Guide · updated August 24, 2026
How to animate a photo with AI
Animating a photo means handing a still image to a video model and letting it invent the seconds either side of the moment the shutter fired. The model keeps your composition and adds movement: a blink, a breath, hair shifting, the camera drifting slowly forward.
It works far better than most people expect, and it fails in one very predictable way — asking for too much motion. This guide covers which photos animate well, how to write the prompt, which model to choose, and how to avoid the melted faces that ruin most first attempts.
How to do it, step by step
- 1
Pick a photo that can move
Sharp, well-lit, with an obvious subject and some space around it. A crisp portrait or a clear product shot animates well; a dark, blurry, cluttered photo does not.
- 2
Upload it as the start frame
Every video model in renza accepts a start frame. The image becomes the first frame of the clip and the model works forward from it.
- 3
Write a short motion prompt
Describe the subject briefly, then the movement, then the camera, then the pace. "A woman in a red coat, hair moving gently in the breeze, slow push in, subtle motion" is a complete prompt.
- 4
Pick a model for the job
Seedance for faithful, subtle motion with audio. Kling for the sharpest detail and start-plus-end-frame control. Veo when the surrounding scene has to do something complex. LTX at 5 credits when you are still testing.
- 5
Generate, then re-generate
These models are non-deterministic — the same prompt gives different motion each run. Two or three attempts is normal. Draft on LTX, then run the winning prompt on a premium model.
Which photos animate well
- ·Sharp focus on the subject. Motion amplifies softness; a slightly blurry photo becomes a very blurry video.
- ·Good lighting with clear separation between subject and background.
- ·A single, obvious subject. Busy group scenes give the model too many things to move at once.
- ·Some empty space around the subject, so a camera move has somewhere to go.
- ·Higher resolution. If your source is small, upscale it before animating.
Which photos fight back
- ·Damaged or heavily scratched scans — restore them first, then animate.
- ·Very low resolution. The model invents detail that was never there and the invention wobbles.
- ·Extreme close-ups with no headroom. There is nowhere for movement to happen.
- ·Complex reflections, glass and mirrors, which physics engines still handle badly.
- ·Dense crowds where a dozen faces all need to move independently.
Old, damaged or faded photo? Restore and upscale it first — the results are dramatically better.
Read the old-photo guide →The prompt formula that works
A motion prompt has four parts, in this order: subject, movement, camera, pace. Subject anchors the model to what is in the frame. Movement says what should change. Camera says what the lens does. Pace tells it how much — and "subtle" is the most valuable word in image-to-video prompting.
Keep it to one or two lines. Long prompts help text-to-video, where the model is inventing a whole scene; in image-to-video the scene already exists and extra words mostly introduce conflict.
Portrait
a man in a dark blazer, slow blink and a slight smile, camera pushes in very slowly, subtle natural motion
Landscape
mountain valley at sunrise, mist drifting through the trees, clouds moving slowly, gentle drone push forward, cinematic pace
Product
a perfume bottle on stone, the camera orbits slowly to the right, light glinting across the glass, everything else still
Pet
a golden retriever looking at the camera, ears twitching, head tilts slightly, tail wagging out of frame, gentle handheld motion
Thirty more prompts, grouped by subject and ready to copy.
Get the prompt list →Choosing a model
- ·Seedance Pro (12 credits, 5 or 10s, audio): the default. Holds the source composition faithfully on subtle motion and adds ambient sound in the same pass.
- ·Kling 2.5 Turbo Pro (12 credits for 5s, 22 for 10s): sharpest fine detail, best camera movement, and the only model that also accepts an end frame.
- ·Veo 3.1 Fast (25 credits, 8s, audio): when the environment around the subject has to behave convincingly, or the clip needs dialogue.
- ·LTX Video (5 credits, 5s): drafting. Cheap enough to run a prompt four times while you work out what motion you actually want.
The end-frame trick
Kling accepts a start frame and an end frame, and it will interpolate the motion between them. This is the most underused feature in AI video: instead of describing what should happen and hoping, you show the model where the shot begins and where it ends.
It is ideal for controlled transformations — a product rotating to a specific angle, a portrait turning from profile to camera, a before-and-after that has to land exactly on the "after". If you can generate or supply both frames as images, you get near-deterministic motion out of a non-deterministic system.
Why faces warp, and how to stop it
Warping is almost always over-request. Ask a model to make someone dance, turn all the way around, or perform a large gesture, and it has to invent geometry the photo never contained — so it guesses, and the guess drifts. Ask for a blink, a breath and a two-degree head turn and it has almost nothing to invent.
The practical rule: the more you love the photo, the less motion you should ask for. Add "subtle motion", "gentle", "minimal movement" to the prompt, keep the camera move slow, and let the stillness do the work. A photograph that breathes is more affecting than a photograph that performs.
Mistakes to avoid
Asking for too much movement
The single biggest cause of bad results. Subtle motion reads as premium; dramatic motion reads as AI.
Animating a low-resolution source
Upscale first. Motion magnifies every flaw already in the file.
Writing a paragraph
Image-to-video prompts should be one or two lines. The scene is already decided by your photo.
Generating once and giving up
Every run produces different motion. Two or three attempts is normal, and drafting on a cheap model makes that painless.
Forgetting the camera
A slow push-in or a gentle orbit makes a clip feel filmed rather than warped. Name the camera move explicitly.
Upload a still. Watch it move.
Kling, Veo 3.1, Seedance and LTX all accept a start frame. From $9.99/month with 200 generations.
Frequently asked questions
How do I animate a photo with AI?
Upload the photo as a start frame, write a short prompt describing the motion you want, pick a model and generate. The clip comes back in 5 to 10 seconds of footage depending on the model. You can do it directly on the animate a photo page.
Which model is best for animating photos?
Seedance Pro for faithful, subtle motion with audio at 12 credits. Kling 2.5 for the sharpest detail and if you want to pin the final frame. Veo 3.1 when the surrounding scene needs to behave convincingly. LTX at 5 credits for drafting.
Why does the face look distorted?
You asked for too much motion, or the source photo was too small. Reduce the request to a blink, a breath and a slight head turn, add the word "subtle", and upscale the source first if it is low resolution.
How long is the resulting video?
5 to 10 seconds per generation: LTX does 5, Kling and Seedance do 5 or 10, Veo does 8. For something longer, generate several clips and cut them together.
Does the animated clip have sound?
It can. Seedance Pro and Veo 3.1 generate audio in the same pass — ambience, effects, and dialogue in Veo's case. Kling and LTX are video only, so you would add your own audio afterwards.
Can I animate an old family photo?
Yes, and it is one of the most affecting uses of the technology. Restore and upscale the scan first, then animate with a very gentle motion prompt. The old-photo guide walks through the whole sequence.
What does it cost?
It depends on the model: 5 credits for LTX, 12 for Kling or Seedance, 25 for Veo. Plans start at $9.99 a month with 200 generations, and the cost of a clip is shown before you generate it.