The Tool for the Job
A wildlife photographer once described the experience of switching from an inadequate autofocus system to the right one as “the difference between gambling and working.” That description is not hyperbole. In the fraction of a second between a kingfisher's dive and its emergence from the water, an AF system either acquires and holds subject lock — or it hunts, hesitates, and delivers a technically perfect photograph of an empty river.
Autofocus is the single most misunderstood camera specification in photography. It is reduced, in reviews, to AF point counts and acquisition speeds, as though focus were simply a matter of raw processing power. It is not. Autofocus is a complex system of hardware architecture, machine learning algorithms, and subject-specific tuning — and the manufacturers who excel at it have done so by spending years optimising for particular categories of subject, in particular lighting conditions, moving at particular speeds.
The question, again, is not which camera has the best autofocus. The question is best at tracking what? A Sony A9 III will track a Formula 1 car through a chicane with a reliability that no other camera can currently match. But ask it to softly, continuously pull focus on a walking subject during a handheld video interview and it may stutter and breathe in a way that Canon's Dual Pixel CMOS AF handles with effortless elegance. These are not deficiencies. They are design priorities.
"First, decide what moves. Then, and only then, choose your autofocus."
This article is a guide to the architecture and philosophy behind modern autofocus — the hardware it runs on, the subject recognition models that drive it, and which systems will serve you best for the specific subjects you dedicate your photographic life to.
What Is Autofocus, Really?
Autofocus is not a single technology. It is a layered system in which hardware, firmware, and machine learning interact to answer one question, thousands of times per second: where is the subject, and how far away is it? The answer changes the lens's focus motor position, which changes the plane of sharpest focus in the image. Do this fast enough, accurately enough, and intelligently enough — and the subject stays sharp regardless of how it moves.
Phase Detection AF (PDAF)
The dominant AF technology in modern mirrorless cameras. Phase detection works by comparing two slightly offset views of the same scene through the lens. When the image is out of focus, the two views are displaced. The direction and magnitude of that displacement tells the focus motor precisely how far to move, and in which direction, to achieve focus in a single motor movement, without hunting. This is what makes phase detection dramatically faster than older contrast-detection systems. Modern cameras embed phase-detection pixels directly into the imaging sensor itself: the Sony A7R V uses approximately 693 phase-detect points; Canon's Dual Pixel CMOS AF converts every single imaging pixel into a phase-detection element, delivering complete frame coverage with no dedicated-pixel overhead.
Contrast Detection AF (CDAF)
The older approach: the camera moves the lens until image contrast is maximised, indicating peak focus. Contrast detection is inherently iterative — the lens must overshoot and return — making it slower and prone to hunting on low-contrast subjects. It remains more accurate than phase detection in some edge cases (very close focus distances, extremely low-contrast scenes) and is used as a refinement layer after phase detection in most modern hybrid systems. Panasonic's cameras relied exclusively on a contrast-based DFD (Depth From Defocus) system until the S5 II in 2023, which fundamentally limited their tracking capability for fast subjects.
Hybrid AF
Virtually all modern mirrorless cameras use a hybrid approach: phase detection for rapid initial acquisition and coarse tracking, with contrast detection engaged for final fine-tuning at the moment of exposure. The ratio and weighting between the two modes varies by manufacturer and shooting condition — and this is where significant real-world differences emerge between cameras that appear, on paper, to have comparable specifications.
AI Subject Recognition
The transformative development of the last five years is not phase detection density — it is artificial intelligence applied to subject identification. Modern AF systems identify and lock onto specific classes of subject: human eyes, faces, heads, bodies, birds, animals, insects, aircraft, motorsport vehicles. The quality of a manufacturer's AI subject recognition model — trained on millions of images, refined over product generations, and running in real-time on a dedicated processor — now defines the practical gap between AF systems more than any hardware specification. Two cameras with identical phase-detection coverage can produce dramatically different tracking results if their subject recognition AI operates at different levels of capability.
AF Architectures & Their Tradeoffs
Understanding the hardware underneath the subject-tracking software gives you a clearer picture of what each system can and cannot do — and why the same AI model will perform differently on different sensor architectures.
Sensor Readout Speed
An AF system can only track as fast as the sensor can be read. A camera reading at 30fps updates its AF decision 30 times per second; a camera reading at 120fps updates four times more frequently, giving it a significant advantage with fast subjects. This is why stacked BSI sensors — which achieve dramatically faster readout than standard BSI designs — are so consequential for AF performance. The Sony A9 III's global shutter sensor reads the entire frame simultaneously, eliminating the inter-row delay that causes rolling shutter distortion in standard sensors and enabling AF update rates that are physically impossible on conventional designs.
