As someone at the war front—that’s the autofocus capability battle line—it strikes me that many of you need a more accurate guide to what’s happening there than a bunch of tweets and videos from random sources, some of them anonymous, some of them biased.
I say that because while looking at today’s Internet scrapings I found YAFV (Yet Another Focus Video). In fact, today produced four of them. Two for, and two against.
Wait, what?
Two of the videos—one was from a self-reported influencer, so take that for what it’s worth—basically were of the “absolutely amazing autofocus, it follows everything perfectly” variety. The other two—one from a long-time, top-level content producer—were essentially “here’s everything you need to know to get decent performance out of [product name’s] focus system.” In other words, half the videos were of the “it just works” variety, while half were of the “it doesn’t work unless you take control” variety.
Just to be clear up front, I’m in the “it doesn’t work unless you take control” camp. In every camera I’ve tried, and that’s now measured in the hundreds, I’ve always found that I can get better results by studying, practicing with, and mastering the nuances of taking control from the autofocus system. That includes the Nikon Z9 and Sony A1.
If all you’re doing is reading randomly on the Internet, you’re probably confused by today’s autofocus systems. I get dozens of emails a week that seem to verify that. And heaven help you if we start talking about the autofocus system of a previous generation, as according to the dogs on the Internet, that must be trash.
So let’s try to state some truths. Yes, these are truths as I know them. Based upon decades of personal experience, decades of working with others, and decades of writing about the nuances of autofocus systems.
- Autofocus has constantly evolved. From its early appearance in the 1980’s, autofocus has been one of the most researched, developed, and deployed changes in interchangeable lens cameras. From single sensors to multiple, from single focus to continuous, from following to predicting, from brainless to subject aware to subject detecting, from screw drive motor to wave to stepper to voice coil motors, we’ve been subjected to a constant stream of autofocus system changes. I haven’t even begun to list all the things that were being iterated, as the list is just too long for even one of my articles ;~). “Evolved” is the correct word, too, as the strongest have survived, while those unable to adapt have died off.
- Autofocus has always worked. You might start off doubting me on this one, but the operative word in this assertion is “worked.” Just as your office has some not-so-hard-working folk and some hyperactive-always-working folk, how well autofocus systems “worked” has changed over the years. But they always worked according to design. For example, the very first autofocus system I remember using in the 70’s “worked” to get focus acceptably close on the thing I pointed it at, to which I always just added a quick manual retouch to get pinpoint focus. Over time, autofocus systems worked at more things and in more cases and in better ways. What I had to do in interacting with them and controlling what they were doing changed over time. I remember the first time that a Nikon body did the right thing with Closest Subject Priority when I wasn’t expecting that given previous autofocus systems. Instead of correcting the system, I was scratching my head wondering how it did what it did. Of course, not every composition needed focusing on the closest subject, so I instantly got back to the study, practice, master sequence I’ve now espoused for decades. For those of you still not getting it: it’s how well the autofocus system worked in each and every situation it is presented that’s been changing. But they’ve always done what they were designed to do. That just might not be what you expected or needed.
- Autofocus has tolerances. No matter how you design a system, you eventually run into the price/performance/speed type of conundrum. Pick two. Maybe it’s not those three specific things that start to come in conflict, but it will be three of something, and when you hit that design problem, you have to make compromises. We usually see that in autofocus systems as tolerance of something. For instance, some autofocus systems will stop if they detect that they’re within some level of depth of field (unspecified to the customer!). Some will assume that an object couldn’t have moved faster than they can detect so keep focus in a predictable place or with a particular speed of change. The list goes on. At some point the 1’s and 0’s essentially line up in a way that the camera says “okay, I’ll live with that.” The interesting point is that three decades ago the user tolerance was “did it focus on the face?” where today the user tolerance has shifted to “did it focus on the pupil of my selected eye”? This makes the tolerance game a bit of back and forth between camera maker and camera user.
- Autofocus makes assumptions. The one that started having a real impact on DSLR users was often one or a combination of two things. With fast lenses the phase detect cells didn’t properly see the outer edge of the lens due to the restrictions in the mount cutoffs and multi-mirror light bounce, and many lenses had focus shift in them. It took me a long time to figure out all the variables that were going on. One of the first DSLR pros I helped at the turn of the century was struggling with back focus in some situations and front focus in others. Early Nikon’s didn’t have AF-Fine Tuning, but even more important, different uses of the lens in question were producing different results, which couldn’t have been tamed with a single AF-Fine Tuning to start with. Later in the Nikon DSLR history we learned about a hidden table in the camera when about a third of the D800’s out of the factory couldn’t focus correctly on their left side. Turns out that a machine in manufacturing that was populating that table with data based upon how the focus and image sensor were aligned was programming the data wrong. So the focus system was assuming it had been set up right, but it simply wasn’t!
