Bring Old Family Photos Back to Life With AI (Restore + Animate)

July 31, 2026

There’s a particular feeling when a scratched, faded photo of your grandparents comes back sharp and full of color — and then, a minute later, they blink and smile. It’s the most meaningful thing we’ve done with these tools, and it’s a straightforward two-step process: restore the photo in Gemini, then animate it gently with an image-to-video model.

Here’s the exact workflow, with the prompts we use.

Before You Start: Scan It Properly

The restoration is only as good as your input. Don’t photograph the print at an angle under a lamp — lay it flat in indirect daylight and shoot straight down, filling the frame, or use a flatbed scanner if you have one. Avoid glare on glossy prints by tilting slightly until reflections disappear.

One more thing: these are usually photos of relatives. If they’re living, ask before you edit and share. If they’ve passed, use your judgment about what they and the family would be comfortable with — restoration and a subtle smile is a lovely tribute; putting words in their mouth is a different thing entirely.

Step 1: Restore in Gemini

Upload the scan to Gemini and use a prompt like this:

Restore this old damaged photograph. Repair scratches, tears, dust spots,
and creases. Fix the fading and restore natural contrast. Add realistic,
historically appropriate color if the photo is black and white. Keep the
people's exact facial features, expressions, clothing, and the original
composition unchanged. Do not modernize anything — preserve the authentic
look of the era, just clean and clear.

The “do not modernize” clause is important. Without it, models sometimes sharpen a 1950s photo into something that looks shot on a phone, which ruins the point.

Check the faces carefully. Restoration involves reconstructing damaged areas, and if damage crosses a face, the model is partly guessing. Compare against the original — does the nose still look right? The eyes? If something feels off, regenerate, or restore in two passes: damage repair first, colorization second. Smaller steps mean fewer guesses per step.

Keep your originals. Save the untouched scan and every intermediate version. Restoration is interpretation, and you may want to redo it as models improve.

Step 2: Animate — Gently

Take your restored image to an image-to-video tool. Kling and Google Veo both handle this well, and both typically offer free starting credits. The single most important rule: subtle wins. A blink and a soft smile is magic; a wave and a walk-off is uncanny.

Animate this photo with very subtle, lifelike motion: the person blinks
naturally, breathes gently, and forms a soft warm smile. Minimal head
movement. Keep their exact facial features and the background completely
unchanged. Slow, calm, realistic movement. 5 seconds.

Generate the short clip first. If the motion is too strong, add “extremely subtle motion, almost still.” If a hand or hair does something strange, just regenerate — motion glitches are usually bad luck, not bad prompting.

For group photos, animate the whole frame with “each person blinks and smiles softly” but expect more retries; more faces means more chances for one of them to drift.

Sharing the Result

When you share these with family, say what you did — “restored and animated with AI” — both for honesty and because, in our experience, relatives find it more moving when they understand it. A nice format: original photo, restored photo, then the video, as a three-part post or a short slideshow.

If you want to go further, our prompt library has restoration variants for specific damage types (water stains, mold spots, torn corners), and the prompt generator can build a restoration prompt matched to your photo’s era and condition.

What This Workflow Can’t Do

Honesty corner: if a face is mostly destroyed or the photo is a thumbnail-sized blur, the AI will invent a plausible face rather than recover the real one. The result may look great and still not be your great-grandfather. For badly damaged photos of people you have other pictures of, some tools accept a reference image — otherwise, treat heavy reconstructions as an artist’s impression, not a recovery.

FAQ

Is colorization accurate? It’s plausible, not factual. The model infers likely colors from context — skin, foliage, common fabric colors — but that blue dress might really have been green. If you know the true colors, state them in the prompt: “her dress was pale yellow.”

Can I restore a photo of a large group? Yes, but zoom in and check every face afterward. For precious photos, crop and restore in sections, then evaluate each. Some faces may need their own regeneration pass.

How long should the animation be? Five seconds is the sweet spot. Identity drift in video grows with duration — the longer the clip, the more the face wanders from the photo. Short, subtle, looping clips look best and cost the fewest credits.