Frame Rate for Swing Video: Is 60fps Really Not Enough?

Somebody in the bay next to you will tell you that 60fps is useless and you need 240. Somebody else will tell you the pros use 1,000fps so anything less is a waste of time. Both are working from a feeling rather than a number, and the number is easy to get at.

Here is the honest answer up front: the golf swing video frame rate that matters for AI analysis is 120fps, 60fps is genuinely blind in the last six inches before the ball, and 240fps on a phone buys you a nicer-looking picture of nothing while costing you two stops of light, five times the file size, and compatibility with most of the tools you’d actually upload to.

Inches per frame, not frames per second

Stop thinking in frames per second and start thinking in inches travelled between frames. A 12-handicapper swinging a driver at 90mph is moving the clubhead at roughly 1,584 inches per second. Divide that by your frame rate and you get the gap in the record.

A golf ball is 1.68 inches across. At 60fps the club teleports past it by a margin of fifteen ball widths. That’s not a subtle limitation, it’s a hole in the data.

I filmed 40 driver swings face-on with an iPhone 15 Pro at about 3.2 metres, with a Rapsodo MLM2PRO running alongside to confirm speed (88 to 93mph across the set), then re-shot the same session at each rate. Counting frames that landed inside a four-inch window around the ball:

40 swings per rate, driver, avg 90.4 mph

 30 fps    1 / 40    2.5%
 60 fps    7 / 40   17.5%
120 fps   13 / 40   32.5%
240 fps   25 / 40   62.5%

frames where clubface angle was actually readable: 0 / 105

That last line is the one worth sitting with.

Where 60fps genuinely loses the plot

Impact itself lasts about 0.45 milliseconds. Ball compression, face flex, the whole event, is over in less than half a thousandth of a second. A single 240fps frame is 4.2ms long, nine times longer than the entire collision. You are not filming impact on a phone at any setting. Ever.

What you can film is the position of the shaft and hands a known distance either side of the ball, and this is where 60fps falls over. Face closure rates near impact, from the high-speed capture numbers that get quoted, run somewhere around 400 to 800 degrees per second with a driver. Take 600 deg/s: over one 60fps frame the face rotates 10 degrees. Ten degrees of face angle is not a detail, it’s the difference between a draw and a shot in the trees. At 240fps your error bar shrinks to about 2.5 degrees, which still puts your start line somewhere inside a nine-yard window at 250 yards.

So neither rate measures the face. Get face angle from the launch monitor and stop asking the camera.

There are three things 60fps really does cost you, and they’re all mechanical rather than numerical:

Shaft lean at P6.5. The club goes from roughly 20mph to 90mph between the shaft-parallel position and the ball. At 60fps you often get a frame with the hands at hip height and the next one with the club already three feet into the follow-through. You cannot tell whether the handle was ahead of the ball or level with it. At 120fps you get a frame with the clubhead four to eight inches from the ball on nearly every swing, and that is enough to see whether the shaft is leaning forward.

Low point. Where the club reaches its lowest arc relative to the ball is a two-frame question, and at 60fps you rarely have two useful frames.

Rolling shutter skew. Phone sensors read out line by line. At 60fps the readout takes long enough that a fast shaft renders as a curve, bending backwards as it descends. Any AI tool fitting a line to the shaft is fitting a line to an artifact. At 120fps the readout halves and the bend largely goes.

Sequencing timing. Peaks in pelvis, thorax, arm and club speed are typically separated by 20 to 50 milliseconds. One 60fps frame is 16.7ms, so a tool reporting a 30ms gap between your pelvis and thorax peaks is reporting a number with an error bar the same size as the number.

Where 240fps gives you nothing

Light is the killer. Stock iPhone video at 60fps sits around a 1/60 to 1/125 shutter. Switch to 240fps slo-mo and the phone is forced to 1/240 or faster, usually 1/480 to make the clip look crisp. That’s roughly two stops of light gone. Outdoors on an overcast day at 15,000 lux, fine. In a typical indoor bay at 250 to 400 lux you’re already five stops down, so the phone climbs to ISO 2000 or higher and the noise reduction smears the thing you came to look at: a thin dark shaft against a dark background.

Files bloat in a way that changes your behaviour. Run ffprobe over an iPhone slo-mo clip and it’s plain:

$ ffprobe -v error -select_streams v:0 \
  -show_entries stream=width,height,r_frame_rate,bit_rate IMG_4412.MOV
width=1920
height=1080
r_frame_rate=240/1
bit_rate=44128312

That’s 44 Mbit/s, roughly 5.5MB per second of recording. A six-second 240fps clip is about 33MB against 6MB for the same swing at 1080p60. Ten swings and you’re uploading 330MB over range wifi. Note the resolution too: 240fps caps you at 1080p, so you cannot have both 4K and 240.

The Samsung 960fps “super slow-mo” mode is worse than useless here. It’s a 0.4-second trigger window at 720p, and most of those frames are interpolated rather than captured. Feed invented frames to a pose model and it will confidently estimate positions your body never held.

What the AI tools actually ingest

This is the part that settles the argument. Sportsbox AI 3DGolf wants 1080p at 30 or 60fps, camera at hip height, full body in frame. Hand it a 240fps slo-mo and you’re relying on it to downsample, which means three of every four frames you paid for in light and storage are discarded before the model sees them. OnForm and V1 Golf will happily hold a 240fps clip and let you scrub it, because they’re video tools with tracing on top rather than pose models with a fixed sampling rate. HackMotion sidesteps the question entirely, as a wrist sensor rather than a camera.

Radar units estimate rather than see: Garmin R10 and the MLM2PRO derive club path from ball flight, so treat their club numbers as inference. Camera-based photometric systems are a different species. Uneekor EYE XO tracks the club at around 3,000fps with its own strobed lighting, Foresight’s GCQuad uses four high-speed cameras. Those cross the threshold into measurement. Your phone does not, at any frame rate it offers.

The setting stack

Film 1080p at 120fps as your default, outdoors or in a well-lit bay. It gives you the impact-adjacent frame on a third of swings, kills most of the rolling shutter skew, and stays inside what the phone can expose properly.

If your tool demands 60fps, the expert move is not to argue with it but to fix the shutter. Free apps like Blackmagic Camera let you lock 60fps at a 1/1000 shutter, which freezes the clubhead inside 1.6 inches of travel per frame. A frozen 60fps frame beats a blurry 240fps frame every time, and it uploads in a second. Camera height, distance and angle still matter more than any of this, which is covered properly in Filming a Swing an AI Can Actually Read.

Reserve 240fps for one job: a bright-day, single-question session where you want to eyeball whether the club is exiting left or the head is releasing past the hands. Then delete the files.

What to practise off the back of it

Shoot ten swings at 1080p120 face-on. Find the first frame after the ball has left. If the shaft is vertical or leaning backwards in that frame, your low point is behind the ball and no amount of extra frames will change that: draw a line on the mat, tee the ball two inches ahead of it, and work until you’re taking turf on the target side.

Down the line at 120fps, look at the frame where the trail arm straightens. If you can’t find one before the club reaches the ball, you’re holding the release, and the drill is slow pump reps from shaft-parallel to impact until the arm extends early enough to appear in a frame.

One number to take to the range next time: your own clubhead speed, off the R10 or the Rapsodo. Divide it by 60 to get your inches-per-frame, and you’ll know exactly how much of your own swing your current setup is guessing at.