2026년 7월 23일 • ☕️ 7 min read
Record a weekend tennis match and you’ll walk away with an hour of footage, easily. Play it back later and you’ll notice something deflating: the ball is in play for maybe fifteen minutes of it. The rest is collecting balls, drinking water, switching ends, and standing around replaying the sitter you just missed.
The trouble is that those fifteen minutes are the only part anyone wants. Checking your form, showing a friend, posting to YouTube — what you need is the rallies, not the water breaks. And getting those fifteen minutes out turns out to be harder than it should be.
For a while I cut the footage by hand. Then I got tired of it and built a tool, and somewhere in the process of feeding that tool a few hundred match videos I learned a second lesson: how you film matters more than how you edit. The fate of a video — whether it edits cleanly or fights you the whole way — is mostly sealed the moment you set the camera down.
So this post comes in two halves. The first compares three ways to edit a tennis match down to its rallies. The second covers the filming habits that make any of those methods work. One confession before we start: one of the three methods is a tool I built, and I’ve tried to spend more words on its weaknesses than its strengths. Discount accordingly.
The most reliable method, and by far the slowest.
You drop an hour of footage on a timeline and scrub through it, marking where each rally starts and ends. An amateur match holds somewhere between 40 and 60 rallies, and cutting them one by one takes two or three times the length of the video. An hour of footage means an evening of editing.
If time were the only cost, you could live with it. It isn’t. Does the bounce before a serve count as part of the rally? What about a point that dies in the net? You end up relitigating the rules with yourself on every cut. Somewhere around the fortieth rally your attention gives out and good points start sliding past unnoticed. And the fatal part: having done it once, you will never want to do it again. The footage piles up in your camera roll, and nobody watches it.
Best quality of the three. Just not something you can sustain every week.
Upload the whole thing and let the machine sort it out. Short version: it doesn’t work for tennis.
Auto-highlight features generally key off crowd noise, commentator excitement, score graphics — signals that broadcast footage has in abundance and an amateur court has none of. Faced with a video that offers no crowd, no commentary and no captions, these features have nothing to hold onto. They can’t tell a rally from a water break, and the output comes back looking much like the input.
It’s simply a different job. Those tools hunt for moments worth cheering; we’re looking for the stretches where the ball is in the air.
This is the part I built. It’s called OnlyRally.
The core idea is almost embarrassingly simple. Two signals are enough to separate a rally from downtime. One is player movement — during a rally, players cover real ground; between points they mostly stand still. The other is the sound of the ball, because nothing else on a court sounds like a racket meeting a ball.
Find the stretches where both signals are alive, splice them together, and that’s your highlight reel. The user does exactly one thing: upload the match. No marking corners, no timeline, no reviewing rallies one by one. The output keeps the original quality, too — no re-encoding mush.
I set one rule for myself while building it: after upload, the user does nothing. The moment you ask someone to click “cut here,” you’ve walked them right back to Method 1.
Since I promised to dwell on the weaknesses.
The failure that hurts most is a lost rally. Mistake a good point for downtime and it’s gone for good, so the tool is tuned to keep anything ambiguous. The price of that choice is that a little dead time occasionally sneaks into the output. It is not surgical yet.
Bad footage in, bad output out. A shaky camera, a court sitting tiny in the corner of the frame, wind burying the sound of the ball — any of these and detection wobbles with it. Indoor courts remain genuinely hard: the ball echoes off the walls, sounds bleed over from the next court, and false positives creep up.
Which brings me to the second half.
Plenty of people record their tennis. Very few record it with the edit in mind. I certainly didn’t — I used to prop my phone against a fence and hope, and what I accumulated was a pile of footage where players wandered out of frame and the image shook too much for even me to tell where the rallies were.
Remember the two signals — movement and ball sound? That isn’t just how my tool finds rallies. It’s what your own eyes look for when you edit by hand. And bad filming kills those signals first. Hold the camera and the whole frame shakes, blending the players’ movement into the camera’s. Frame the court too small and a player’s footwork shrinks to a few pixels of flicker. Let wind hit the microphone and the ball sinks under the noise.
Capturing clean signals is eighty percent of the edit. Here are the rules I keep now.
If I could keep only one rule, it would be this one. Two to five meters behind one baseline, centered left to right — the spot where the camera lines up with the middle of the net.
Get too close and the near player fills the frame while the far court disappears. Drift to one side and the opposite sideline gets clipped. From the right spot, both players stay in frame for the entire rally, and “he ran out of frame, this point’s unusable” stops being a sentence you say.
Leave the camera on the ground and the near player’s body eclipses the far player. A little above standing eye level — 1.5 to 2.5 meters — gives you a gentle downward angle where the two never overlap.
Extend the tripod as far as it goes, or set the camera on a fence or a bench. But remember: stability beats height.
If you take one thing from this post, take this. Lock the camera down.
Handheld footage — or worse, a friend panning along with the action — poisons the movement signal exactly as described above. This is where automatic editing accuracy collapses fastest. No tripod? Use a gimbal. No gimbal? A wall, a fence, a bag. Anything that doesn’t move.
Shoot landscape; vertical video clips both sidelines, and any rally that goes wide simply exits the frame. 1080p is plenty — 4K makes the files heavier and the upload slower without improving detection at all. And let the court fill 70–80% of the frame’s width. When sky and fencing eat half the picture, the part that matters ends up at postage-stamp resolution.
The ball’s sound is half of rally detection, and it’s the half everyone ignores.
On windy days, face the mic away from the wind or fit a windscreen — wind hitting the capsule directly buries the ball entirely. Set up on the side farthest from neighboring courts, whose ball sounds otherwise bleed in and masquerade as your rallies. Indoor courts are a hard environment to begin with, so get close and keep the mic facing the play.
I have personally made every one of these.
The editing choice comes down to three options. For occasional, maximum-quality edits, cut it yourself in CapCut or Premiere and budget the evening. Don’t bother with auto-highlight features; they weren’t built for amateur tennis. And if you want rallies-only every week without the chore, use a tool where uploading is the whole job — I couldn’t find one, so I built it.
Whichever you choose, the filming rules fit in three lines:
Looking back, what decided my editing method was never quality. It was whether I’d still be willing to do it next week. The finest result means nothing if you can’t face doing it twice. And five minutes of care at filming time buys back an hour at the edit. Those two sentences are the whole post.