How it works
Leg Length does not send the portrait through a full-frame generator. It requires one clear head-to-toe person and cross-checks mirrored MoveNet Thunder shoulder, hip, knee, and ankle landmarks against separate face detection. Six-class portrait segmentation must then trace two continuous, separated silhouettes from mid-thigh through both shoes, with verified background beside each leg and enough clean floor below each sole. Two independent vertical source-pixel meshes begin about 40% down each hip-to-ankle chain and lengthen those bands by up to a restrained 7% at maximum intensity. Each shoe is translated downward without being intentionally scaled, while nearby source background is compressed only inside the measured side-and-floor envelope and displacement fades to zero at every boundary. The face, hair, arms, torso, hips, upper thighs, distant background, and every pixel outside both envelopes are copied unchanged into a lossless PNG. Cropped, bent, touching, occluded, transparent, or ambiguous legs are refused instead of guessed.

























