How to Animate Photos With AI, Beyond Deep Nostalgia

Deep Nostalgia is a MyHeritage feature that animates the face in a still photograph, part of its genealogy service, and MyHeritage lists LiveMemory as a related tool (as of October 2026). When it launched in 2021 it worked by applying prerecorded facial movements to the face in your photo, which is why so many animated ancestors turn, blink and smile in the same few ways. It is simple, it moved a lot of people, and it lives inside MyHeritage's own plans and terms. MyHeritage and Deep Nostalgia are its owner's trademarks; DreamClip is not affiliated with either.

Pixbim sells desktop software for the same job, as of October 2026. Both are built around motions the product provides. DreamClip does not run Deep Nostalgia or Pixbim. It animates the photo from a start frame on Kling V3 Turbo, Veo 3.1 or Kling 3.0, from a line you write, and makes it speak with the talking avatar.

Each step is one tool on DreamClip, and you can stop after any of them: an enhanced scan is already worth keeping, and a silent clip is often all a family wants.

Step 2 is the one most people search for, and it is the same image to video step as for any other photo; our image to video guide covers the general craft, and the image to video page shows photos beside the clips they became. What is different about old photos is the scan, the respect and the voice, which is where this guide spends its time.

The clip inherits every flaw in the still. Blur stays blur, a crease becomes a crease that moves, and dust crawls. Ten minutes on the scan saves every run after it.

• Scan flat, or photograph the print in daylight by a window with no flash and no glare. Fill the frame with the print and keep the phone parallel to it.
• Crop to the person who matters. A whole wedding party animates badly; the couple animates well.
• Leave the colour alone. A faded or sepia print is part of the memory; a model asked to colourise it invents colours, and you can say keep the sepia tone and grain in the prompt.
• Run Enhance image in the Enhancer. It sharpens the still and doubles its size, which gives the video model a face with detail to hold.
• For a tear or a scratch across the face, open the still in the Editor and use Retouch, which fills a marked area to match what is around it. Fix only the damage; a face restored too far stops looking like them.

Judge the enhanced still at full size before you animate it. If the eyes are not sharp, start again from a better scan rather than hoping the clip will hide it.

In the Video studio, switch the start frame on and drop the enhanced still on it. The model takes your photo as the first frame and renders what follows. Your prompt describes only what moves. Nothing about who they are or what they wear belongs in it; the photo already says that.

Which model to open

• Kling V3 Turbo first. It is silent, runs anywhere from 3 to 15 seconds at 720p or 1080p, and renders 1:1, which fits most old prints without a crop. Plan on a second and a third take; a face usually needs them.
• Veo 3.1 when you want the room with the picture: a clock, a radio, a street outside. It renders 4, 6 or 8 seconds with sound on by default, in 16:9 or 9:16 only, so crop a square print first.
• Kling 3.0 when you want sound and a longer take together, 3 to 15 seconds with sound on by default, square as well.

Settings that suit old photos

• Short. Five to eight seconds is a moment; fifteen seconds of a smile is a loop, and the face drifts past ten.
• The print's own shape. 1:1 for a square print on Kling V3 Turbo or Kling 3.0, 9:16 for a portrait print, 16:9 for a wide one.
• 720p for the first take, 1080p for the keeper.

Only a delivered clip is charged, so a failed take costs nothing. Watch the first second and the eyes. If either slips, make the prompt calmer and run it again. The Kling and Veo pages have more prompts for people.

Three video models for the clip and two voices for the speech, as the studio sells them today.

Each card opens its model in the Video studio; drop the enhanced scan on the start frame there.

• KlingKling V3 Turbo3 to 15 s · 720p, 1080p
• GoogleVeo 3.1 (with audio)4, 6 or 8 s · 720p, 1080p
• KlingKling 3.03 to 15 s · 720p, 1080p

Hearing someone is a bigger thing than seeing them move, so ask the family first. If they want it, the route is the talking avatar: upload the portrait and an audio file, and the tool renders the face speaking with the mouth matched to the words. One face looking toward the camera works best. Keep the line short, a sentence or two; the run is priced by the length of the audio.

Where the audio comes from

• A preset voice. ElevenLabs v3 reads the line you type in one of its preset voices; pick one of the right age and warmth, and the text to speech page plays them.
• A real recording. If you have the person on tape, a voicemail or a home video, you can use that audio as it is, or have Voice Clone read a new line in that voice from a sample of at least 10 seconds of them talking normally, with no music behind it. Our voice cloning guide covers the sample.
• A family voice. Many families record a grandchild reading the line instead. The voice cloning page plays a clone next to its sample.

The avatar takes an uploaded audio file, so download the read from the voice studio and upload it there. For a clip you have already animated, Lip Sync in the Editor fits the mouth in that video to the new audio, at 720p; the avatar is the simpler route when you start from a still.

A voice only from a recording you have the right to use, and only with the family's agreement. Our terms forbid using the service to impersonate real people without their consent, and a clip of a relative saying words they never said should stay with the people who loved them.

The best prompts for old photos are the smallest: one motion, a steady camera, and the light left as it is. These are written for the three models above and have not been rendered by us; use your own photo and change one thing at a time.

Say what should not move when it matters: nothing else moves, the camera holds still. Say what to keep: the sepia tone, the black and white, the grain. These phrases stop the model from modernising a photo that should stay old.

A group photo asks the model to hold several faces at once, and it holds each one worse than it would hold one. Crop in until the faces that matter fill the frame. For the whole group, give the prompt one shared motion, everyone turns to the camera and smiles, rather than an action per person, and keep it under six seconds. Faces at the back of a crowd are the first to drift.

Faded prints need the enhance step more than any other, and they need the prompt to say keep the fading and the soft contrast. A faded print sometimes comes back stronger than it was; if the family prefers it that way, keep it, but make it a choice. Hand tinted photos sit between the two: say keep the hand tinted colours, and expect the tint to hold better on the face than on the background.

Your uploads, prompts and finished clips are stored in your DreamClip account so you can get back to them, and they are not shared publicly. To render a clip, the photo and the prompt are sent to the model provider that runs it. Your data is kept while your account exists; deleting the account removes your personal data, prompts and stored generations from our systems. Our privacy policy has the full wording, and you keep the rights to what you make.

A cloned voice follows the same rule: the sample lives in your account and is deleted with the voice; the voice cloning guide has the detail.

• Teeth and open mouths. A wide smile from a closed one is where every model slips; ask for a small smile and let the avatar do the speech.
• Big head turns. A portrait taken face on never showed the side of the head, so the model invents it; keep turns to a glance.
• A face that is not sharp in the scan. No model recovers detail that is not there; enhance first, or accept a soft clip.
• Length. A face that holds for six seconds can soften by twelve; two short clips cut together beat one long one.
• The voice is a guess. A preset voice is a stranger's; a clone is only as close as the recording it comes from. Say so to the family before they hear it.
• Veo 3.1 and the square print. It renders 16:9 and 9:16 only, so a square photo is cropped; Kling V3 Turbo and Kling 3.0 render 1:1.