Camera manufacturers have gotten very good at describing autofocus.
The language is confident: intelligent subject detection, advanced tracking, AI-powered recognition, reliable eye detection even in profile. The demonstrations look seamless. Then you take the camera on a real job and discover that the system has opinions of its own.
I spent years measuring autofocus systems before products shipped and now use them under the less polite conditions of paid work. Here is a translation of the most common claims into what they usually mean in practice.
“AI Subject Detection” / “Intelligent Subject Recognition”
What the claim suggests: The camera understands what it is looking at and will correctly prioritize the important subject.
What it usually means: The camera has been trained on large datasets of faces, eyes, animals, vehicles, and sometimes other objects. It can identify those categories with impressive accuracy when the subject is clear, well-lit, and reasonably large in the frame. It does not “understand” the scene the way a person does. It classifies patterns.
Practical result: Excellent when the subject matches what the system was trained on and the background is not competing. Less reliable when multiple people overlap, when the subject turns away, when lighting is mixed or low, or when something in the background happens to look more like a preferred target (a face-shaped highlight, a second person, a reflective surface). The system is making a statistical best guess, not reading your intention.
“Real-Time Tracking” / “Sticky AF”
What the claim suggests: Once you lock onto a subject, the camera will follow it wherever it goes.
What it usually means: The camera will attempt to maintain focus on the detected subject as it moves across the frame, provided the subject remains sufficiently visible and the motion stays within the system’s ability to predict.
Practical result: Works very well for subjects moving in predictable ways against relatively clean backgrounds. Struggles when the subject is briefly obscured, when another detectable subject crosses in front, or when the motion is erratic. On long jobs I often reduce the aggressiveness of tracking so the camera does not make sudden decisions I did not ask for.
“Eye Detection Even in Profile” / “Animal Eye AF”
What the claim suggests: The camera can find and focus on eyes under almost any orientation or species.
What it usually means: The system can locate eyes in a useful range of angles and, in many current cameras, can identify the eyes of certain animals. Performance drops as the eye becomes smaller in the frame, as the face turns further from the camera, or as light falls off.
Practical result: A genuine improvement over older systems, especially for portrait and event work. It is still not magic. When the eye is partially hidden, poorly lit, or very small, the camera may fall back to face detection or to a general subject area. Checking focus remains your job on critical frames.
“Low-Light AF Down to –X EV”
What the claim suggests: The camera will focus accurately in extremely dim conditions.
What it usually means: Under controlled test conditions with a specific target contrast, the system can still acquire focus at very low light levels. Real-world scenes with low contrast, mixed color temperature, or moving subjects are harder.
Practical result: Modern systems focus in light that would have been impossible a decade ago. They still slow down, hunt, or lock on the wrong plane when the scene lacks clear edges or sufficient contrast. The EV number on the specification sheet is a laboratory best case, not a guarantee for every dim reception hall.
“200+ AF Points Covering Most of the Frame”
What the claim suggests: You can place your subject almost anywhere and still get fast, accurate focus.
What it usually means: The camera has a high density of focus areas and can use them across a wide portion of the sensor. Edge performance is often still weaker than center performance, and the camera’s subject-detection system may override your chosen point if it decides something else is more important.
Practical result: Greater framing freedom than older systems, especially when combined with eye detection. You still need to understand when the camera is likely to ignore your selected point and take control itself.
How I actually use these systems
I leave useful detection tools available—eye detection in particular has earned its place. I turn down or disable the most aggressive continuous-tracking behaviors on jobs where I need predictable control. I keep a simple AF area mode that I can direct instantly with a joystick or touch. And I still verify critical focus when the light is difficult or the moment cannot be repeated.
The goal is not to reject the technology. The goal is to keep the final decision with the photographer.
A clearer way to read the marketing
When you see a confident autofocus claim, translate it this way:
“Under favorable conditions, this system can identify and follow certain subjects with a high success rate.”
“It will still make mistakes when the scene is messy, the light is poor, or multiple possible subjects are present.”
“You remain responsible for noticing those mistakes and correcting them.”
That translation is less exciting than the brochure. It is also more useful when you are the one who has to deliver the files.
The best camera remains the one you understand. Understanding autofocus includes knowing what the system is actually optimizing for—and when its priorities diverge from yours.
