Animate a Face in a Photo with AI: What Actually Works

Ask a tool to animate a face in a photo and it will almost always try. Whether you end up with something convincing or something unsettling comes down to two things: the photo you started with, and how much motion you asked for.

What's actually happening to the photo

AI portrait animation works by building a simplified, moving rig from a single photo: eyelids, eyebrows, mouth, the tilt of the head.

From there, the model drives that rig through whatever motion you specified. A blink, a small smile, a turn toward the camera.

Everything around the rig gets pulled along for the ride. Hair, ears, and collars tend to look off in these clips, mostly because there's no dedicated model tracking them at all.

What makes a usable source photo

A face shot mostly straight on. A profile photo just doesn't give the model enough to work with.

Open, visible eyes. Closed eyes are the single worst starting point you can give it.

Sharp focus on the face itself. A soft or low-resolution photo tends to fall apart the moment it starts moving.

Even lighting. A hard shadow across half the face stays locked in place when the head turns, and it reads as a stain rather than a shadow.

One face in the frame. Group photos get processed in ways that are hard to predict.

How much motion you can realistically get

Minimal movement, a blink or a hint of breathing, is the safest bet. It almost always looks believable.

Facial expression is harder: a smile, raised eyebrows. Facial expression is harder to pull off, and it tends to look shakier on older faces.

A head turn works for a few degrees before the model starts warping features.

Speech with synced lips is its own problem with its own set of limits, covered in detail here.

The pattern holds across all of it: ask for less motion, get a more believable result.

Animating old family photos

Scan them instead of rephotographing them. A scan gives you the resolution the model actually needs to work with.

Keep the original somewhere safe, and don't let the animated version replace it.

Label every file: who's in it, what year, where it came from. A decade from now, that information will matter more than the image quality ever did.

And ask whoever's still around to remember. They'll catch details no model is ever going to guess correctly.

What to do once you have an animated portrait

Keep the source photo as a separate file. The animated clip is a new piece of media, not an upgraded version of the old one.

Label it too, so years from now it's clear this file was generated and not the original photo.

Don't post it without context if the person in it is recognizable.

And check with family before sharing an animated portrait of someone they loved. Someone's reaction to an animated portrait of a loved one can be very different from what another family member feels seeing the same clip.

Where the line actually sits

Animating someone else's face without asking first isn't something to do. It's a legal problem, and it also shapes how you look if the clip ever gets out.

A portrait of someone who's passed away is a family call, and there's no clean answer here. I've done this kind of work myself, and reactions from people in the same family can go in completely different directions.

Using a public figure's likeness in an ad without permission is a direct conflict, and there's a separate piece on rights around generated footage.

A fully generated character who doesn't resemble a real person sidesteps all of this, and for commercial work, it's the calmer option.

Though the line between "doesn't resemble" and "resembles" is thinner than it sounds. More than once, I've caught myself thinking a generated face reminded me of someone real.

Frequently asked questions

What kind of photo works best?

A forward-facing shot, eyes open, sharp focus, even lighting, and just one face in the frame.

Which type of motion comes out best?

Minimal motion, like a blink or light breathing. The more movement you ask for, the higher the risk of distortion.

Why do hair and ears look strange in the result?

The moving rig is built from facial features alone. Everything around it just gets dragged along with no model of its own.

Can I animate a photo of someone else?

Only with their consent. For commercial projects, a generated character is the safer route.

Why do closed eyes ruin the output?

The model builds its rig from visible features. Without open eyes to reference, it has to guess their shape, and the face ends up looking like someone else entirely.

Sources

1. Flami: talking photo AI
2. Flami: AI photo animation
3. Flami: copyright on AI video