- Autofocus requires data live in a certain range. This gets tricky. For instance, today’s mirrorless cameras generally need the subject to be properly exposed. Too little exposure and noise dominates the data. Too much exposure and non-linearity dominates the data. It isn’t so much about how little light there is in low light, it’s about whether at the camera settings the data stream is usable by the autofocus system. I’ll give you one such example: on the Nikon Z9 the focus data stream comes from the image sensor in real time separate from the image data (the sensor data is duplicated into two separated channels, one for the focus/viewfinder, one for the image). If you’ve set the camera so that the subject is going to be severely underexposed, by the time the data stream is demosaiced by EXPEED and gets to the focus component there might not be enough contrast—particularly on dark subjects—to make a proper decision on focus.
- Subject detection autofocus is generally using a form of Machine Learning. Machine Learning (ML) is a way of teaching a system about something. You generally train these systems up until they meet some threshold of acceptability you’ve set, but usually you don’t attempt to get them to 100%, as that can introduce other issues you have to deal with. Indeed, what you want is “most of the time it recognizes the subject, but it doesn’t produce (m)any false positives.” The good news is that the camera companies do retrain their autofocus detection systems these days as they learn about interactions and issues that arise, and issue firmware updates that make the cameras get better.
- You and I will disagree on exposure, and we’ll disagree on focus. There’s this myth that there’s a perfect exposure for any scene, and that’s now expanded to include the myth that there’s a perfect focus for any scene. Even an individual photographer may change their mind: over his lifetime, Ansel Adams tended to print his images with deeper blacks and more contrast as he re-worked them. (This is the genesis of my insistence on “capture optimum data.” You can’t make some decisions in post processing unless you captured optimal data from which to do so. If your data is sub-optimal, some later choices will be either more difficult to accomplish, or worse, impossible.) Over time, many of you have come around to agreeing that “exposure” is in the eye of beholder. You can’t have high key versus noire takes on a scene without exposure and post processing decisions differing, after all. Focus not being an absolute is tougher to explain, but perhaps you’d agree that depth of field is indeed variable, and so what’s “acceptably in focus” may vary between me and you. More nuance than that is a subject for another article. However, the reason why this point is important is:
- You and the camera will sometimes disagree about focus. We saw this a bit when the camera makers first touted “eye detection” but astute observers noted that this usually meant “eyebrow detection.” It may come as a surprise to you that I’m not an absolutist about focus point position. The example I’ll give is the elephant. More often than not having the camera focus absolutely on the elephant’s pupil—which, by the way, almost no autofocus system will do—is not what I’d want. The most important details on an elephant tend to be anterior (trunk) and posterior (ear) to the eye. Which I might want to emphasize, typically through depth of field, depends upon what my composition is trying to say. Your composition may need a different focal (focus) point. As with composition (rule of thirds), focus has lots of shortcuts (eye in focus, closest thing in focus) that can lead you very astray in creating great images.
So what should we conclude from the above? I’ll tell you what I’ve concluded, practice, and preach: you need to know what your autofocus system is doing and why, plus when and how you should override it.
Which brings me to this:
- Those who have had trouble focusing in the past love current autofocus systems.
- Those who haven’t had trouble focusing in the past love current autofocus systems.
The difference between those two tends to be that proclamation of “amazing autofocus” (first bullet group) versus the “here’s what you need to know about autofocus” group (second bullet group) that I noted up front in this article. In other words, some folk are getting better results from current systems than they’ve gotten from previous ones, while others enjoy the better recent autofocus systems yet still take matters into their own hands when the camera doesn’t quite do what they want. I’m tempted to write “novices” versus “pros,” but that might sound a little too much like looking down my nose. Nevertheless, there’s some truth in that.
Finally, I have to point out something else. Those in the first group (past autofocus was challenging, now solved by modern systems) have a tendency to conflate brand with success. They used a Canon or Nikon DSLR in the past, struggled to get good focus, then more recently switched to a Sony mirrorless camera and are achieving better results. This leads to many of the “Sony has the best autofocus” comments you see (the others tend to be Fan Boy trolling).
As someone who’s used every brand, both historically as well as today’s latest products, my assessment is different. I’ll specifically speak to the Nikon Z9 and Sony A1, the two top models from those brands that have their best autofocus systems to date (I’d even throw in the Nikon D6 DSLR, which is the ultimate DSLR solution). Each is really good. Each produces slightly different results if left in all-automatic. Neither is perfect. And for each I get the best and virtually identical results if I do what I preach: learn, practice, master. Learn how the system works and what it can and can’t do. Practice with the system so that your learning of it is reinforced. Over time configure your system so that you can take control and alter what the system is doing when you need to.
This, by the way, is one reason why “how’s it handle” is one of the key points in my reviews, of both cameras and lenses. If I need to be in control, I need that control to be flexible, full, and at my fingertips.
/Truth out
