ISO for video: more complicated than it seems
There is a video version of this article on our YouTube channel, if you prefer.
As more and more video shooters switch from photo-type cameras to professional digital cinema cameras, correct exposure is becoming more and more of an issue. People shoot in LOG and RAW simply because they can, not really understanding how these formats work and what they require in terms of exposure. Many still operate on the notions and concepts passed down to us from our photographer brethren, concepts that lose their relevance or become simply incorrect when it comes to shooting video. One such concept is ISO.
ISO makes the video brighter, right? Well, yes, kind of. But not in the way you think or want. ISO is a surprisingly complex topic. And one that took me a long time to really wrap my head around. And some of the things I learned are completely counter intuitive, like using lower ISO in low-light situations. Wait, what? Yes, we’ll get to that. This article is for those who aren’t satisfied just knowing which button does what on the camera. If you prefer to understand what is happening “under the hood”, then read on.
ISO came from film
First of all, let’s bust some myths. You’ve probably heard from all those “exposure triangle” videos, that ISO changes the sensitivity of your camera. Well, that is just plain wrong. And let’s see why.
I’m not gonna bore you with history, but it’s important to understand that ISO came to us from the film days. It’s a measure of a film stock’s sensitivity to light, which is determined by the size of the grains of silver halide in the emulsion. Bigger grains absorb more light. Pretty straightforward.
ISO also serves an important purpose for shooting film because it’s the number that you need to put into your light meter to get the correct F-stop for neutral exposure. Because you can’t actually preview exposure on a film camera. You have to use a light meter.
So at the beginning of the digital era ISO was used as an easy way to communicate sensor sensitivity to people who were used to shooting film. In terms of exposure ISO 100 on a digital camera should be equal to 100 ASA film stock. And that’s about the point where film comparisons lose most of their usefulness or relevance.
ISO does not measure or change sensitivity
ISO for film is exact science. ISO for digital is mostly the camera manufacturer’s opinion. That’s why, for example, Sony’s 800 ISO and Canon’s 800 ISO are not exactly the same thing, there can be a difference as large as a stop of exposure depending on the intricacies of sensor design. And on some cheaper cameras ISOs are just plain wrong or mislabeled.
Second and more important reason is that film stock analogies were misleading in the first place. Because you can’t swap the sensor in your camera like you can swap out a roll of film. You are stuck with whatever sensitivity your sensor had when it left the factory. (And we’ll talk about dual ISO later). And vice versa, you can’t change the ISO of film stock, but you can change the ISO in a digital camera.
Then what does ISO even mean then in a digital camera? If you can’t change the sensitivity of your camera’s sensor, then what does changing the ISO do? Well, it reflects the change in the digital gain. Upping the ISO amplifies the signal being recorded. And the difference between ISO and gain is not just semantics, we are not talking about words here. Understanding how gain works has some pretty serious implications for how we should expose when shooting on digital cameras.
ISO really means gain
To understand gain let’s quickly brush up on how a digital imaging sensor works. It consists of photo-sensitive diodes arranged into a grid. Any light that falls onto the sensor registers as a signal. More light means stronger signal. But if there is not enough light, the signal is drowned in noise. After all, the sensor produces some stable level of noise as a side effect of its normal operation. You’ve probably heard of the signal-to-noise ratio. So, if your signal is well above the noise floor, you have high SNR, the image is clear. Conversely, if the signal is mired in noise — your SNR is low and the image is unusable.
Now here is where gain comes in. You can supply more voltage to your sensor to make it work harder. It’s literally the same as turning up the gain knob on your guitar amp.
Gain is measured in decibels and each 6 decibels of gain is equal to 1 stop of light or doubling of ISO.
Some professional video cameras like my FX6 can be switched to show gain instead of ISO, but even if yours can’t, it is better to start thinking about ISO in terms of gain and it all has to do with native ISO.
The signal gets amplified, so your picture gets brighter on the monitor, but there is a catch. And a huge one — you see, the noise floor also rises. If you double the signal you double the noise. Now, this is hugely important to understand and if you take away only one thing from this article it should be this: you can’t increase exposure with ISO. Your camera does not get any more sensitive.
Brightness of the picture on your monitor is not the same thing as exposure. Increasing gain does not let in more light.
