From our ongoing exploration of 10 Things You Should Know Before Shooting with RED Epic:
5. How Epic’s Auto-Focus Works Against You When Off
One of the big reasons the Epic (and Scarlet, too) made such a splash in the low-budget realm was the Canon lens mount swappable with the standard PL Mount. Suddenly, thousands of filmmakers who already owned still lenses could put their glass to use on a high-quality digital cinema camera.
And, with that, a collective groan from camera assistants was heard ’round the world.
That’s because still lenses are a fickle bitch compared to their cinematic counterparts when it comes to filmmaking — the iris is operated internally, the focus ring often free-spins, and the throw between distances focus-wise is absurdly small.
With Epic, all those problems exist, as well as another one I’d like to bring to your attention today.
(Just a quick heads-up, this article isn’t about how to use the Epic’s auto-focus, but rather how it affects you when you are pulling focus manually and also the theory behind the mechanism.)
The Most Annoying Part of Pulling Focus Manually on RED Epic with Still Lenses with Auto-Focus Off
A few weeks ago, I found myself on a set in New York working on an Epic with a few Canon zooms. Because what we were shooting was largely unscripted B-roll, I was worried very little about the usual trappings of still lens filmmaking.
But that doesn’t mean I didn’t notice the potential pitfalls if we had been shooting scripted material.
The main annoyance that arose was the Epic’s undeniable urge to reset the focus ring on the lens everytime it was powered down and booted back up. I had made sure to set the focus mode to “MANUAL,” in which absolutely everything is manual with no assists, but without fail, the camera would still reset the focus ring to its closest distance on boot-up.
Basically, everytime you restart the Epic with a still lens, it will reset the focus ring to its closest distance to calibrate the auto-focus should you decide to turn it on.
While I understand the mechanical need for this process, the lack of need for it on our shoot couldn’t be more overstated.
Take, for instance, a shot we had lined up and rehearsed with a few actors posing on a guard rail looking at the Manhattan skyline. With a small slider, there was enough of a focus pull to justify making marks on the lens with a grease pencil. Using the 1:1 zoom, I got my marks and we rehearsed the shot.
Then the REDVOLT battery reached 5%, so I swapped it out.
As the camera booted back up, I watched it — against my will — spin the focus ring and mis-align all my marks. We were back at square one.
This phenomenon is a combination of most still lenses having a free-spin focus ring (as in, it will continue spinning even after you have reached infiniti) combined with the Epic’s complete disregard for where the lens was focused before you shut it down.
The fact that nobody else talks about this blows my mind because it was increasingly frustrating.
If there is a way to prevent this from happening, please let me know in the comments. I’d much rather admit I’m wrong on this point and find a solution than to know this is the way it actually works!
Bonus Lesson: How the Auto-Focus Mechanism Works
While writing this post, I did a few Google searches for better insight into how auto-focus on the Epic worked. Because I’m not a lens-technician, I was curious of the process the camera and the lens go through to achieve sharp focus automatically.
So I was pleased to stumble across a well-informed post over at REDUser.net by Alexander Imbrahim. Even though it doesn’t directly relate to the issue I discussed above, it’s still incredibly fascinating to read Alexander explain the process behind the Epic’s auto-focus. Here’s his post:
Remember that DSLR lenses are not meant for continuous focus tracking. They are meant to focus at discrete instances and snap off a shot.
As a result they have different operating characteristics, many of which are non-linear. In English, the position of the AF servo doesn’t have a direct correspondence to the focal distance.
DSLR’s dont “care” because they have time to hunt for maximum AF sharpness between frames. Motion imaging cameras don’t have that liberty … they have to continuously be in focus.
My understanding is that Red cameras address this in one of two modes, which could be called “knowing” and “guessing.”
With a fully characterized lens, the camera can make the most accurate corrections straight away, because it knows which lens motor position corresponds to a given distance.
In the guessing mode, the autofocus still works, but it hunts more. So as a subject moves … the camera doesn’t know how much to move the servo to make it be focused at a certain distance. So, it moves the servo, and then evaluates the image for sharpness at the AF point. (i.e. where you touch the monitor) Then it adjusts the servo again … repeating this until the camera decides the image is in focus.
This guessing mode is ALWAYS how DSLR cameras operate. Thus they overshoot focus and come back to sharp, sometimes several times. We call that “hunting.”
