Give a photo just enough motion to feel alive: a breath, a blink, a slow drift of light. Restraint is the skill here, and this page is mostly about how to get it right.
Input
JPG · PNG · WebP · HEIC, 20 MB
Motion levels
Subtle · moderate · dramatic
Models available
13, selectable per job
Clip length
5–30 s depending on model
Audio
Available on most models
Free tier
3 drafts daily, no account
Typical time
Seconds to 3 minutes
Our one-line verdict: less motion, better result. Every disappointing photo animation we've seen was asked to do too much — the Subtle setting exists because it's usually the right answer.
Animating a photo is the most reliable thing AI video does, and the most commonly botched. The failure is almost never technical: it's asking for too much movement. A portrait that breathes, blinks once and drifts a few centimetres reads as a living moment. The same portrait asked to turn its head, change expression and walk forward reads as a rubber mask. This page defaults to restraint for that reason, and it's the one honest tip that beats every setting: describe less motion than you think you want. If you want the full model-by-model breakdown for heavier work, the image-to-video page covers that in detail — this one is about getting a single photo to feel alive.

Sharpest copy you have. For prints, scan rather than photograph them — the difference in the final clip is bigger than you'd expect.

Start on Subtle. Name what moves and what must not: 'she breathes and blinks once, the camera drifts in slightly, nothing else changes.'

Run it on the free tier first to check the motion reads. Then re-run the same prompt on a paid model for the clean version.
A box of scanned prints and an afternoon. At the subtle setting these come out quiet and strange and rather moving — the most-loved use of this page by some distance.
A feed rewards motion. Animating a photo you already have is the shortest path from a static post to a video-native one.
A slow drift across a packshot, steam rising from the cup, a light sweep across the surface. Cheaper than a video shoot and reuses the photography you already paid for.
The right column is half technical limits and half rules we've chosen. Knowing the failure modes before you upload saves more time than any guide.
• Subtlety, when you ask for it.
Breath, blinks, stray hair, a few centimetres of camera drift. The motion nobody notices consciously, which is exactly why it works.
• Old photographs.
Scanned prints animate surprisingly well at the subtle setting — grain and all. It's one of the most affecting things you can do with a family archive.
• Free to test.
Three drafts a day, no account. Motion ideas are quick to try and cheap to abandon, which is the correct way to work on this.
• Thirteen engines behind one box.
Same upload, different model when the job changes — speed, fidelity to your photo, length, audio. The routing lives on the image-to-video page.
• Too much motion breaks faces.
Ask for a head turn plus an expression change and you'll get drift, warping or a different person. This is the single cause of most bad results.
• Low-resolution sources.
Animation doesn't add detail. A soft photo produces a soft clip — start from the sharpest copy you have, and scan prints properly.
• Complex scene changes.
A still can be animated, not rewritten. If you need a different setting or new elements, generate the video from text instead.
• Photos of people who didn't agree.
Your own photos or ones you have the right to use. No public figures, no minors, nothing designed to misrepresent someone.
Upload the image, ask for less motion than you think you want, and generate. Start on the Subtle setting and the free draft tier — the whole loop takes under a minute, and restraint is what separates a living photograph from a rubber mask.
The full thirteen-model routing table.
Two photos instead of one.
Restore an old scan before animating it.