Dedicated AF Processors
Top-tier AF systems run their subject recognition models on dedicated AI accelerator chips separate from the main image processor. The Sony A7R V was the first consumer mirrorless to include a dedicated AI processing co-processor on the main board, enabling on-board subject detection without burdening the main BIONZ XR image processor. Canon's DIGIC X processors similarly devote significant silicon to real-time subject tracking inference. The practical consequence: high-end cameras run more complex, more accurate recognition models at higher update rates than entry-level bodies, even when those bodies nominally share the same sensor silicon.
PDAF Coverage & Density
Early on-sensor phase detection covered only a central portion of the image frame, leaving corner subjects prone to AF failure. Modern systems from Sony, Canon, and Nikon cover effectively 100% of the frame with phase-detect pixels at densities high enough that even small, partially obscured subjects are reliably acquired. The practical consequence: where you place your subject in the composition no longer determines whether the AF system can find and hold it.
Eye Control & Human Interface
An overlooked dimension of AF architecture is how the photographer communicates intent to the system. Sony uses touch-tap or joystick selection. Nikon inherits DSLR-era 3D Tracking paradigms. Canon's R3 introduced Eye Control AF — the viewfinder tracks the photographer's own iris position to determine which area of the scene to focus on, requiring no button press or joystick movement. For sports and wildlife photographers who cannot divert attention from the scene to operate a joystick during peak action, this represents a genuinely different philosophy of how human intent and machine execution connect.
The Major Manufacturers: Their AF DNA
With that foundation in place, we can assess the AF DNA of the major manufacturers honestly — not as a ranked list, but as a taxonomy of different strengths, priorities, and design philosophies.
The Tracker's Camera
Sony's autofocus is the result of a fifteen-year engineering programme that has produced the most significant AF innovations in camera history. The company pioneered Real-Time Eye AF in 2019 — the ability to detect, lock onto, and track a human eye across the frame with no manual AF point placement required — and that technology has been refined across eight camera generations.
Sony's current subject recognition covers humans (eyes, faces, heads, bodies), animals (eyes, bodies, heads), birds, insects, vehicles, and aircraft — each with its own dedicated neural network model, updated with each camera generation. The A9 III's global shutter sensor enables AF tracking at frame rates that make fast-moving subjects tractable in ways previously impossible: a tennis ball mid-flight, a bird at full extension, a sprinter at peak stride. The camera is not guessing about future subject position — it is updating its focus at a rate fast enough to simply follow.
In short: If your subjects move unpredictably, quickly, or erratically — Sony's AF is the most reliable system available.
The Versatile Tracker
Canon's Dual Pixel CMOS AF is one of the most consequential autofocus innovations since phase detection itself. By engineering every imaging pixel to function as both a light recorder and a phase-detection element, Canon created an AF system with no coverage gap, no dedicated-pixel overhead, and phase information available from every corner of the frame at all times. In practice, Canon cameras acquire focus on subjects positioned anywhere in the frame with equal speed and reliability — a meaningful advantage over systems with coverage gaps at frame edges.
Canon's subject recognition AI is considered the finest for human subjects of any manufacturer. The R3, R5 II, and R1 prioritise human eye detection with an aggression calibrated for professional event and wedding work: if there is a human eye in the frame, Canon's system will find it and hold it, even as subjects move rapidly, turn away, or become partially obscured. Canon's unique Eye Control AF (R3, R1) removes the last physical interface step between observation and focus acquisition, selecting focus zones based on where the photographer's own eye is looking in the viewfinder.
In short: If you photograph people — in any condition, at any speed — Canon's AF is designed to make that the camera's primary obsession too.
The Predictive Tracker
Nikon's approach to autofocus is rooted in its DSLR legacy of sophisticated subject tracking, now rebuilt entirely for mirrorless on a stacked BSI sensor architecture. The Z9 and Z8 combine 120fps blackout-free electronic shooting with EXPEED 7's deep learning subject recognition across humans, animals, birds, aircraft, and vehicles.
Nikon's distinctive contribution is 3D Tracking reimplemented for mirrorless: the system builds a three-dimensional spatial model of the subject's position relative to other scene elements, allowing it to maintain lock across occlusion events — a bird flying behind a branch, a rugby player obscured by another athlete — that cause competing systems to lose and re-acquire the subject. Nikon also offers Pre-Capture: up to one second of frames is buffered before the shutter is pressed, ensuring peak-action moments that occur fractionally before the photographer reacts are preserved.