Think of it this way: exposure happens BEFORE the light hits the sensor. If something is happening to the signal after the sensor — it is not exposure by definition.
So if you are shooting a dark low-key scene, where there’s a lot of important detail in the shadows, increasing your ISO actually ruins your picture. Your scene is already sitting in the noise floor and now you are raising it. No bueno. You are better off pushing the noise floor down by lowering your ISO below the native which means applying negative gain.
Negative gain works exactly as it sounds it should. It’s the complete opposite of increasing gain — your signal gets weaker, you lose dynamic range in the highlights, but the noise floor also gets lowered. So the dynamic range is now distributed more towards shadows — which is where your image lives, if you are shooting a dark scene. Of course, you will need to compensate for the loss of exposure with aperture, shutter speed, or light. But remember — lower the gain to get cleaner shadows. I know it’s a mind-bender and completely counterintuitive, but believe me. Or better yet — test this yourself.
By the way, negative gain is part of how the ARRI Alexa gets its legendary 14 stops of usable dynamic range. The sensor is read simultaneously at zero gain (to preserve color reproduction in the highlights) and negative gain (for the sake of low noise in the shadows). The two datasets are then merged into a single high dynamic range image. A similar (but not identical) dual gain technology is implemented in the Canon C300 mk.III and Canon C70 cameras.
Native ISO = 0 gain, always
If you’ve seen at least one filmmaking channel on YouTube you’ve probably heard that you should shoot at native ISO — you get the cleanest image and highest dynamic range of your camera, right? Yeah, but why? What is native ISO? Every digital camera has one. Native ISO means one thing — 0db of gain on the sensor. It’s the power level at which the sensor is calibrated by the manufacturer to capture the highest dynamic range the sensor is capable of. Different cameras may have different native ISO numbers but all of them equate to 0db of gain.
Some sensors are equipped with 2 sets of gain circuits. That gives us the ability to switch between them without adding any gain and thus — any more noise. This is called dual native ISO and it’s a killer feature. It’s like having two sensors in one camera. By the way, this is one of the main reasons why Sony Venice is gaining popularity among top DPs and stepping on the heels of ARRI Alexa a little bit.
If you forget about ISO and think in terms of gain, your life gets that much easier, because you always know, that 0db of gain means native ISO regardless of which picture profile or dual ISO mode you are in. You don’t need to do any math or remember crazy sets of numbers. And you always know how far away from native ISO your gain is. For example, I know through testing, that for my camera, at lower base ISO 6db of gain is completely fine, 12db is about as high as I would go and if I need 3 or more stops I’m better off switching to higher base ISO and putting on ND.
Change of ISO doesn’t always mean change of gain
You might have noticed that native ISO is different for different picture profiles. For example, on my FX6 the native ISO is 800. But that’s only for Slog-3. If i’m shooting S-cinetone — which is a “what you see is what you get” type of profile, the native ISO is 320. Why 320? Again, with ISO numbers, it doesn’t actually mean what you think it means. It doesn’t mean that the camera applied negative gain. And it doesn’t mean that S-cinetone is 1,3 stop darker than LOG. The sensitivity did not change, it is still 0db.
ISO320 is just a number that you would have to put into a light meter to get correct exposure for middle grey for that particular picture profile.
LOG is flat, S-Cinetone is not, and middle grey falls in a different place on the waveform. If you want to match their exposure, you use different ISO. But the sensitivity did not change. Still 0 gain.
Note the gain value - it does not change when switching gamma curves and ISO bases
I know this is super confusing and that’s what I’m trying to tell you. ISO is a misleading number. And I firmly believe that for video people, it’s not the correct metric to use. What you really want to know is gain. Because 0 gain is 0 gain, regardless of picture profile, dual ISO and even the camera. So whatever you are shooting, if you are at 0 gain, you know you are in the native ISO for that.
Shooting in LOG? Use native ISO. Like, seriously
LOG gamma curve allocates the data more evenly between the stops of exposure.
