Face-On and Down-the-Line: The Only Two Angles You Need

Your phone is fine. Genuinely. The iPhone 13 sitting in your golf bag records 240fps slow motion at 1080p, which is more than enough temporal resolution for every AI swing analyser on the market to track your hands, your shaft, and your hips through impact. The problem was never the camera.

The problem is that you put it three feet too far left, tilted it up to get your head in frame, and now the AI thinks you’re 14 degrees over the top when you’re actually neutral.

I want to make an unpopular argument here: golf swing camera angles matter more to your AI feedback than every spec on the box. Sensor size, frame rate, lens quality, HDR, the works. If your camera is in the wrong place, a £900 setup gives you worse data than a five-year-old Android propped against a bag with a towel, placed correctly. And “correctly” is a much narrower window than most club golfers assume.

Why the AI cares so much about where you stand

Every pose-estimation model used in golf apps (Sportsbox 3DGolf, Swing AI, HackMotion’s video companion, the analysis layer in Rapsodo MLM2PRO, Golfer’s Swing Buddy) works by finding 2D joint positions in each frame and then inferring depth. Sportsbox is the most aggressive about the 3D reconstruction; it claims millimetre-level measurement of pelvis sway and thorax bend from a single camera. That claim holds up reasonably well, but only inside its calibration assumptions, which are: camera at a known height, at a known distance, square to the line it thinks it’s square to.

Here’s the failure mode nobody mentions. Pose models estimate depth from the apparent size and overlap of body segments. When you shift the camera off the target line by even a small amount, the model’s depth priors get systematically biased, and that bias shows up as a plane or path number. It doesn’t show up as an error message. The app returns a confident measurement that is simply wrong, and you go and practise against it.

A rough sense of scale from my own testing over a few range sessions with a 7-iron, filming the same swing from deliberately wrong positions:

That third row is the one that ruins most people’s practice. Move the camera toward the target and the shaft appears more leaned than it is, because you’re looking at the shaft from a rotated viewpoint. Move it away and you get the opposite. Neither error is visible in the video. You look at the frame and think “yeah, that’s my swing.”

Down-the-line: pace it out, don’t eyeball it

Stand behind the ball. Your camera needs to sit on the extended ball-target line, behind the ball, at a specific height and distance. Every word of that matters.

Finding the line. Lay a club on the ground pointing at your target, ball on the toe-end. Walk back along that shaft’s extension. Not roughly parallel to it: on it. The single most common DTL mistake among the golfers I’ve watched film themselves is standing behind their own feet rather than behind the ball, which puts the camera about 20 to 24 inches inside the line for an iron. That’s the top row of the table above, and it’s the reason so many people think they have an out-to-in path they don’t have.

Distance. Eight feet back from the ball, measured properly. Pace it out heel-to-toe: an adult foot is roughly 10 to 11 inches in shoes, so nine to ten heel-to-toe steps gets you close, or use the club you’re hitting. A 7-iron is about 37 inches, so two and a half club lengths plus a bit. Too close and the lens distortion at the edges of frame bends your shaft. Too far and the app loses joint confidence on your hands, which is exactly where you need precision.

Height. Hand height at address. For most club golfers that’s 32 to 36 inches off the ground. This is non-negotiable and it is the rule that separates usable AI data from noise. Put the camera at hand height and the shaft plane you see in the video is the shaft plane that exists. Put it at chest height because that’s where the bag strap happens to hold your phone, and the entire swing looks shallower than it is. Sportsbox will tell you off for this directly, and it’s right to.

Orientation. Portrait, not landscape, for DTL. You need the full arc of the club above your head and the ball at the bottom of frame. Landscape crops the top of the backswing on almost everyone.

Here’s a rig that works and costs nothing. Phone leaning against your golf bag, bag laid flat on the turf, phone propped at hand height on a wedge head or a stack of two headcovers. Check the height with the grip end of the club you’re hitting: hold it against the lens, the butt should sit level with your hands at address.

DOWN-THE-LINE (viewed from above)

        target
          ↑
          |
          |   ball
          |    ●
          |   /|\  ← you, feet parallel to line
          |
          |
          |  ← 8 ft on the extended line
          |
        [📱]  camera, portrait, 34" high

Face-on: the reference is your hands, not your chest

Now rotate the whole picture 90 degrees and change what you’re aiming at. For face-on, the camera goes perpendicular to the target line, directly opposite your hands at address. Not your sternum. Not the ball. Your hands.

