Restore Old Photos with AI: Scratches, Faded Colors, Better Light

An AI editor can only work with the pixels you give it, so scanning is half the restoration. Use a flatbed scanner at 600 DPI if you have one; otherwise photograph the print in bright indirect daylight, perfectly flat, camera parallel, no flash, no shadows from your hands. Fill the frame and shoot RAW or highest-quality JPEG — every megapixel you capture is detail the AI gets to keep.

Crop to the photo edges, straighten, and save the untouched scan forever before editing anything. Professionals call this the digital negative. You will make mistakes mid-restoration, and the only undo button that truly works is starting over from a pristine original.

Scratches, dust spots, creases and stains are the safest AI edits because they remove information rather than inventing it. Ask for "remove scratches, dust and creases, clean background damage" and the editor heals the photo the way a retoucher would — sampling nearby texture instead of guessing new content. Work from the edges inward: backgrounds first, then clothing, faces last.

Faces are sacred ground. The moment an edit changes who someone looks like, you have failed the restoration no matter how pretty the result. Always include identity anchors in face edits: "keep the same face, same expression, same features, natural skin texture." If a face comes back looking like a different person, discard it and retry with a weaker, more conservative prompt.

Old prints fade toward yellow, red, or washed-out gray depending on the chemistry and the sun. The beginner error is demanding "vivid colors" and getting neon postcards that look nothing like the day. Restrained language wins: "restore natural balanced color, remove yellow age tint, true-to-life skin tones." Think 1960, not 2060.

Fix in this order: white balance first (whites should be white, not cream), then contrast, then saturation last and gently. Black-and-white originals deserve a decision, not an accident — a faithful grayscale restoration beats a garish colorization for anything solemn, while cheerful colorization suits casual snapshots. Either way, keep one faithful copy before experimenting.

Age softens everything: eyes lose their catchlights, fabric loses weave, backgrounds melt into haze. Targeted sharpening — "sharpen facial features, clear eyes, natural detail" — recovers surprising presence, but global over-sharpening creates crunchy halos around every edge. Sharpen people, leave skies and walls alone.

Relighting is the quiet superpower: "soft even daylight, lift deep shadows, keep original mood" rescues photos shot under harsh noon sun or dim bulbs. Do not change the time of day or invent dramatic sunsets unless the family asks for art — restoration serves memory, and memory had its own lighting.

Over-restoration is the epidemic of free AI tools: plastic skin, instagram teeth on grandparents, backgrounds replaced wholesale. If a stranger could tell the photo was "AI-enhanced," you went two steps too far. The test that matters: show it to the oldest relative available. If they say "yes, that is exactly how I remember it," you are done.

Hard limits to respect: faces smaller than a postage stamp cannot be faithfully rebuilt (AI will invent a plausible stranger — label such results clearly); text on signs and documents regenerates as gibberish (transcribe important text separately instead); and fully missing halves of photos are reconstruction, not restoration — beautiful, but honest labeling matters for family history.

Restoration projects sprawl: originals, attempts, finals, colorized variants across a dozen folders with names like "final_v2_REAL." Future you — or your children — will curse present you. Adopt a boring system now: one folder per photo, containing scan-original, working attempts, and a single file named FINAL, plus a one-line text note of what was done ("removed crease, rebalanced color, kept grayscale").

Back everything up twice, in two places, with one copy offsite or in cloud storage. House fires, dead hard drives and lost phones have destroyed more family history than time itself. A restoration nobody can find is a restoration that never happened — and prints keep fading while you procrastinate the backup.