CommunityCoding & Developmentgithub.com

Portrait Retouch

Refine existing portrait photos using image-to-image techniques. Suitable for removing acne, reducing shine, lightening dark circles, improving backlit faces, and achieving natural beauty enhancements that make you look refreshed yet like yourself. Supports targeted adjustments by area and intensity, delivering standalone retouched images and before-after comparisons with identical composition.

What is Portrait Retouch?

This skill, named 'Portrait Retouch,' uses image-to-image generation to retouch existing portrait photos with natural and subtle enhancements. It is designed for common needs like removing acne, reducing oil shine, lightening dark circles, fixing backlit selfies, and achieving a 'refreshed but still me' look without heavy makeup or morphological changes. The skill excels at interpreting everyday language (e.g., 'I look tired' or 'fix these pimples') into precise editing actions based on the actual photo content. Users can specify areas, intensity, and what to preserve (e.g., moles, face shape, expression). It supports common developer platforms like Claude Code, Cursor, or Codex where image editing tools are integrated. The workflow involves analyzing the original photo, applying targeted edits (e.g., shadow correction for tired eyes, matte finish for shiny forehead), generating a before-after comparison with matched composition and scale, and delivering the standalone retouched image. The skill prioritizes natural results, preserves facial geometry, skin tone, age features, and expression unless explicitly overridden by the user.

Works with~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/hzw-object/portrait-retouch/tree/main

Installed? Explore more Coding & Development skills: steipete/bluebubbles, steipete/eightctl, steipete/blucli · View all 6 →

Ask in your favorite AI

Open a new chat with this agent skill pre-loaded.

Documentation

Natural Portrait Retouch

Users provide a photo and a casual description, and retouching can begin. By default, natural light retouching is applied; user-specified makeup style, intensity, and preservation items take priority.

Translating Everyday Language to the Photo

First examine the original photo, then select adjustments that actually exist in the image and relate to the request.

User SaysObservation and Processing
"Fix these pimples for me"Repair temporary red pimples based on location, referencing the texture and brightness of adjacent skin. Preserve moles and freckles.
"My eyes look a bit tired"Find the darkest area below the lower eyelid, gently lift the shadows, while preserving the boundary of the under-eye bags and fine lines from smiling.
"My forehead is too shiny"Reduce harsh highlights, preserving the light-to-dark transition on the nose bridge and forehead so the light direction remains readable.
"Backlit selfie, help me look more vibrant"Prioritize adjusting facial exposure and local color cast, using the ears and neck as skin tone reference, and preserve the original background brightness.

Execution and Delivery

  1. Use the user's original photo as input to the image editing tool. When there are multiple images, distinguish between the photo to retouch and the makeup reference image. If retouching only one person in a group photo and the description is unclear, ask for their position. If no original photo is provided, ask the user to upload one; only use sourced example portraits if the user wants a demonstration.
  2. Write specific areas, degree, and preservation items for the editing tool. By default, preserve face shape, distance between facial features, skin tone, age characteristics, and expression. For example, "look refreshed" can usually be achieved by correcting shadows around the eyes; slimming face, enlarging eyes, and changing makeup must be explicitly specified by the user.
  3. After generation, compare face shape, gaze, mouth corners, and picture edges at the same display size. Then zoom in to check pores, hairlines, and color differences in edited areas. If noticeable drift occurs, go back to the original photo and redo with a smaller editing scope.
  4. Save the retouched image separately and keep the original. When a comparison is needed, use the actual input and final output side by side in equal proportions, label before and after, and keep the full frame; also deliver the standalone image. When highlighting local differences, add one or two detail crops below the full-face comparison. Both before and after should use the same coordinates, magnification, and display size. Describe visible differences. If the user only wants a prompt, provide an editable editing command directly.

Image-to-Image Example · Light Makeup Retouch

The example original is an AI-generated fictional portrait. The retouched result is edited from that original image. See the side-by-side comparison.

Instructions that can be used directly with the original photo

Perform a visibly clear, avatar-suitable transparent light makeup retouching on this original photo to generate a single image. Completely remove the red pimples and surrounding red marks on the right cheek and chin, naturally filling them with fine texture from adjacent skin. Tone down the large shiny forehead oil to a soft matte finish. Significantly brighten the blue-brown dark area under the eyes, evening out skin tone. Lift the lips to a clear, natural rose-beige color, matching the original lip shape and retaining lip texture. Preserve the small mole under the eye, pores, facial bone structure, eye shape, original skin tone, and smile. The composition and facial features must be precisely aligned with the original photo, preserving the curtains, wall, hair, earrings, collar, and window light direction.

Related Skills