AI Old Photo Restoration

What It Solves

Old photos often have fading, yellowing, scratches, tears, noise, and weak facial detail. Manual retouching requires separating damage from original character, while simple filters can over-smooth faces or make the image look artificial. This skill turns a common restoration workflow into a scriptable process: analyze the photo, choose color, face, and damage repair strategies, then export comparable versions.

How It Works

The core entry point is scripts/restore_photo.py, which loads an image and applies restoration options. Key steps:
- Photo analysis: detect fading, color casts, scratches, blur, missing areas, and face locations; decide whether colorization or face enhancement is needed.
- Repair strategy: use histogram equalization, white balance, and color correction for tone issues; use OpenCV for face detection and models such as CodeFormer or GFPGAN for facial enhancement; use inpainting and denoising for small scratches and noise.
- Output control: options include --enhance-faces, --color-correct, --colorize, --remove-scratches, and --intensity, with PNG or high-quality JPEG output and before/after versions for review.

Limits To Note

It fits mild-to-moderate damage, especially frontal faces, fading, scratches, and black-and-white colorization. Large tears, large missing areas, or very low-resolution images still need manual editing. Colorization accuracy depends on model training and may not match historical reality, so use conservative intensity settings and keep the original file.