In short: For photographers who need confident tracking across complex, occluded scenes — and who value the intelligence to capture the moment before they see it — Nikon's Z9 and Z8 are unmatched.
The Improving Pragmatist
Fujifilm's autofocus history is one of the most dramatic improvement stories in modern camera engineering. Early X-Trans bodies had AF systems genuinely inadequate for any subject moving faster than a walking pace: phase detection coverage was limited, subject tracking rudimentary, and low-light acquisition unreliable. Professional wildlife and sports photographers simply did not consider Fujifilm a viable option.
The X-H2S changed that. Fujifilm's first stacked APS-C sensor, with 120fps readout and a completely rebuilt subject recognition AI, delivered AF performance that was, for the first time, genuinely competitive with Sony and Canon for a narrower range of subjects. Human and animal eye tracking is now reliable under normal conditions; bird detection has improved substantially across firmware updates. The X100VI brings face and eye tracking to a fixed-lens compact, making it a credible street and documentary tool.
Where Fujifilm still trails the top tier is in tracking erratic, high-speed, or partially occluded subjects in poor light. The APS-C sensor's smaller photosites mean phase-detection performance degrades faster than full-frame systems as light diminishes. Fujifilm's AI models, while now capable, are trained on narrower datasets and show their limitations at the edges of tracking difficulty.
In short: For street, portrait, travel, and documentary work, Fujifilm's AF is now more than sufficient. For wildlife and action at the highest level of difficulty, it remains a compromise.
The Computational Pioneer
OM System (formerly Olympus) is responsible for a development that is significantly underappreciated in mainstream photography discourse: the company pioneered AI-based bird and animal subject recognition before any other manufacturer — including Sony. The E-M1X, released in early 2019, gained pioneering AI-based bird-in-flight recognition via a late-2020 firmware update, preceding Sony's widely praised Bird Eye AF on the Alpha 1 by a crucial few months.
The OM-1 Mark II continues this tradition with a computational approach unique in the industry. The Micro Four Thirds sensor enables extraordinarily deep depth of field at equivalent apertures, making it possible to track fast subjects in dense environments — birds through tree branches, insects in flight — with a keeper rate that larger-format systems cannot always match despite superior individual AF metrics. The OM-1 Mark II's 50fps burst with continuous AF tracking (and up to 120fps with focus locked) delivers speed capabilities that exceed most full-frame competition, and its ProCapture buffers up to 35 frames before the shutter press.
The trade-off is physics: the smaller Micro Four Thirds sensor performs less well in low light, and AF acquisition in very dim conditions is not competitive with BSI full-frame systems. For dawn or dusk wildlife work in low contrast light, this is a meaningful limitation that must be weighed against the system's considerable strengths in daylight conditions.
In short: For daylight wildlife, macro, and computational high-speed photography, OM System's AI tracking is consistently and significantly underestimated.
The Video-First Camera
Panasonic's autofocus history is a cautionary tale in design priorities. For years, the company insisted that its DFD (Depth From Defocus) contrast-based system was competitive with phase detection. It was not. DFD calculates subject distance by capturing two quick sequential frames at different focus positions and comparing them against a stored database of the attached Panasonic lens's out-of-focus blur characteristics (defocus profiles). This theoretically enables rapid distance determination without dedicated phase-detection pixels. In practice, it was significantly slower, more prone to hunting, and entirely inadequate for subjects moving at speed. Wildlife and sports photographers had to look elsewhere.
The S5 II (2023) was a turning point: Panasonic's first full-frame body with on-sensor phase detection, and the improvement in tracking capability was immediate and dramatic. Human subject tracking is now fully competitive for video work, and stills AF is capable for subjects at moderate speed. For fast action and unpredictable subjects, Panasonic still trails Sony, Canon, and Nikon.
Where Panasonic genuinely excels is in video AF behaviour. The smoothness and naturalness of focus transitions — the cinematic quality of a pull from background to subject — is, for many videographers, the finest of any manufacturer. Panasonic tuned its AF for how video focus is supposed to look, not just for how fast it can change.
In short: For video and hybrid shooters who prioritise cinematic focus behaviour over tracking speed, Panasonic's S5 II and S5 IIX are compelling choices no other manufacturer currently rivals.
