The shoebox of prints every family owns can be repaired, colorized and even gently animated with nothing but a phone. Here's the complete workflow, including the mistakes that ruin restorations before the AI ever sees them.
Restoration quality is decided here, not in the AI. Lay the photo flat near a window in indirect daylight, hold the phone parallel so the print fills the frame, and never use flash — glare burns out highlights no model can recover. If the print is curled, hold the corners down with two coins kept outside the crop.
Before any AI pass, use the built-in photo editor: crop the border, straighten the horizon, and tap out obvious dust with the healing tool. Clean input means the AI spends its effort on real damage — scratches, tears, fading — instead of on your thumb shadow.
AI restoration models have learned from millions of photographs what faces, fabric and film grain should look like, so they can rebuild scratched or missing areas and sharpen soft focus convincingly. Expect excellent results on fading, scratches and small tears; damage covering more than about a third of a face is where restoration becomes invention — good tools warn you instead of guessing.
Colorization adds natural tones to skin, clothing and sky. Two tips: reference reality where you can (grandmother's cheongsam was red — tell the tool if it lets you), and be skeptical of over-saturated results. The goal is "how it looked", not "Instagram 1962".
A breath, a blink, the hint of a smile — subtle motion is what makes people cry at family gatherings, in the good way. Keep it minimal for portraits of relatives who have passed, keep the untouched original alongside, and ask close family before sharing animated versions to the group chat.
Export at full resolution and back up in two places (cloud plus a family shared album works). And the nicest ending we know: set the restored memory as a live wallpaper, so the photo that lived in a shoebox now lives on the screen you see a hundred times a day.
Memory Revive is our lab experiment that runs this whole workflow — repair, colorize, animate, living wallpaper — from a single photo of the print. Joining the early list is the vote that gets it built, and early members restore free at launch.
Or read the full pitch: Memory Revive experiment page
Yes, for most damage — scratches, tears, fading and blur respond well. Missing areas larger than about a third of a face are the honest limit: past that, results are plausible invention rather than restoration.
A flatbed scan at 600 DPI is ideal; a careful phone photo in indirect daylight is more than good enough and is how most family shoeboxes actually get digitized.
Families differ. Subtle motion lands very differently from talking-head effects — keep it gentle, keep the original, and ask before sharing.