How Omuscle rates you
Everything runs locally in your browser. No video is ever sent anywhere. The AI is a pose estimation model — it sees your skeleton, not a photo.
Your webcam feeds into MediaPipe Pose
MediaPipe Pose is a Google model that runs entirely as WebAssembly in your browser — it never touches a server. Every frame of your webcam feed is analyzed to extract 33 body landmarks: your shoulders, elbows, wrists, hips, and a few head anchors for posture.
33 landmarks. 8 that matter for scoring.
We only score based on upper body landmarks: both shoulders (11, 12), both elbows (13, 14), both wrists (15, 16), and both hips (23, 24). The small faint dots near your head — nose and ears — are used purely as a posture anchor to check shoulder level and spine alignment. They have zero effect on your physique score.
Your hips must be in frame for a full scan
Three of the seven scoring factors — V-Taper, Shoulder Width, and Chest — require both hip landmarks to be detected at high confidence (≥65%). If your hips aren't visible, those factors are excluded from your score entirely and a yellow banner prompts you to step back. A scan without hips is marked as partial.
The golden ideal is overlaid on your body
Based on your detected torso centre and height, we compute where the ideal 10/10 physique's landmarks would sit — using the golden ratio (1.618) for shoulder-to-hip, and classical bodybuilding proportions for arm development. These appear as gold dashed dots and lines on screen. The gap between your green detected dots and the gold ideal dots is what drives your score.
7 factors are scored with anti-gaming limits
Each factor compares your detected ratio against a research-calibrated ideal. Some factors have hard ceilings: Arm Development is capped at 7.5/10 because elbow-spread alone can be gamed by arm position or lighting — the remaining headroom is reserved for vision-model muscle definition scoring. V-Taper is capped at 9.5/10, and frames where the shoulder:hip ratio exceeds 2.0 (physically impossible) are rejected entirely.
Score is smoothed over 20 frames, then you see the result
Raw per-frame scores fluctuate as the model tracks you. We average over a 20-frame rolling window so your displayed score is stable. You see three things: your overall 0–10 score, your DOM (the factor you score highest on), and your FLAW (the factor dragging you down most). Individual sub-scores are intentionally hidden.
Weights are hidden in the UI. Shown here for transparency.
Shoulder width divided by hip width in pixel space. Linear scale: even (1.0) → 0, golden ratio (1.618) → 9.5. Frames where the ratio exceeds 2.0 are rejected as detection noise. Hip landmarks must hit 65% visibility confidence — lower than that and V-Taper is excluded for that frame.
Your shoulder span in pixels divided by your torso height (shoulder midpoint to hip midpoint). Scale: 0.85× → 0, 1.45× → 10. Research baseline (Hughes & Gallup 2003): average male = ~1.00–1.10×, elite aesthetic = 1.35–1.45×.
Similar geometry to Shoulder Width but scored on a slightly more lenient scale (0.80× → 0, 1.40× → 10). Captures how imposing your upper body looks relative to your overall frame.
Elbow-to-elbow width relative to shoulder width. Bigger arms push elbows further out. Scale: 0.72× → 0, 1.15× → 7.5 max. If elbow spread exceeds 1.25× shoulder width, it's flagged as likely gaming (arms held wide) or lighting artifact and capped at 6.5. The remaining 2.5 points are reserved for when vision-model muscle definition is added.
Compares total arm length (shoulder→elbow + elbow→wrist) left vs right, plus shoulder height difference. Both contribute to a combined deviation score. Based on bilateral proportion research. Falls back to shoulder-level only when elbows or wrists aren't detected.
Ratio of upper arm length (shoulder→elbow) to forearm length (elbow→wrist). Bell-curve scored around 1.18 — the McCallum / Steve Reeves classical ideal. Sensitivity: 30 points per unit of deviation. Excluded when elbows or wrists aren't visible.
Shoulder tilt angle (height difference left vs right, relative to shoulder width) plus how well your nose aligns over the centre of your shoulders. Tilting, leaning, or standing off-axis all reduce this score.
A 0–10 composite from all available factors. Partial if hips aren't detected.
Your highest-scoring factor — your physical standout. Green.
Your lowest-scoring factor — what's pulling your score down. Red.
- →Show your full upper body — waist to neck — so hip landmarks are detected and V-Taper can be scored
- →Wear fitted or no shirt — loose fabric hides shoulder width and obscures elbow position
- →Stand 4–6 ft from the camera — too close cuts off your hips, too far reduces landmark precision
- →Keep arms slightly away from your body — arms pinned to your sides compress the elbow spread reading
- →Use front lighting — shadows shift where the model detects your elbows and can inflate or deflate Arm Dev
- →Stand square to the camera — side angles distort every width measurement simultaneously