Regarding native ISO, it’s important to note that the “highest dynamic range” claim is only true if you’re shooting in the flattest LOG profile available on your camera. The camera sensor registers light linearly: the more light, the more information. But we do not need the information from the brightest areas to occupy the vast majority of the file, because the human eye is best at distinguishing between midtones. Therefore, the camera converts the linear gamma to a logarithmic one by bending the top and bottom of the graph. Thus, the information in the container (file) can be evenly distributed between the steps of the dynamic range. There will be about as much information in the shadows as there is in the highlights. This is the whole point of the logarithmic gamma — to cram the maximum dynamic range into the file. But this is not always good (especially if you have a tiny 8-bit container) and not always necessary, more on that later.
You’ve probably seen charts like this. They basically show the allocation of information in your LOG image between shadows and highlights. Now the highest dynamic range will be pretty evenly distributed between shadows and highlights. The middle point is middle gray which is your 18% gray card. When you up the gain, you lose stops in the shadows to increased noise. But you don’t actually gain any new stops in the highlights, because the sensor’s sensitivity hasn’t changed, so the clipping point is still the same. If you clip your highlights with gain — they’re gone. So it bears repeating — increasing gain doesn’t increase light sensitivity, it only pushes both the signal and noise up the waveform.
Maximum theoretical dynamic range of S-Log3 is 14 stops, but factual DR is limited by the noise floor of the sensor. For example, Sony FX6 has a dynamic range of 12,8 stops at SNR:1 (as measured by CineD). Upping the ISO raises the noise floor, thus reducing the DR.
Again, if you’re shooting in LOG format, using a native ISO is paramount. So much so that if you touch ISO at all, it means you’re cutting dynamic range. And if you don’t care as much about dynamic range as you do about overall image brightness, then why shoot in LOG at all? Honestly — in this case, shooting in a standard profile becomes the best option. After all, the standard profile takes linear data directly from the sensor, before applying the logarithmic function, that is, the file container will be filled with information more rationally. More information where there is more light. Remember: the gamma curve does not affect the size of the container, it only determines how it will be filled.
why you should overexpose LOG
Your camera’s LOG curve and its native ISO are interrelated, set in relation to each other to capture the most information in both shadows and highlights with a neutral exposure. For example, for S-Log 3, neutral gray falls at 41%.
But the dynamic range of the scene being filmed does not always go beyond the DR of LOG profile. In this case, there are “empty shelves” in our file. Let’s say there are 10 shelves in the file, but we only have shelves 2–7 filled. In this case, it is better to expose with a margin, because we already know that the bottom two shelves are always occupied by noise. It is necessary to give more light to the sensor and “shift” the information a couple of shelves higher.
That’s why you may have heard all this “expose to the right” advice from photographers. For video, this should be called “expose to the top”, because we look at the waveform. This means that your logarithmic image must be above the noise floor in order to record the cleanest signal.
You will get the cleanest image out of the camera by overexposing the LOG waveform by 1-2 stops (while still protecting the highlights). Any further overexposure does not increase the benefit, however.
But here’s the problem — our low key scene no longer looks dark on the monitor, and we wanted a dark scene. What do? Easy. Just load an underexposed 3D LUTs into your external monitor (it’s very easy to make your own). Let’s say you have three LUT versions: 0EV, -1EV, and -2EV. Want to overexpose your LOG by one stop? Look at the scene through -1EV LUT. A similar function is performed by EI, which will be discussed below.
After shooting, you can adjust exposure in post. Just make sure to lower it BEFORE applying a LUT or a transform. Again, because LUTs and transforms expect you to correctly expose the gray card for your particular LOG gamma. This way you will get a clean image with correct exposure in the developed image.
ISO in RAW is really Exposure Index (EI)
Now, for reasons we just discussed, in a lot of high-end cinema cameras like Arri Alexa, you can’t even control gain. You are stuck in the native ISO forever. This is to be expected because for cinema applications dynamic range is everything. And if you are shooting in RAW — same thing, there is no gain, no ISO control. But that only goes for true RAW formats that record signal data before gain is applied. Some cameras, like Canon’s cinema line for example, apply gain earlier in the imaging chain, so their Cinema Raw Light actually has gain backed in. But in true RAW formats, like R3D or ARRIRAW, ISO is only recorded as metadata, there is always 0db gain on the sensor and the ISO wheel only affects the image on your monitor, but not the one being recorded to the card.
