Pitch Lab

Upload video

Shoot from behind the pitcher. Slow-motion (iPhone 120/240fps, Sony HFR) tracks much better than normal speed.

Seed the tracker (3 points)

0 / 3 seed points

Scrub to three moments of the flight and mark the ball at each: the start (just thrown), the middle (around the peak), and the end (where it's caught). At each, click Add Seed Point, then click the ball. These three set the start/end frames automatically and let the tracker follow the ball forward and backward — so the descent stays locked even as the ball changes size.

Detect ball trail

no trail yet

If a frame is mis-located, switch to Manual Correct and drag a box around the ball at that frame — the box sets both its position and size, so distant/blurry balls stamp at the right scale. A quick click drops a point at the last size used. Switch to − Remove and every trail point shows as a red dot — click the one you want gone (it shows an ✕ when you hover it). After fixing points, hit ↻ Refresh to re-capture the ball patches at your corrected sizes.

Trail style

Stamps: 25

Stamps are drawn translucent so the live ball is still visible during playback.

Save pitch

Distance: ft Capture FPS: fps

iPhone slo-mo = 120 or 240 · Sony HFR = 250/500/1000

How this prototype works

Upload your videoiPhone slow-motion clips work best — the ball stays visible for many frames. Normal 30fps clips will work but you'll need manual correction more often.

Find the release frameScrub forward (Right arrow steps one frame) until the ball has just left the pitcher's hand. Click Mark Release Frame.

Find the end frameScrub forward until the ball reaches the target (or leaves the frame). Click Mark End Frame.

Auto-detectThe tool walks every frame between your two markers, compares each to the release frame, and finds the bright spot that wasn't there before — the ball. It records the ball's position on each frame.

Manual correct (if needed)Switch to manual mode, scrub to any frame where the trail looks wrong, and click on the ball. That position overrides the auto-detect for that frame.

Save the pitchName it, tag the pitch type, save. It shows up in the Pitch Library tab.

What this prototype is NOT yet

• Cloud sync is on — saved pitches (and their videos) now sync to your Supabase project, so you can open them from your phone and laptop. Videos download on demand the first time you open a pitch on a new device. Login/multi-user comes in the next phase.
• No multi-pitch overlay — once the tracking is verified, Phase 2 adds the ability to superimpose multiple saved pitches on top of each other to compare release-point consistency.
• No velocity / pitch ID — Phase 3 adds rubber-to-target distance calibration so the tool can estimate pitch speed and break.
• No Claude coaching — Phase 4 hooks up Claude to give feedback on arm slot, target accuracy, and release consistency.

Why this approach

Ball tracking from behind-the-pitcher video is normally hard because the ball moves fast and gets small. Three things make it tractable here:

Slow-motion footage gives 40–80 frames of ball flight per pitch instead of 5–10
A high-contrast white ball against dark backgrounds lets simple frame-differencing find it
The phone is static, so the background is stable — anything that's bright and moving must be the ball

The auto-detect doesn't use any machine learning model. It's just: "diff this frame against the release frame, find the brightest moving blob in the area where the ball is expected to be."