Short answer: record 20-30 minutes of a hitting session with the phone in landscape on a tripod above net height, upload it, and read the report in this order — heatmap first, shot-mix second, one stroke-quality breakdown third. Don't try to read the whole report. Don't compare to anyone. Don't change your strokes after one report. The first session is calibration; sessions two and three are where the pattern shows up.

This post is the longer version. If you've just signed up, this is the one to bookmark.

TL;DR — the actual 5-minute checklist

Minute 0-2: Phone in landscape on a tripod, 6-10 ft behind the baseline, above net height. Press record.
Minute 3-25: Play tennis. Don't film the warm-up. Do film at least 20 minutes of actual rallies or sets.
Minute 26-28: Stop recording, open AceSense, hit "Upload video". Wi-Fi recommended for first upload.
Minute 29: Pipeline runs. You can close the app — push notification when ready.
Minute 30 (or whenever the report's done): Open the report. Read the heatmap, then the shot mix, then one stroke-quality breakdown. Stop there.

Why a starter guide exists in the first place

The "first time using" problem is real and well-documented. There's a r/10s thread literally titled "First time using SwingVision" where the user is asking what to do with the report they just generated. There's a follow-up thread "Tips when using SwingVision" asking the same thing in a different phrasing. The common pattern: a player generates their first report, sees a wall of statistics and per-shot breakdowns, and bounces off because they don't know which numbers matter.

This post is the page we wish those users had had on the way in. It's not specific to AceSense — most of the advice is "what do you do when an AI tennis tool produces a report" — but the workflow assumes you're using AceSense.

What to film

For your first session, don't film a full league match. Film a hitting session, one set of practice tennis, or 30 minutes of structured rally drills. The reason: a full match has too much going on for a calibration session. You want a short, predictable clip so you can compare what you remember of the session to what the report says.

Specifically:

20-30 minutes is the sweet spot. Long enough to give the AI a sample, short enough to upload over normal Wi-Fi without you waiting around.
Singles, not doubles. AceSense's heatmap is calibrated to a singles court layout; doubles works for shot detection but the heatmap is less useful.
Either you on one side, or you focused on one side. If you and a partner are both filmed, the report will analyse both — that's fine, but for your first calibration it's easier to focus on one player's data.
Hard or clay surface. Both work. Clay is slightly noisier (we cover that in filming on clay courts). Indoor works too, with one frame-rate adjustment (the indoor post covers it).

Where to put the camera

This is the bit most first-time users get wrong, and it's the single biggest determinant of whether your first report will be useful or frustrating.

The non-negotiables:

Landscape orientation. Lock it before you press record. Portrait crops out the sidelines and breaks court detection.
Behind one baseline, centred on the court's lengthwise axis. Not on the side fence, not at the net post, not in the corner. Behind the baseline.
6-10 ft (1.8-3 m) above the surface. Above the net cord, looking slightly down at the court. Below net height = bad framing.
6-10 ft behind the baseline. Far enough back to see the whole court in frame; close enough to resolve player pose.

The filming guide has the full breakdown with diagrams. For your first session, "tripod with phone clip, 6 ft tall, weighted at the base, behind the baseline" gets you 95% of the way there. If you don't have a tripod, the no-fence post covers the alternatives.

How to upload

Open the AceSense app. (iOS or Android — works on both. We're not SwingVision; nothing here is locked to one platform.)
Tap "New Session" or "Upload video" — wording varies by app version.
Select the video from your phone's gallery. AceSense will show you the file size and an estimated upload time.
Wi-Fi recommended. A 30-minute 1080p clip is roughly 3-4 GB; on home Wi-Fi, that's 3-7 minutes to upload. On 5G it's faster but uses your data.
Once the upload finishes, you can close the app. The pipeline runs server-side. You'll get a push notification when the report is ready, usually within 5-15 minutes for a 30-minute clip.

There's no "wait, what does this button do?" moment. The upload flow is the upload flow.

How to read your first report

This is the part most first-time users skip and then complain about. The report is several pages long. Don't read the whole thing on session one. Read these three things in this order, and stop.

