How to Recreate Any Video’s Motion with Your Own Photo (2026)

Most AI video tools ask you to pick a movement from their library. That works when you want a trending dance. It does not work when you want a specific motion — your own choreography, a friend’s gesture, a clip you saved months ago, the way someone in your family moves their hands when they talk.

Recreating an arbitrary motion works differently: you supply two files and the model maps one onto the other.

• The photo supplies identity — the face, hair, body, clothing.
• The reference clip supplies the movement — timing, posture, rhythm, gestures.

The output is a new video of the person in the photo performing the motion from the clip.

The two ways to do it

Both approaches use the same two inputs. What changes is which one wins when they conflict.

If you are unsure, start with motion transfer. It preserves the thing most people care about — that the result still looks like the person in their photo.

Step by step

1. Choose the reference clip first, not the photo

Counter-intuitive but it saves time: the clip is the harder input to get right, and its requirements constrain which photo will work.

A good reference clip has:

• One person — a second subject in frame makes the model guess which body to follow
• A fixed camera — handheld footage and pans get transferred as camera chaos
• Visible limbs — arms and legs not hidden behind props, tables, or other people
• One continuous shot — no cuts, no transitions, no montage
• Five to fifteen seconds — long enough to show the motion, short enough to stay clean

If the clip you want to use has a cut in it, trim to the single best shot before uploading.

2. Match the photo to the clip’s framing

Now pick a photo that gives the model what the clip needs:

Mismatched framing is the single most common cause of strange results. A seated photo plus a full-body breakdance reference forces the model to invent legs that were never visible.

3. Check the hands

This is the detail almost every guide skips, and it matters more than resolution.

Hands holding an object — a phone, a bag, a drink, a pet — transfer badly. The model cannot separate the hand from the object, so fingers bend the wrong way, duplicate, or smear. A photo where both hands are visible and empty produces visibly cleaner output.

The same applies to the reference clip: if the performer is holding a microphone or a phone, that shape tends to leak into your result.

4. Keep the face clear

Face landmarks anchor identity. Photos where the face is small, blurred, turned away, heavily shadowed, or covered by a phone produce weaker identity retention. Scanned vintage photos are usually fine — grain is not a problem, but a face smaller than a thumbnail is.

5. Generate, then change one thing at a time

Upload both files, choose the mode, and start the task. Generation typically takes 10 to 25 minutes.

If the result is not right, resist changing both inputs at once. Change the reference clip, or the photo — not both. Otherwise you cannot tell which input was responsible.

Failure modes and what actually causes them

Notice that nothing in that table is fixed by a better prompt. There is no prompt — the reference clip is the instruction. This is why the workflow is easier to control than text-to-video, and why bad inputs produce predictable failures.

What this is good for

Because the movement is supplied rather than imagined, the workflow suits jobs where the motion already exists:

• Recreating a specific routine — a choreography, a walk, a gesture you have on video
• Vintage photographs — one old portrait plus a short clip makes a moving tribute
• Consistent content at volume — one reference clip, many different photos
• Test shoots — check framing and timing before booking a real production
• Character work — a mascot, avatar, or illustration performing real human movement

What it is not good for

Being honest about the boundaries saves you a wasted credit:

• Inventing a scene from a description — that is text-to-video, a different tool category
• Long-form video — these are short clips (up to 60 seconds), not productions
• Crowd shots or group choreography — one subject per generation
• Fixing a bad photo — a blurry, tiny, or mostly hidden face stays a problem

FAQ

Does the person in the reference clip matter?

No. Identity comes from the photo. Many people deliberately use a professional dancer’s clip to get readable, well-lit movement, then apply it to their own portrait.

Can I use a clip from social media?

Only if you have the rights to it. The clip is an input to your output, and using someone else’s footage without permission is a copyright problem regardless of what the model does with it. Short clips you filmed yourself, or that you have licensed, are the safe route.

Can I chain several motions into one longer video?

Not inside a single generation. The usual approach is to generate each motion as a separate clip — one reference per generation — and join them in an editor.

How much does one clip cost?

With Animaker, one credit produces one video. Credits start at roughly $0.30 per video on the 100-credit pack ($29.90); the smallest pack is 20 credits for $9.90 ($0.50 per video). Failed generations are refunded automatically, so a task the service cannot complete does not cost you anything.

What if the result looks distorted?

Distortion almost always traces back to the inputs: hidden or occupied hands in the photo, a face that is not clearly visible, more than one person in the reference clip, or heavy camera movement. Change one input at a time so you can tell which one fixed it.

Can I copy the motion from any video onto my photo?

Yes, as long as the reference clip shows one person clearly and the motion is readable. The system extracts the movement from the reference and applies it to the person in your photo. Short clips with a fixed camera and one subject work best; clips with fast cuts, multiple people, or heavy motion blur transfer poorly.

Do I need a video of myself to make this work?

No. The person in the photo does not have to be you, and the person in the reference clip does not have to be the same person. The photo supplies the identity and appearance; the reference supplies the movement.

What is the difference between motion transfer and character replacement?

Motion transfer keeps the scene and identity from your photo, and copies the movement from the reference onto that subject. Character replacement keeps the scene from the reference clip and places your photo’s subject into it. Motion transfer is the better fit when the photo matters; character replacement is the better fit when the scene or setting matters.

How long should the reference video be?

Five to fifteen seconds is the sweet spot. Animaker accepts reference clips up to 60 seconds, but longer clips take longer to process and are more likely to contain cuts, camera movement, or multiple subjects that reduce quality.

Try it

Prepare one clear photo and one short clip whose movement you want to copy, then create a motion video with Animaker. New accounts get 2 free credits and no card is required, so the first attempt costs nothing.

If you want to see what the two modes look like before uploading anything, the photo dance examples page uses the same workflow end to end.