Scenarios: Matching AF to Subject
Theory informs, practice decides. Here is how AF system strengths map to the specific photographic scenarios where autofocus is genuinely the deciding factor between a kept image and a discarded one.
Birds in Flight
Sony A9 III or Nikon Z9/Z8. Birds in flight are the ultimate AF stress test: unpredictable direction changes, partially occluded by foliage, small relative to frame, often against complex sky backgrounds. Sony's A9 III offers the most reliable bird eye detection across the widest range of distances, with global shutter readout eliminating the wing distortion that plagues rolling-shutter designs. The Nikon Z9's 3D Tracking excels specifically in occluded environments — birds passing through tree branches — where Sony's system occasionally drops lock. OM System's OM-1 Mark II is a serious contender: its 50fps burst with continuous tracking and the 2x crop reach advantage make it a high keeper-rate system for birds in dense cover under daylight conditions.
Sports & Motorsport
Sony A9 III or Canon R3/R1. Motorsport demands the fastest sensor readout to eliminate banding and distortion under artificial track lighting at high shutter speeds. The Sony A9 III's global shutter makes it the technically definitive answer for eliminating rolling shutter artefacts entirely. For team sports — rugby, basketball, football — Canon's R1 Action Priority AF introduces predictive trajectory modelling that anticipates a subject's next position rather than simply reacting to its current one. Canon's Eye Control AF removes joystick interaction from the photographer's workflow during peak action. Nikon's Z9 with Pre-Capture is the choice when the decisive moment tends to occur fractionally before the shutter is pressed.
Wedding & Events Photography
Canon R5 II or Sony A7 IV. Weddings demand AF that finds the human eye reliably in mixed and shifting light: harsh flash at the reception, dim candle-lit ceremony, bright outdoor group shots. Canon's Dual Pixel CMOS AF is built for this: full-frame coverage with no gaps, exceptional human eye prioritisation, and graceful behaviour under subject transition. Sony's A7 IV brings mature Real-Time Eye AF at a more accessible price point. For documentary-style wedding photographers who also shoot video, the Canon R5 II's smooth video AF makes it a dual-purpose choice that no other system at its price point currently matches.
Portrait & Studio Photography
Any modern mirrorless — but Canon or Nikon for consistency. For controlled portrait work where subjects move slowly and lighting is consistent, the gap between AF systems narrows dramatically. Any current Sony, Canon, Nikon, or Fujifilm full-frame body delivers reliable eye detection for static or walking subjects. Canon's edge is in the naturalness of eye tracking across face angles — holding an eye even as the subject turns profile — and in group scenarios, where it transitions intelligently between multiple subjects without confusion. Nikon's Portrait mode delivers similarly refined face tracking with a preference for the nearest, largest subject that suits environmental portraiture beautifully.
Street & Documentary Photography
Fujifilm X100VI, Sony A7C II, or Leica Q3. Street photography rewards rapid acquisition of unpredictable subjects at moderate distances, ideally without the photographer having to think about focus at all. Fujifilm's X100VI combines face detection with zone AF and a physical aperture-ring workflow allowing hyperfocal pre-setting while face detection handles closer subjects opportunistically. Sony's A7C II offers full Real-Time Eye AF in a compact, inconspicuous body. The Leica Q3's AF is slower than Sony or Canon, but the near-silent shutter and fixed 28mm f/1.7 lens make it the most unobtrusive camera possible in human environments — a distinct advantage when raising a camera changes what the subject does.
Macro & Close-Up Photography
OM System OM-1 Mark II or Sony A7R V. Macro AF demands acquisition of tiny subjects at very close focus distances — where depth of field is measured in millimetres — with sufficient speed to follow subtle subject movement. OM System's Insect Detection mode is unique: the camera specifically identifies insect anatomy, prioritising the compound eyes of a dragonfly rather than hunting randomly across the insect's body. Sony's A7R V delivers accurate macro subject tracking for recognisable animal subjects, and the 61MP sensor provides extraordinary headroom for cropping when AF is operating at the limits of its close-focus capability.
Travel & General Photography
Sony A7C II or Fujifilm X-T5. For photographers whose subjects span landscapes, people, architecture, and opportunistic wildlife in a compact travel package, the modern generation of mid-tier mirrorless bodies is genuinely remarkable. The Sony A7C II's full Real-Time Tracking in a compact body covers virtually every shooting scenario with sufficient capability. The Fujifilm X-T5's face and eye detection is now mature enough for mixed travel demands, and the X-Trans CMOS 5 HR sensor delivers a pixel density and colour quality that makes it the travel body with the highest creative ceiling of any APS-C system currently in production.