That is why a lot of RED users experimentally found that using ISO 50 gives a cleaner image. This is really more miscommunication and weirdness on RED’s part. Because every other major cinema camera manufacturer uses the term Exposure index or EI instead of ISO for that purpose. EI basically only regulates the brightness of your monitoring LUT. For example, if you are shooting at a native ISO of 800 but your EI is set to 1600 you are basically telling your camera — show me the picture as if I had one more stop of light. EI is a monitoring tool designed for literally fooling yourself into under or — more commonly — overexposing your image. But it does not effect what is being recorded on the card. You are always still recording at native ISO.
High ISO NOISE and noise reduction
Ok , you might say, that is all very interesting, but I’m shooting a wedding in a dark restaurant tomorrow with a budget mirrorless camera, and I need to increase the ISO, so what should I do now? Well, it really do be like that sometimes. But we need to have a serious talk about noise reduction. LOG and noise don’t go well together, at all. Any weirdness recorded in LOG gets greatly amplified when conforming to REC709 in post production. When shooting in LOG, you really want to turn off all in-camera noise reduction and stay clear above the noise floor. This way, you get the cleanest camera negative to work with in post.
If you are increasing gain, and your image does get muddled by noise, then your in-camera noise reduction becomes the superior option. You are better off leaving noise reduction and color interpretation to your camera, rather than struggling with messed up LOG footage in post.
You see, camera manufacturers know their sensors very well, they can fine-tune their noise reduction to their particular noise pattern at each gain level. And since a lot of the noise is not only luma but also chroma, that means that color information also gets messed up by increasing gain. And we are working with Bayer pattern sensors here — so the color information is already somewhat incomplete, there’s a lot of guessing going on by the camera. That’s why at high ISOs it’s better to let the camera do the work and present you with a complete image, rather than trying to fix it in post. And that means using a what-you-see-is-what-you-get picture profile. This is where you let your camera apply that pesky “color science” we’ve been hearing about for years — a term I hate, but it’s actually applicable in this case.
How high ISO causes bad image
And this is exactly why you might have thought that ISO is an exposure tool. Because you can increase it, and up to a point your image gets brighter but not noticeably worse. And people are saying stuff like “this camera is good in low light, you can shoot up to 6400 ISO pretty safely”. All that means is that built-in noise reduction is well tuned for that sensor.
But the image does get worse and it IS degraded, for you see — noise reduction is not a free lunch. It comes at a price — by fighting noise you average out the pixels losing fine detail and color fidelity: your image gets blurry and the colors start falling apart. And to combat that, your camera paves over with digital sharpening and raises the saturation to make up for lost color gradations. At high enough gain levels, those two processes combined give you that dreaded ugly-ass “video” look.
10 Practical takeaways
Well, if you survived that information deluge, my respect and thanks to you. Now for the TL;DR crowd I want to summarize everything we discovered so far into 10 practical actionable takeaways that you can implement in your everyday shooting right now:
ISO is not an exposure tool. Increasing gain doesn’t get you more light. Want better image? Pour more light on that sensor. Don’t fool yourself! Brightness is not the same thing as exposure
Native ISO is a misleading number that depends on your picture profile and how sensor is calibrated. What you really want to know is gain. 0 gain means native ISO in all circumstances
You can lower the gain to get cleaner shadows but lose DR in the highlights.
Dual native ISO should be way higher on your camera features priority list. Higher than IBIS or AF. Exposure > everything else. That is why we see more and more Sony Venices being used.
What makes a camera good in low light is the sensor & photosite size resulting in higher base sensitivity, not how it deals with noise from high ISOs.
If you want maximum dynamic range — shoot in LOG exposing for middle gray. If you have highlight headroom — overexpose. Move the LOG waveform to the top until you clear the noise floor (~0-20% IRE).
When shooting LOG always shoot at native ISO, turn off all in-camera noise reduction and sharpening.
If you are in a run-and-gun situation and have no choice but to expose with ISO, abandon LOG and use a WYSIWYG picture profile instead.
But remember — excessively high ISO leads to crappy image via NR, sharpening and saturation. Same goes for tiny sensors. Learn your camera! Study the ISO performance and determine what is the acceptable range.
When shooting RAW, ISO means EI, it only affects the brightness of your monitoring LUT and metadata recorded. Use lower EI or underexposed MLUTs to overexpose LOG or RAW in a controlled fashion. Lower the exposure in post to the correct level BEFORE applying LUTS/transforms.