This distinction sounds pedantic and it isn’t. With a 7-iron, your hands sit maybe four to six inches ahead of the ball and your sternum sits behind them. If you line the camera up on your chest, you’ve introduced a small parallax that gets read as shaft lean, weight shift, or hip slide depending on which app you feed it. Line up on the hands and the vertical centre of your frame becomes a genuine reference line for the whole swing.

Distance: ten feet, a little further than DTL because you need width. Your trail side needs room at the top of the backswing and your lead side needs room at the finish. Height: same as before, hand height. Orientation: landscape for face-on, because the swing’s widest dimension is now horizontal.

The tilt rule deserves its own sentence. Keep the phone perfectly vertical. Not angled up to fit your head in, not angled down to fit the ball in. If you can’t get both in frame, step back another two feet. A 12-degree upward tilt tripled the sway measurement in my testing, and sway is the single metric that club golfers most often try to fix based on video. You end up drilling a fault that doesn’t exist while the actual problem (early extension, usually) sits unaddressed.

FACE-ON (viewed from above)

              target →
    ●  ← ball
   /|\ ← hands here
    |
    |
    |  10 ft, perpendicular
    |
  [📱] landscape, 34" high, zero tilt

What each angle can and cannot tell you

Getting the placement right earns you the right to trust specific metrics, and only those metrics. This is where most club golfers over-read their footage.

From down-the-line you can trust: shaft plane at halfway back and halfway down, club path tendency, head depth (whether you’re moving toward or away from the ball), early extension, and the relationship between your trail forearm and the shaft at the top. If a tool reports these from a properly placed DTL camera, take it seriously.

From down-the-line you cannot trust: weight distribution, hip sway, or anything about lateral movement. You’re looking down the axis of that motion. Apps will still give you a number. Ignore it.

From face-on you can trust: sway and thrust (pelvis lateral and vertical movement), head movement, shaft lean at impact, tempo ratio, and where your low point sits relative to the ball.

From face-on you cannot trust: club path, face angle, plane, or anything about the club’s relationship to the target line.

Which brings me to the actual practice recommendation, and it’s short. Film both. Every session, same two positions, same two distances, same height. The reason isn’t completeness for its own sake; it’s that your AI tool’s confidence scores improve dramatically with consistent framing, and more importantly, you can compare week three to week one and know that any change in the numbers is a change in your swing rather than a change in where you put the phone.

Uneekor, Foresight and Trackman all give you club path directly and you don’t need video for it. What video gives you, correctly framed, is the body sequence that produces the path. That’s the loop: launch monitor tells you the ball did something wrong, DTL video tells you the club did something wrong, face-on video tells you your pelvis did something wrong, and the pelvis is the only one of those three you can actually practise.

A worked example from a 14-handicapper

A mate of mine, 14 handicap, Rapsodo MLM2PRO owner, convinced for about eight months that he had a steep, over-the-top move. He’d been filming DTL from behind his feet, at roughly chest height, four feet back, because that’s where his bag naturally stood. Every video confirmed the diagnosis. He’d taken maybe 200 practice swings drilling a shallowing feel that made his strike worse and cost him about 8 yards of carry on a 7-iron.

Twenty minutes of repositioning: eight feet back, on the extended line found with a club on the ground, lens at 34 inches. Same swing, same day, same session. His plane was fine. The shaft sat right on plane at P5. What the new footage showed instead, immediately obvious once the angle stopped lying, was his trail hip moving toward the ball in transition, which was making the club feel steep while being nothing of the sort. He was compensating with a flip, hence the thin strikes.

Two months of a single drill (trail glute against a chair back, half swings) and he’s a 10.4. The camera didn’t get better. It moved four feet backwards and dropped 18 inches.

If you want the rest of the setup detail, including frame rate settings per app and the lighting conditions that break pose estimation, that’s covered in Filming a Swing an AI Can Actually Read.

Your next range session

Bring a tape measure the first time. Measure eight feet for DTL, ten for face-on, and measure your own hand height at address with a 7-iron. Write those three numbers on the inside of your bag tag or in your phone’s notes app. Mine read: 8ft / 10ft / 34in.

After that you never measure again. You pace it: nine heel-to-toe steps back for DTL, eleven for face-on, and you know from muscle memory where the lens needs to sit because you’ve checked it against the grip end of your club twenty times.

What you’ll notice within about three sessions is that your AI tool’s outputs stop contradicting themselves week to week. That consistency is the whole game. A tool that tells you the same thing twice is a tool you can build a practice plan on, and no amount of sensor quality substitutes for a phone that’s standing in the same place it stood last Tuesday.