesterday in film school and my friend mentioned it would be good to always expose a little over, so as to capture more shadow detail. Then in post bring it all down. He mentioned most if not all industry DOPs over expose by a couple stops in recognition that it will be taken down in post.

can someone please explain this process to me. Is it common? If you shoot on arri raw is this even necessary? Wouldn’t raw be capturing everything anyway?

all responses appreciated

silvan schnelli · November 1

I personally highly recommend readings ARRIs white paper on Dynamic Range, which talks about Exposure Index, SNR (Signal to noise ratio) and other things.

The logic of overexposing and then bringing it down in post is valid, for reasons such as increasing SNR (less noisy appearance), the fact that making a scene darker is much more forgiving than brightening a scene and and overexposing allows for more information in the shadows.

However, I do think that especially with todays digital cameras which have very good shadow latitude overexposing for the shadows is not as necessary, perhaps for a dark scene like yours it would be. An interesting point was brought about by David Mullen ASC in his recent appearance on the Team Deakins Podcast. He notes that it’s okay to underexpose as long as you know that, you or anyone else isn’t planning on bringing up the shadows later down post production pipeline.

Going back to the paper, a digital sensor to my knowledge only has one ISO speed, sensitivity to incident light, and that different EI settings are simply different curves, which change what signal value the camera interprets as middle gray. But seeing as the dynamic range of camera does not change (the maximum light a camera can capture before the photodiode becomes saturated and the minimum light to give a usable signal, what changing the EI does is give you either more highlight latitude or shadow latitude. I think this is also why in ARRI Raw you can still change the EI, you have exposed the sensor and changing the EI now just is choosing what is middle grey.

Practical Use:

A lot of people quote the EI of 800 to be the Alexa’s base ISO, meaning that if you want to overexpose it, you could set an EI of less than 800, e.g. 400. This would theoretically bias you to overexpose the sensor by one stop. The benefit though is that the image is overexposed but the EI already pulled down the image for you, so you have an overexposed image, but one that is correctly exposed to the way you want to light it. If you film in RAW this correction is not baked in, as the sensor data has not been debayered yet.

This process is very similar to analog film and I think it was specifically designed to be so similar to maybe allow Cinematographers to film in the same way that they have been.

I hope this explanation helps and just wanted to note that your shot is very nice and moody.

A question I have myself is that I’ve always wondered if filming at EI 400 and correctly exposing would give me the same result as filming at EI 800, overexposing 1 stop and pulling it down in post. Intuitively I would think yes and the differences may be minor, but I guess depending on the the log encodings (e.g. logc3 the different curves have slightly different slopes) and the way the image is pulled down, minor differences may be noticeable.

https://www.arri.com/resource/blob/295460/e10ff8a5b3abf26c33f8754379b57442/2022-09-28-arri-dynamic-range-whitepaper-data.pdf

https://teamdeakins.libsyn.com/technical-knowledge-for-cinematographers

Albion Hockney · November 1

overexposing a bit is a common practice- not everyone does that though and its really dependent on the final look you want to achieve. if your happy without having detail in the shadows and want them to be true black in places OR you like the added grain or noise you get when lifting them a little then thats great. If you want clean blacks with lots of detail in the tow of the curve then you might want to over expose.

Do some tests.

Karim D. Ghantous · November 2

12 hours ago, silvan schnelli said:

A question I have myself is that I’ve always wondered if filming at EI 400 and correctly exposing would give me the same result as filming at EI 800, overexposing 1 stop and pulling it down in post. Intuitively I would think yes and the differences may be minor, but I guess depending on the the log encodings (e.g. logc3 the different curves have slightly different slopes) and the way the image is pulled down, minor differences may be noticeable.

IMHO, there is almost no difference. It obviously can depend on the sensor - some sensors are not invariant, some are. Alexas use dual gain which is another thing. Albion said it best:" do some tests".

FWIW, I used to have a Leica M8 for a short while. From what I saw, it was better to underexpose than to change the ISO setting.

silvan schnelli · November 2

6 hours ago, Karim D. Ghantous said:

Alexas use dual gain which is another thing.

Interesting, I’m currently reading up on this.

David Mullen ASC · November 2

A lot of cameras now have dual gain but ARRI doesn't use it for high ISO settings, they use it for extended dynamic range -- the two outputs from the sensor, high and low gain, are combined into one signal.

Personally, because I shoot series and features with many weeks of dailies, if not months, I don't want to shoot anything that looks "wrong" (too bright) with the idea of fixing it later, after editing is complete. Too many people will be working with the footage before it is color-corrected. I want it to look correct on the set monitors and in dailies.

Yes, sensors, like film, prefer more light but you have to take into account the headroom where detail gets clipped.

I think a better method is to find an ISO rating that gives you the balance of noise-to-headroom you like and then LIGHT for the amount of shadow and highlight detail you like. That way you won't be fixing it in post later anyway, you won't be trying to pull detail out of the shadows (but if you think that is likely, then you could work at a lower ISO in general knowing that your recording has less noise.)

