Loading pose estimation model
MoveNet Thunder · runs entirely in your browser
Step 1 -- Select athlete sex for grading ranges
Male
Uses male reference ranges
Female
Uses female reference ranges
Selecting a sex loads the relevant reference dataset for angle grading.
Step 2 -- Upload videos (all three required)
Side view

Film from directly beside the runner

Best: Someone moving alongside you at the same pace.
Good: Someone panning the camera to follow you.
OK: Stationary camera, you run past it.
Wear clothes that contrast with the background.
REQUIRED
Front view

Film from directly in front

Run straight toward the camera. Do not run at an angle -- even a slight angle will skew hip drop and knee alignment measurements. Wear contrasting clothes against the background.
REQUIRED
Back view

Film from directly behind

Run straight away from the camera. Keep the camera dead center behind you, not at an angle. Wear contrasting clothes against the background.
REQUIRED
Recording tips for the best results
Keep clips short: 3 to 8 seconds. We only sample 20-40 frames from each video, so longer clips just add unnecessary footage. A few seconds of steady running is ideal.
📷Stay close to the camera. The runner should fill most of the frame. If you are too far away, the pose model cannot accurately identify joint positions.
Nothing should block your body. Treadmill handrails, other runners, or objects in front of you will hide joints and reduce detection accuracy. If filming on a treadmill, position the camera so the display and handrails are not covering your arms or legs.
🏃Consider filming both outdoors and on a treadmill. Treadmill running can change your form -- your stride length, foot strike, and trunk lean may differ from outdoor running. If you want a complete picture, analyze one of each and compare the results.
🎨Use a uniform background that contrasts with your clothing. The pose model works by identifying your body against the background. If your clothes blend into the background (e.g., dark clothes on a dark track, or grey clothes against concrete), detection will suffer. A turf field, a plain wall, or any backdrop that is a clearly different color from what you are wearing will give much better results.
Avoid very bright or very dark lighting. Overexposed (washed out) or underexposed (too dark) video makes it harder for the model to distinguish your body from the background. Even, natural lighting works best. Avoid filming directly into the sun or in deep shadow.
All video processing happens locally in your browser. Nothing is uploaded or stored anywhere.

MP4 / MOV / WebM -- Fully private -- No data uploaded

Scanning stride phases
Analyzing your videos for key moments of the running gait cycle.
Preparing...
Preparing...
Before you finish: review every card below
We made a first pass at detecting each phase of your running gait. Now you need to open each card, check the frame, and make sure detection is strong. This is what makes your results accurate.
1
Click on every card below
Each card represents a specific moment in your stride. Click on each one to open it. You will see a skeleton overlay on your video and a detection quality reading.
2
Get every card to "Strong detection"
The large detection quality indicator at the top of each card should read "Strong detection" in green. If it does not, click the Analyze button again, or use the scrubber to pick a clearer frame.
3
Make sure each frame matches its description
Each card describes what the frame should look like (for example, "the moment the left foot contacts the ground"). If the auto-selected frame does not match, use the scrubber to find the right moment and click Analyze again.
4
Then click "Complete analysis"
Once every card shows Strong detection and the correct frame, scroll to the bottom and click Complete analysis to generate your full report with coaching recommendations.
Why does this matter? The accuracy of your final report depends entirely on the quality of these frames. A frame with weak detection or the wrong moment selected will produce inaccurate angles and unreliable coaching cues. Taking a minute to check each card is the difference between useful feedback and noise.

If you are consistently getting weak detection across multiple cards, the issue is usually the video itself -- the runner may be too far from the camera, partially blocked by equipment, the clip may have too much motion blur, or the clothing may blend into the background. Wearing clothes that contrast with the background (e.g., a bright shirt on a green turf field) and filming in even lighting will make the biggest difference. Re-recording a short, close, well-lit clip with good contrast will help more than adjusting frames.
Left side
Right side
Trunk lean
▸ Within range
▴ Above range
▾ Below range
All phases analyzed -- ready to complete
Review the report below to finalize your analysis
1
Phase
⚠ Not auto-detected
⚠ Not automatically detected
Use the scrubber to find the best matching frame, then click ↻ Analyze.
Frame adjustment
0.00s
frame
jump to:
Overlay key Left side Right side Trunk lean Center
What to look for
Estimated angles for this frame compared to elite reference ranges. The bell curve shows where your value falls relative to the reference population. Accuracy depends on video quality, camera angle, and keypoint confidence.
Analyze a frame to see angles
My Sessions
Sign in to see your saved sessions.
Compare analyses