How to animate an old photo

Old photographs are the one case where "turn a photo into a video" is not a novelty. A scanned print from 1962 is usually the only moving-image record that will ever exist of the person in it, and a slow push across their face, in a video with their name and a piece of music, is genuinely moving in a way a scan sitting in a folder is not.

It is also the case where doing it badly is most obvious. Here is the order that works.

Restore first. Always first.

Every step below gets harder once there is motion involved. A scratch on a still is one repair; a scratch on an animated clip is a repair on every frame.

Scan properly, if you still can. Whatever you feed in is the ceiling. A phone photo of a print, taken at an angle under a lamp, carries the glare and the perspective into everything afterwards. Flat, even light, square on.

Upscale it. Old prints are small and soft, and generators work from what you give them — a soft input produces mushy motion. Enlarging first is cheaper than discovering the problem three generations later.

Be clear-eyed about what upscaling does: it adds plausible detail, it does not recover information the negative never captured. On a face, that means it is inventing pore-level texture. Usually convincing, occasionally subtly not the person. Look closely at the eyes and mouth before you accept it.

Repair the damage. Creases, tears, foxing, the missing corner. Image editing handles these well — and unlike upscaling, a repaired crease is genuinely just gone rather than reimagined.

Decide about colour. Colourising a black-and-white photo is a choice, not a restoration: every colour in the result is a guess. Sometimes a lovely one. But if you know she wore a blue dress and the model picks green, it is now a picture of someone else's afternoon. Keep the monochrome version too.

Then choose the motion

Two genuinely different options, and the quieter one is right more often than people expect.

A camera move

The photo does not change. The frame moves across it — a slow push into a face, a drift along a group, a pull back to reveal the room.

Use it when the photograph is the point. Nothing is invented, nobody's face shifts, and it costs nothing to render. Ken Burns built a career on this and nobody watching a documentary thinks the stills are a limitation.

For a group photo, a slow drift across the faces is far stronger than a zoom into the middle — it mimics how someone actually looks at an old picture.

Keep the move slow, and give it a reason. Push in to end on a face. Pull back to end on the room. A zoom into nothing in particular just reads as an accident.

Generated motion

The subject actually moves — a blink, a head turn, a smile.

This is image-to-video with the restored photo as the start frame, and it is the impressive one. It is also where this goes wrong, because a face invented by a model is not that person's face. Two rules that help a great deal:

Ask for very little. "Subtle movement, camera holds still" gets you a photograph that breathes. Anything more ambitious and you are generating a stranger.

Keep it short. Two seconds. Generation degrades the longer it runs, and it degrades on faces first.

Where to stop

Worth saying plainly, because most guides to this do not.

Animating a photograph of someone who has died is emotionally loaded in a way that animating a landscape is not, and the line between moving and unsettling is different for every family. A gentle breath and a blink usually land. A generated smile from someone who cannot consent to smiling frequently does not — and putting words in their mouth with lip sync is a step further again.

The practical version of this advice: show it to one other person who knew them before you show it to everybody. They will tell you immediately if it is wrong, and you will not be able to tell on your own after an hour of looking at it.

There is no technical rule here and I am not going to pretend there is one. But "we could" and "we should" are different questions, and the second one is the family's.

Assembling it

For a full memorial or anniversary video, the same shape works nearly every time:

1. Restore everything first, as a batch. Do not mix restoring and editing — different headspaces.
2. Lay the audio down first. Music, or a recorded voice telling the story. The pictures get cut to it, never the other way round.
3. Mostly camera moves, with two or three generated shots for the people it matters most for. Viewers cannot tell which is which.
4. Vary the movement. Push, then drift, then hold. Three identical zooms in a row reads as a slideshow template.
5. Hold longer than feels right. Five to seven seconds per photo. People want to look; a fast cut on a face they loved is frustrating.
6. Name and date on screen at least once for each person. In twenty years, that caption is the most valuable thing in the file.

A voiceover telling the story in a real voice — yours, or a recording of theirs if one exists — is worth more than any effect in this entire post.