Page 1 — the court heatmap (read this first)

The heatmap shows where your shots actually bounced, broken down by shot type. It's the single most useful page in the report, and the one your eye should land on first.

What you're looking for, on session one: does the heatmap pattern match what you remember of the session? If you remember spraying backhands wide, do you see a cluster of backhand bounces wide? If you remember hitting deep cross-court forehands, is there a cluster in the deep cross-court zone?

If the heatmap matches your memory, the report is reliable for you. Move on to the next read.

If the heatmap doesn't match your memory — for instance, the report shows backhands going deep when you remember them all going short — the most likely cause is camera position, not AI failure. Check that the phone was above net height in landscape behind the baseline. If it was, then your memory is the unreliable one (which is exactly why video review works).

Page 2 — the shot-mix summary (read this second)

This is where session one earns its keep. Look at the shot counts: how many forehands, backhands, serves, volleys did you hit?

The single most common pattern at NTRP 3.0-4.0 is a 3:1 forehand-to-backhand ratio. Most players genuinely don't know they're running away from their backhand. If your report shows that ratio (or worse), congratulations — you've just identified the highest-leverage single fix in your game. Your opponent figured it out by game three; now you know too.

Don't try to fix it on session one. Just notice it.

One stroke-quality breakdown (read this third)

Pick one shot type — your forehand or your serve — and read the stroke-quality breakdown. AceSense scores stroke quality across preparation, contact, and follow-through, with sub-scores for kinetic-chain components (legs, hips, shoulders, racquet).

For session one, you're not trying to fix anything. You're trying to calibrate your eye to what the scores mean. A 7.2/10 forehand prep score: is that good? Is it bad? On session one, you don't know. You'll know after session three, when you have three reports to compare.

Don't read every shot type's breakdown. Don't compare to anyone else. Pick one shot, read the breakdown, close the report.

What to ignore on session one

The week-over-week comparison panel. You don't have a week-over-week yet. It'll be empty.
The "AI coaching tips" beyond the headline. Read the headline tip; ignore the rest until session three.
The opponent analysis. Useful for match charting, not useful for calibration.
Speed and spin numbers in absolute terms. They're directionally useful (your serve is faster than your second serve) but the absolute number requires comparison data to mean anything. Don't get hung up on whether your serve is "really" 95 mph.
The doubles report (if you accidentally filmed doubles). Singles is the supported path; doubles works partially.

What changes on session two and three

Session two, you record again — same court if possible, similar opponent tier, same camera position. Now the report has a comparison column. Are your shot counts shifting? Is the heatmap creeping deeper? Did your forehand stroke-quality score move?

Session three is where the trend shows up. Three data points is the minimum to call something a pattern instead of noise. By session three, you'll know: is my forehand actually getting more consistent, or am I chasing variance?

This is also when most players start noticing one specific thing the report keeps surfacing — usually a shot-mix imbalance or a placement leak — and start working on it. That's the 4-week self-coaching loop earning its keep.

What if my report has obvious errors?

Three common ones, in order of how often we see them:

Shot count is too low. Almost always a camera position issue. Check landscape, height, framing. Re-shoot.
Heatmap looks rotated or stretched. Court detection failure. Almost always a camera position issue (too low, off-axis, or framing missing the back baseline).
Stroke-quality scores look weirdly low across the board. Check that the camera resolved the player's full body in the frame — if you're cropped at the head or feet, pose detection degrades.

Camera position fixes 80% of "weird first report" issues. The remaining 20% is genuine model error, which we surface on the accuracy page along with the failure modes we know about.

After session three: read the rest of the report

Once the heatmap, shot mix, and stroke-quality breakdown are familiar, then read the rest. The opponent analysis becomes useful when you've played the same opponent twice. The week-over-week panel earns its keep at session three. The full AI coaching tips become readable once you have a baseline.

If you're a coach setting players up with this workflow for the first time, the junior coaches page has the between-lessons version of this loop. If you're a parent of a junior tournament player, the parent guide covers the supportive (not coachy) version.

For everything else, give it three sessions. The pattern shows up.