This is all compounded by the fact that DSLR lens manufacturers send metadata about focus and focal length back to the camera that isn’t quite right. Photographers don’t use that data the same way cinematographers and motion imaging post artists do, so they don’t need accuracy. They effectively just need a gross reminder.
Red has to fix all that for these lenses to work using their built in features for cinematography.
DSLR cinematography almost always depends on a spectacularly powerful computer with a fantastic bit of software to manage all that. Its called the first assistant camera’s brain. As a 1st AC on DSLR projects, you learn what lenses will do, and you start to move non-linearly in response it works out. (By the way … this is a fantastic reason to spring for some manual Nikkor AI or Zeiss primes as opposed to cheap modern AF lenses.)
Red has to recreate that, but with a lot less processing power and flexibility than a human brain.
One thing they must do is empirically measure the actual FOV and focal distances at each FOV of DSLR lenses. Primes aren’t so bad … a Canon 50mm prime is pretty reliably 50mm, so you only have to calibrate the entire focal range in the smallest steps the motor can give. I don’t actually know the precision of DSLR lenses, but a 14 bit 2/3″ ENG lens has 16384 positions. For each position, Red must map the actual focused distance against the reported distance in metadata. Then Red has to use mathematical regression to figure out a curve that can be calculated quickly that accurately represents how motor position corresponds to actual and reported focal distance.
For a zoom lens … that becomes hugely more completed. Taking the 24-70, you have to divide the focal length range and divide it into a large number of steps. (I’d pick the steps available in typical cine zoom motors, which I believe is 16 bit or 65536 positions, then for each of them plot true focal distance versus motor position and reported focal distance.)
That could be over a billion data points which have to be accurately regressed into a polynomial.
That also makes clear why you’d want to calculate this data as opposed to looking it up. One focal position needs 16 bits for focal length, 32 bits for true focal distance (INT in mm), 16 bits for motor position, 8 bits for aperture. That means a billion position table would be 9 Gigabytes. For every zoom lens.
Now, my numbers here are just estimates. I didn’t look up the precision of any of these digital motors … I also didn’t account for the possibility that analog motors are in use. I also didn’t account for lenses whose focus characteristics are affected by aperture. (A cine lens or even a good manual SLR lens wouldn’t be, but in search of lower costs some DSLR lenses are affected by aperture and compensate with AF.) I also didn’t account for breathing, which also affects focus accuracy.
So, that’s an outsiders guess at why they characterize lenses, and why it takes so damn long.
What’s important to note in Alexander’s post are two ideas, one of which will be popular with readers of this blog:
1.) Auto-focus is a complex computational calculation dependent on a camera’s system and lens being able to exchange information about each other freely.
2.) Auto-focus, at least on RED Epic, is not an absolute replacement for the instinct and skill of a camera assistant.
So if you’re a producer, director, or cinematographer thinking auto-focus might save you from shelling out the dough for a camera assistant, well, I don’t think that’s going to happen anytime soon.
And even if the auto-focus were perfect, you still need a crew member who understands the technology inside the camera and the practical applications of said technology. That person — whether they ever pull focus or not — is likely going to be a camera assistant.
There will be some shoots and sets that can get by with the Epic’s auto-focus, but I imagine these are the same shoots already choosing not to hire focus pullers because the demand for intricate focus pulls just isn’t in their shotlist or script.
That’s fine and completely understandable.
I just want to emphasize that the more complicated your shot, the shallower your depth-of-field, the more spontaneous the movement within the frame is, the harder it is going to be to depend on computerized auto-focus than a camera assistant’s touch.
So Let’s Review…
Because this post has been several times longer than the other posts in this series, let’s take a look at the quick takeaways from it…
- Still lenses, even when in manual focus mode, will re-calibrate their focus whenever the camera boots
- Not all still lenses will have the same degree of auto-focus accuracy with RED Epic
- The Epic in auto-focus mode is not a guaranteed replacement for camera assistants or focus pullers
Phew.. ok that was a lot to write just to convince you to shoot cinema lenses instead :P
After all, you know how I love my Superspeeds!
How often do you find yourself working with still lenses on a cinema camera? Or cinema lenses on a stills camera? Share your thoughts on the two styles in the comments!