Camera moves that survive into a video render
Animating a still is not "make it move" — it's "move the way a camera would." Image-to-video models learned from footage, so the vocabulary of a shot list — push-in, pan, parallax — is the vocabulary that lands.
One move per clip
The reliable shape is one camera move plus one subject action: "slow push-in while she turns to camera." A clip is a few seconds of one idea — stack two moves and the render splits the difference into mush.
If the shot needs a push-in and an orbit, render the push-in, then feed the result back in for the orbit as a second clip. Chained clips are how sequences get built.
Moves that land
Slow push-in is the safest instruction in the vocabulary — the camera eases toward the subject and the scene breathes. On a portrait, "slow push-in on her face" almost always renders clean.
Pull-back widens the frame to reveal the scene around the subject — strongest when the still has environment worth showing. A slow pan slides the frame laterally; keep the word "slow" on every camera instruction, because fast pans smear.
A partial orbit — "slight orbit around her" — holds. A full 360° does not; the far side of the subject is invented and drifts.
Parallax is free depth
Layers moving at different speeds is what reads as cinematic: rain in front, the subject mid-frame, neon behind. You don't say "parallax" — you name the layers: "rain streaks past the lens while she holds still, traffic lights shifting behind her."
Ambient motion is where fidelity lives: hair, cloth, water, crowd, smoke, particles. A still subject inside a moving world is the most faithful clip these models make.
Prompts that work
"Slow push-in — she holds the pose, neon reflections drift on the wet street." — camera move plus an ambient layer.
"Gentle pull-back, the crowd shifts behind her, she turns to camera at the end." — a reveal with one subject beat.
"Camera drifts right — foreground leaves cross the frame, she watches them pass." — a foreground layer doing the depth work.
The honest limits
Fast anything drifts — whip pans, running subjects, big rotations. The last seconds of any render are the least faithful to the first frame.
Faces at speed are the fragility point: a subject who stays put while the world moves keeps a clean face far longer than one striding through the shot.
Cost rule: prove the motion at 480p and four seconds before paying for the flagship tier. The full budget playbook is in animate a photo and the image-to-video page.