Fell off my chair laughing at that one.
I was having a staring contest with the personal text and blinked first.
No offense intended DEJ, I blink first all the time.
Hey I operate a Flame, I'm a badass editor. (Though some may say that the movies I worked on suck.)
I'm pretty sure you took some footage from a different movie, with the same characters, and edited it into the end of From Dusk Till Dawn.
Getting back to the subject, it seems to me that what we'd really like is a camera that would never overexpose a photograph.
That would help with one aspect of the noise problem. Since by definition, a fully overexposed image = 100% noise.
The trick is being able to capture the amount of light that we want, without violating our preferred motion and DOF preferences.
Think of the sunny 16 rule. If you want to have a narrower DOF then f/16, you must adjust the shutter to compensate, which adjusts the capture of the motion. To defeat this adjustment of shutter, you can use an ND filter, but the ND filter reduces the amount of light captured, which means that the light being captured contains more noise.
In this image (representing the noise profile of a 6D @ ISO 100), the red line represents the noise level of the camera, the faint yellow line represents the maximum quality obtainable if all of the noise was only shot noise (Poisson noise), with the faint grey line representing the maximum quality obtainable if the noise was only electronic noise.
Left to right is darker to brighter, with top to bottom representing quality, higher meaning more quality.
The very right hand side side of the red line represents the saturation point of the sensor (at lowest ISO), or the saturation point of the downstream electronics when gain is applied (ISO is increased). A 1EV difference represents a doubling or a halving.
So we can see that at the saturation point of the sensor in this ISO 100 profile, any noise that the camera produces is below the noise level contained in the light itself. The light that we are capturing has more noise then the camera electronics are making. If we look closely enough, it's not until we are four stops below this saturation point, that the noise the camera produces starts to become noticeable. For reference, 18% luminance (midtone) is around 3.5 stops from saturation.
As we move further away from this saturation point, we can see that we are no longer shot noise limited, we start to become read noise limited. In other words, the noise that the camera electronics is making is greater then the noise in the light.
Here is a noise profile from the same camera @ ISO 6400.
If you can understand my nerdy stuff above, two things should be immediately obvious. The saturation point is no longer delivering eight point something EV of quality, only five point something EV of quality instead. And that the image is shot noise limited to around six stops below the saturation point, instead of only four stops @ ISO 100.
Two things are happening here. The first is that the amount of light we can capture @ ISO 6400 is less than the amount of light we can capture @ ISO 100, and hence, the very right hand side of the graph in the ISO 6400 graph is representing a smaller amount of light captured. We can see this being represented by the maximum quality of the image being less than the ISO 100 graph.
The second is that as we increase ISO in current Canon cameras, the efficiency of the downstream electronics becomes greater (they produce less noise), which means that even though we capture less overall light, the light that we do capture is being represented more faithfully. That is, the camera electronics don't affect the image quality until we are at some point further away from the saturation point.
edit: We can also see the difference in quality
for the same amount of light hitting the sensor between ISO 100 and ISO 6400. The difference between ISO 100 and ISO 6400 being six stops. For simplicity lets start at the 13EV line on the vertical scale. Measuring back six stops brings the light level to 7EV on the vertical scale. The 7EV on the vertical scale in the ISO 100 graph is the
equal amount of light hitting the sensor at 13EV on the vertical scale in the ISO 6400 graph. In other words, there's around 0.6EV more quality @ ISO 6400 then ISO 100 for that amount of light hitting the sensor.
This is how we can ETTR with ISO on Canon cameras, because ISO controls an aspect of the overall noise. In Sony/Nikon cameras for instance, ISO does not adjust the amount of noise that the camera produces (mostly anyway), so using ISO on these cameras from an SNR standpoint is useless. The only reason to use ISO on these cameras is so that you can actually see the rendered image on the back of the camera, and I know of quite a few people who are asking the powers to be to address this issue.
edit: So as per my edit above, we can see that as we travel further then 6EV from the from the sensor saturation at ISO 100, read noise (camera electronic noise) becomes a significantly larger proportion of the total noise in the image. Whereas,
from this same amount of light hitting the sensor @ ISO 6400 (13EV on the vertical scale) the noise remains entirely shot noise dominated for a further 6EV (six stops).
So if you're still following along, really there are two aspects that need addressing.
Being able to capture more light. More light is always more better from an SNR standpoint. We shouldn't have to worry about deviating away from our required motion and DOF requirements to respect the saturation point of the sensor. The only concern should be making sure we don't nuke our models out of existence with light. When we can capture more light without having to reduce the shutter time, close the aperture or use an ND filter, we can extend the DR by pushing more signal further from the shot noise limit in the shadows.
Reducing electronic noise. This means that whatever light we do capture, the amount of noise contained in the image is always shot noise dominated.
I didn't intend for this post to be so long, but there you go.
