Motion Lab FAQ

Every other part of SwingVantage feeds into or out of it: you capture evidence for Motion Lab, practise what it found, and come back to it to retest.

The tracking runs on your own device, and every result says how it was obtained — estimated from the video or inferred from a pattern — and how confident it is.

If you arrive from a sport page or a link that names a sport and motion, Motion Lab opens with those already selected, so you can go straight to recording or uploading.

You can read what it does first on the Motion Lab overview.

• Golf: Driver Swing, Fairway Wood Swing, Hybrid Swing, Iron Swing, Wedge Swing, Pitch / Chip and Putting Stroke
• Tennis: Forehand, Backhand, Serve, Volley and Return
• Pickleball: Dink, Third-Shot Drop, Drive, Serve, Volley and Reset
• Padel: Bandeja, Víbora, Smash, Volley, Groundstroke / Wall, Lob and Serve
• Baseball: Hitting Swing, Throwing Motion, Pitching Motion, Fielding Transfer and Catcher Throw-down
• Slow-Pitch Softball: Hitting Swing, Throwing Motion and Fielding Transfer
• Fast-Pitch Softball: Hitting Swing, Slap Hit, Pitching Motion, Throwing Motion and Fielding Transfer

If your exact shot is not listed, pick the closest motion. A read built for a different motion is less reliable than one built for the motion you actually filmed.

Uploaded clips can be MP4, MOV and WebM files. Guided recording (RecordAssist) is available once you are signed in and hands the clip straight to Motion Lab without it leaving your device.

Whichever you choose, the same quality check runs on the footage, and it will tell you if the clip limits what can be read.

It then names the specific problems it found (for example "the feet and lower body were not consistently in frame" or "the camera moved during the clip"), lowers confidence in the reads those problems affect, and gives you recording fixes for next time.

This is why a weak clip still gets an honest answer instead of a blank screen or a confident-looking number it cannot back up.

• Excellent — supports a full, high-confidence analysis.
• Good — the core motion reads are well supported.
• Usable — the main reads hold; a few are lower-confidence.
• Limited — partial reads, with the fixes flagged.
• Difficult capture — only the most defensible signals were recovered.
• Not defensible from this angle — the body could not be tracked reliably; any reads are heavily capped, and none are shown if no body was found at all.

Only Excellent and Good captures are marked retest-ready, because a before-and-after comparison is only fair when both clips can support it.

The results come as a guided walk-through: a Summary, Visual Proof (the clip with the tracking drawn over it), the Main Issue, and — with a free account — your One Fix, practice plan and retest.

Video reads are estimates from a single camera. If you import golf launch-monitor data, those numbers stay separate, labelled as measured by that device — they are not blended into the video read. See what SwingVantage does not measure.

For golf, automatic clubhead tracking is not reliable enough yet, so the trace follows your hands, which the body tracking does see. To trace the clubhead itself, choose Mark clubhead on the video and tap the head on a few clear frames — your marks become the trace. A trace shows positions in the image; it does not establish a target-relative club path or a verified moment of contact.

For golf full swings you can also step through key swing positions. Positions that depend on seeing the shaft clearly are shown as unavailable when the clip cannot support them.

When a clip allows it, you may be offered a two-angle capture for a stronger 3D read. See two-camera 3D.

Swings it cannot score are left out of the summary rather than counted as zero, and the summary says how many swings contributed.

For a single focused read, trimming the clip to one rep still gives the cleanest result; several reps are useful when you want to see what repeats.

Two views improve the shape of the 3D motion, but without a verified calibration the overall scale is still estimated — so the result is labelled as triangulated with scale estimated, not as a measurement.

Both clips have to capture the same moments, so the two cameras should record the same rep at the same time.

The label on the 3D view tells you exactly what it is: estimated from a single camera, triangulated from two views with the scale estimated, or — only with a verified calibration — measured.

Downloads are not offered when the clip needs to be re-filmed, because a report built on a capture that cannot support a read would be misleading. Every printed report carries the same confidence labels and disclaimers you see on screen.

What is saved is the analysis, not the video file: the reads, a compact motion track and, for some swings, one still frame used to draw your swing trace. Older analyses drop off once the saved list is full. An analysis you ran before creating an account resumes where you left off when you sign in on the same device.

When you are signed in, saved analyses follow your account: the reads, a compact motion track and, for some swings, one still frame used to draw your swing trace — never the video file. The trade-off of on-device analysis is that a very old or low-powered device may take longer, and the browser tab needs to stay open while it runs.

Two optional features can send a limited set of still frames — never the video file — and only if you choose them: a cloud re-analysis offered when the on-device read came out too poor to use, and an AI second opinion you can start after your results. See what leaves your device.