How do I use this prompt with Gemini Omni?

Copy the prompt below, open the Gemini app or Google Flow, paste it into the video generator, and tweak duration and aspect ratio as needed.

Is this prompt free to use?

Yes. All prompts on this site are free. You only need an active Gemini Omni Flash subscription (or free YouTube Shorts access).

Can I modify this prompt?

Absolutely. See the "How to tweak" section below for suggested variations on style, subject, motion, and lighting.

CHARACTER

4:3 5s Micro-motion image-to-video

character

Old Photo Revival — Subtle Motion from a Still

Bring a vintage family photo gently to life with image-to-video: micro-motion only, no invented actions — the restraint is what keeps it believable.

old-photo image-to-video family-archive restoration memorial

Prompt

Using the attached vintage photograph, create a 5-second 4:3 video.
Bring the photo subtly to life: gentle, natural micro-movements only — a soft blink,
slight breathing motion, hair moving faintly as if in a light breeze. Keep the pose,
framing, clothing, and background exactly identical to the photograph. Static camera,
preserve the original film grain and faded color.

Why this prompt

“Bring an old photo to life” is one of the most emotionally loaded — and most searched — uses of AI video. It’s also image-to-video’s natural home turf: reference images preserve identity far better than any text description (Atlas Cloud), and with a family photo, identity is the entire point. Every design choice here is restraint. The prompt asks only for micro-movements — blink, breath, hair — because they’re simple motions in the zone hands-on testing found reliable (digit.in), and because anything more starts inventing behavior for a real person, which is both an identity-drift risk and, frankly, not what anyone wants from a memorial piece.

Source tier: 🟡 Pattern-composed (medium confidence — built from documented image-to-video patterns and verified failure modes, output not yet video-verified)

The keep... exactly identical sentence is a full “keep X identical” lock — the documented preservation pattern — applied to everything except the three permitted micro-motions. And 4:3 matters: most printed photos aren’t widescreen, and forcing 16:9 makes the model hallucinate scene beyond the photo’s edges. Match the aspect ratio to the print.

How to tweak

Even subtler: drop the blink, keep only faint breathing motion — for very formal portraits, a blink can feel like too much.
Group photos, one at a time: pick one subject — the woman seated on the left blinks softly; everyone else remains completely still. Animating everyone multiplies the identity-drift surface, the same math as the documented multi-subject problems (failure modes).
Weathered prints: add preserve the scratches, dust, and paper texture — the damage is part of the artifact’s authenticity; letting the model “clean” it changes the object’s meaning.
Gentle zoom variant: static camera → extremely slow push in for a documentary-style treatment — the Ken Burns move, animated.

Common failure modes

Asking for actions the photo doesn’t contain. “Make grandfather wave and smile” forces invention: hands (the documented articulation failure) plus an expression the model must fabricate on a real face. Micro-motion is the honest ceiling.
Identity drift on longer clips. Character consistency is documented to degrade over time and shots (Atlas Cloud consistency review) — 5 seconds is deliberately short; regenerate rather than extend if the face wanders.
Wrong aspect ratio. 16:9 from a portrait print = hallucinated furniture beyond the photo’s borders. 4:3 or 1:1 for most prints; crop the scan before upload.
Restoring and animating in one step. Colorizing and repairing a damaged photo is a separate edit — stacking restoration + animation in one prompt runs past the focus ceiling (Seaart). Restore first (any tool), then animate the restored still.

Notes

Consent and dignity: animate photos of your own family, or with the family’s consent for others. Google’s avatar identity rules point the same direction — Omni’s Avatar feature verifies your own face precisely because animating other living people without consent is the abuse case (failure modes). For the deceased, that judgment call belongs to their family.
Output carries a SynthID watermark.
For a fully synthetic period character (no real photo), the Pixar-style walk cycle covers text-driven character work instead.

Sources

Image-to-video identity preservation: Atlas Cloud features overview
Consistency degradation: Atlas Cloud multi-turn review
Simple vs. complex motion: digit.in hands-on
Preservation lock pattern: “keep X identical” post
Prompt focus ceiling: Seaart analysis