How it works
Grooming does not generate, remove, extend, reshape, recolor, sharpen, blur, or inpaint hair or the portrait. It requires one complete near-front face, cross-checks dedicated face detection with MediaPipe Face Mesh V2, and uses MediaPipe six-class portrait segmentation only as a face and non-accessory safety envelope—not as proof that a dark pixel is facial hair. The renderer measures a source-skin luminance reference inside the verified forehead zone, then retains only source pixels sufficiently darker than that reference inside exactly two face-mesh eyebrow polygons. It can also define already-present facial hair inside one optional lower-face zone: the candidate must stay inside the face oval, below the nose, within verified face semantics, outside protected nose and lip masks, and in a sufficiently large connected region whose dark pixels have nearby original fine-detail evidence. Smooth cast shadows cannot qualify. Very light, absent, or uncertain facial hair is left unchanged. The adjustment changes LAB luminance only, capped at three levels for brows and four levels for facial hair at full intensity; it adds no synthetic strands, texture, edge sharpening, or color. Eyes, irises, nose, lips, inner mouth, face skin outside the accepted source-dark detail masks, ears, scalp hair, turban or other headwear, facial geometry, expression, lighting, clothing, background, framing, uploaded alpha, and every pixel outside the two required brow zones and optional facial-hair zone are copied unchanged into a lossless PNG. Profiles, incomplete or tiny faces, uncertain face-skin evidence, obstructions, insufficient source-dark brow evidence, and uncertain facial-hair evidence are refused or left unchanged rather than redrawn.