Occasionally when you shoot a very dim scene with no bright highlights, then you could select a lower ISO, which will give the sensor more exposure with less overall noise in post, and things will look the way you want on the monitor and in dailies.

Nicolas POISSON · November 2

Overexposing the sensor and bringing the brightness down in post allows for cleaner shadows because the main culprit for noise is the sensor itself. This is a very common principle of noise reduction : boost the signal before the source of noise, bring everything down (both signal and noise) after. Shooting in raw, log or 709 does not change this principle : this is too late. The key is that the sensor receives more light. This is at the expense of highlight margin. You gain in shadows what you loose in the highlights.

Say the sensor adds « 1 » of noise (arbitrary unit). You set your camera to have a correct exposure. The optical signal ranges from 1 (shadow) to 10 (hot spots). Your signal/noise ratio is 1:1 in the shadows.

Now you open the aperture by one stop. The optical signal ranges from 2 to 20. Signal to noise ratio is now 2:1. But hey ! The image looks overexposed now ! No problem : apply a x0.5 gain in post. The signal is back to 1:10, and the noise is 1x0.5 = 0.5. The signal to noise ratio in the shadows is now 1:0.5, which is the same as 2:1. This works perfectly as long as the sensor’s clipping threshold is greater than 20. If it is 15, overexposing lead to clip.

Thus overexposing is a good technique when :

- you want clean shadows

- the scene has not much contrast, and the hot spots are far from the sensor’s clipping point.

The RIAA correction for vinyls works the same way : the audio high frequencies are boosted before cutting the master. When playing the LP, the stylus sliding on the track generate a lot of high frequency noise. The RIAA circuit brings the high frequencies down. The signal reverts to linear, but noise is attenuated.

Using cameras with a 0,45 gamma (shadows are boosted with respects to highlights), airing that signal (noise added) and bringing everything back to linear thanks to the inherent gamma 2,2 of the CRT, has been a noise reduction system used for years.

19 hours ago, silvan schnelli said:

A question I have myself is that I’ve always wondered if filming at EI 400 and correctly exposing would give me the same result as filming at EI 800, overexposing 1 stop and pulling it down in post. Intuitively I would think yes and the differences may be minor, but I guess depending on the the log encodings (e.g. logc3 the different curves have slightly different slopes) and the way the image is pulled down, minor differences may be noticeable.

I may be wrong but if you want cleaner shadow, you should lower your ISO/EI, so switch to EI 200. If you rate EI800, this will artificially brighten your image, and you will close the aperture to compensate. With less light, shadows are even more noisy, but y gain 1 stop of latitude in the highlights.

Now with log encoding, there is another problem. The lower part of the log curve is, well, not log. If it were, the log curve would be a straight line, whereas all log curves have a toe. In that zone, the number of code values per stop is low, which makes it hard to grade. This may not be a big deal for 12bit log, it will be a greater problem for 10 and particularly 8 bit log.

Note that this only happens for cameras that set the ISO shifting the log curve left/right. If you use Sony’s CineEI, or if you shift the brightness using exposure compensated LUTs on an external monitor, you will not have this problem.

Jon Pais · November 4

@Jarlath McKernan You definitely don’t want to overexpose anything. In the English language, over- anything usually means excessive. What you are referring to is ETTR, or exposing to the right, which is actually not all that common in the industry. An exception is DP Erik Messerschmidt (Mindhunter), a proponent of ETTR and everything I’ve seen of his looks impeccable.

Nicolas POISSON · November 4

I used to think the same, but it seems sometimes pro guys use the term differently. Overexposure does not (always) mean clipping. It means you expose higher than you would using a light meter, or looking at a grey card on a waveform, compared to a recommended practice. Or higher than your own taste looking at the image with a standard LUT. In that meaning, it is the same as pushing / pulling, or rating a different ISO/EI.

Edited November 4 by Nicolas POISSON

M Joel W · November 4

I think that looks good. If you think it looks good, don't worry about it.

As others have mentioned, overexposing and bringing the imagine down in post will reduce noise, which might be desirable, similar to how shooting on a slower film stock would reduce grain. It would also mean you risk clipping highlights sooner, however.

There is a famous story about how on Mad Max: Fury Road they overexposed the day for night sequences.

Lubezki also overexposed (film) on the New World and eschewed color correction filters I believe.

On the other hand, some DPs will underexpose the Alexa to shoot at an effective 1600 or 3200 ISO to introduce more texture. Others will add the grain in post.

It's an aesthetic choice and there is no right or wrong approach so long as the image isn't completely blown out or underexposed. (Actually, that might be even more interesting.)

Rating the camera at 200 ISO will achieve the same thing as overexposing 800 ISO by two stops if you are shooting in ArriRAW, only the viewfinder's image will appear two stops darker. So just shooting at 200 ISO is an option if you want a cleaner image (but you will clip highlights two stops sooner). Personally, I might do this if I'm shooting on a green screen and there are no highlights that risk clipping, but otherwise I wouldn't bother. But that's just me.

Edited November 4 by M Joel W