How it works
Soft Glam does not regenerate, reshape, inpaint, relight, or generally blur 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 to distinguish face skin from hair, beard, body, clothing, accessories, and background. The renderer combines exactly three broad skin-polish zones across the forehead and both cheeks with exactly five compact makeup zones above both upper eyelids, over both upper cheeks, and inside the outer lip contour. The skin polish uses a skin-mask-normalized local baseline so nearby background, hair, or beard color cannot bleed into the correction; it reduces only measured mid-scale luminance and chroma variation while retaining fine luminance detail and pores. At full intensity, skin luminance correction is capped at two LAB levels and each skin chroma correction at three LAB levels. Blush and lip tint change LAB chroma while retaining source luminance, and eyeshadow darkening is capped at four LAB levels. Face-mesh masks protect the eyes, irises, eyebrows, nose, and inner mouth. Facial geometry, expression, protected features, hair, beard, clothing, scene lighting, background, framing, uploaded alpha, and every pixel outside the eight reported zones are copied unchanged into a lossless PNG. Profiles, incomplete or tiny faces, uncertain or incomplete face-skin evidence, and glasses or other obstructions that leave any required zone without enough safe support are refused instead of redrawn. Already-even skin may receive only the verified makeup preset; a portrait with no safe visible correction is left unchanged and refused.
