When you upload a video of your serve to RacketFit, we run it through an analysis pipeline that extracts concrete data from your movement. Here we explain exactly what happens, step by step.
The problem with traditional video analysis
Asking a friend to watch your serve and tell you what you're doing wrong has two limits: it's subjective, and it misses the details. The serve is the fastest shot in tennis: in less than a second, the arm goes from the trophy position to impact. The human eye can't capture everything.
That's why we use artificial intelligence: we give the expert eye precise data to work with.
The pipeline, step by step
Frame extraction
Your video is scanned and 26 frames are extracted, evenly distributed over time. We look for the moments where you are clearly visible and in action.
Player detection
A pose estimation algorithm (MediaPipe) analyzes each frame and finds where you are in the shot. It measures how visible you are (full body or partial), how large you are in the frame, and whether you are moving.
Player crop
The best frames are cropped around you, with a 3:4 aspect ratio. For the serve, the crop is taller to capture the trophy position and impact. If detection is weak (fewer than 4 good crops), the report omits the image: better no frame than a wrong one.
AI analysis
The crops are passed to a multimodal model (Claude) that describes visible elements such as apparent grip, motion path, apparent contact point and movement patterns. Spin, speed and forces are not measured. The model must say "not determinable" when the footage is insufficient.
Aggregation and report
Observations from the four shots are combined with the player's stated information. The AI produces an informational report with explicit limitations, racket, string and tension options and a test protocol. It does not make technical or medical diagnoses.
What video cannot establish
The pipeline does not measure shock, spin, speed, dwell time or shoulder load and cannot link a setup to pain. If a player reports discomfort, the report treats it as self-reported information and recommends consulting a doctor or physiotherapist. Equipment options remain hypotheses to discuss with a qualified technician.
The key lesson: a wide angle from behind works for forehand, backhand and volley, but not for the serve. The serve needs a three-quarter side angle, closer in (3-4 metres), so the arm and racket stay in frame throughout the whole motion.
Why this approach works
Frame selection is critical: wrong frames = wrong profile. That's why we use a "many frames + aggregation" approach: never rely on a single key frame. The AI is honest: if the footage is not enough, it says so.
The result is a report that explains why your current setup doesn't work for your game style, and gives you a concrete protocol to test the solution on court.
Want your report?
Upload videos of your game and get the full analysis: observable technical profile, setup review and 2-3 racket, string and tension options to test.
Upload your videosFrequently asked questions
How should I film my serve for AI analysis?
A three-quarter side angle, closer in (behind and to the side, 3-4 metres from you), so your arm and racket stay in frame throughout the whole motion. 4-6 serves are enough.
How many frames does the AI analyze for each shot?
We extract 26 evenly spaced frames from the video, then select the ones where the player is clearly visible and in action. The AI analyzes the best ones to build the technical profile.
What happens if the video is not good enough?
The AI is honest: if the footage is not enough for a reliable assessment, it says "not determinable". Better no frame than a wrong one.