At a Glance: AF System Summary
| Manufacturer | AF Character | Human Tracking | Wildlife / Birds | Low Light AF | Video AF |
|---|---|---|---|---|---|
| Sony | Fast, accurate, widest subject coverage | ★★★★★ | ★★★★★ | ★★★★★ | ★★★★ |
| Canon | Full-coverage DPAF, people-first AI | ★★★★★ | ★★★★ | ★★★★★ | ★★★★★ |
| Nikon | 3D tracking, Pre-Capture, occlusion handling | ★★★★★ | ★★★★★ | ★★★★ | ★★★★ |
| Fujifilm | Capable for moderate subjects, improving | ★★★★ | ★★★ | ★★★ | ★★★★ |
| OM System | Computational, pioneering bird & insect AI | ★★★★ | ★★★★★ | ★★★ | ★★★★ |
| Panasonic | Smooth video AF; stills tracking catching up | ★★★ | ★★ | ★★★ | ★★★★★ |
Autofocus in Video: A Different Problem
Video autofocus is not simply stills AF applied to moving images. The requirements are fundamentally different — and a system that excels at tracking a bird in flight may behave badly on camera.
In stills photography, the goal is a single sharp frame. A focus system that acquires in 80ms, drops lock briefly, then re-acquires in another 60ms has failed only if the dropped frames correspond to a decisive moment. In video, that same sequence produces a visible, distracting focus pump that destroys the illusion of cinematic naturalism. Video AF must be not merely accurate, but smooth, natural in its transitions, and appropriately slow to react when a new subject temporarily crosses the frame.
Canon's Dual Pixel CMOS AF has been the video AF benchmark since the original EOS R in 2018. The continuous, fine-grained phase information from every pixel allows Canon cameras to calculate smooth, gradual focus transitions that look as though executed by a human focus puller rather than a computer. Canon's video AF speed settings allow precise calibration of how quickly the system reacts to subject changes — critical for avoiding the jarring focus jumps that occur when an arm crosses the foreground of a static interview setup.
Panasonic's approach, despite historical contrast-AF limitations, produces focus pulls of exceptional cinematic quality. The S5 II's phase-detect system maintains Panasonic's characteristic smoothness while adding the acquisition speed needed for practical video work. For documentary, interview, and narrative filmmaking — where focus behaviour is as much a creative tool as a technical requirement — Panasonic remains a compelling choice that Sony and Nikon have not yet equalled for the specific quality of focus transition feel.
Sony's video AF is fast and accurate but can exhibit a characteristic known as focus breathing: a subtle but perceptible change in apparent focal length as the lens refocuses. On Sony GM lenses this is well controlled; on older glass it can be noticeable in close-up or telephoto work. Sony's Breathing Compensation feature, available on recent bodies with compatible lenses, digitally corrects the field of view change during focus transitions — a software solution to a lens-physics problem that works well in practice, though it slightly crops the frame during the compensation.
Fujifilm's video AF is competent on the X-H2S but is not recommended for critical narrative filmmaking: the AF speed control options are less refined than Canon or Sony equivalents, and the F-Log 2 format requires careful handling in post. Fujifilm's niche in video is its film simulation and colour science pipeline, not autofocus smoothness — a clear expression of the company's priorities.
The Verdict
Every autofocus system represents a set of engineering decisions made in response to a specific conception of what the camera is primarily for. Sony built its AF around the assumption that the hardest thing a photographer will ask of a camera is to track a fast, unpredictable subject through a complex environment. Canon built its AF around the assumption that the most important subject is always a human being. Nikon built its AF around the assumption that the decisive moment may have already passed before the shutter is pressed.
These are not abstract philosophies. They are expressed in the firmware, the training data, the sensor architecture, and the physical interface of each camera system. They produce meaningfully different outcomes in the field — outcomes that determine whether your images are sharp or soft, decisive or missed, publishable or discarded.
"The best autofocus is not the fastest. It is the one that understands what you are trying to photograph."
Before you evaluate any other specification — before you consider resolution, dynamic range, weight, or ergonomics — ask yourself what moves in your photographs. Ask how fast it moves, how predictably, in what light, against what backgrounds. The answers to those questions will eliminate most of the market and leave you with a small number of systems genuinely suited to your work.
Then buy the one with the AF philosophy that matches your own. The camera that understands your subjects before you press the shutter is not a tool. It is a collaborator.