How it works
Skin Texture Preserve is a source-detail treatment, not generated skin texture or a whole-face sharpening filter. It requires exactly one complete near-front face. A dedicated face detector verifies that only one face is present, MediaPipe Face Mesh V2 verifies its full geometry, and MediaPipe six-class portrait segmentation separates face skin from hair, beard, body, clothing, accessories, and background. Exactly three bounded zones cover verified skin on the forehead and both cheeks; dilated face-mesh masks protect the eyes, irises, eyebrows, nose, and lips. A skin-mask-normalized fine-scale luminance baseline prevents background, hair, or beard pixels from entering the texture measurement. The renderer reinforces only fine LAB luminance detail already present in the upload, applies no chroma correction, and caps luminance correction at 1.5 LAB levels. It never synthesizes pores, strands, wrinkles, or any new texture. Broad skin tone, eyes, irises, eyebrows, nose, lips, inner mouth, hair, beard, facial geometry, expression, body, clothing, lighting, background, framing, uploaded alpha, and every pixel outside the three zones are copied unchanged into a lossless PNG. It never smooths, denoises, blurs, or sharpens the whole face, recolors, reshapes, inpaints, or regenerates the portrait. Multiple faces, profiles, incomplete or tiny faces, uncertain skin evidence, obstructions, and skin without enough reliable fine source detail are refused rather than guessed.

























