Sensia's video workspace animates a still you've already perfected — identity locked, pose composed. The only new information the model needs is what changes over time. Yet most first attempts read like a description of a photograph that happens to wobble.
The formula: subject motion + camera motion + pace
- Subject motion — what her hands do, where his weight shifts, the direction of a gaze. One or two beats, no more.
- Camera motion — push in, pull back, orbit, or static. Pick exactly one.
- Pace — 'slow' and 'gentle' are load-bearing words in intimate work; without them models default to music-video energy.
'She turns towards the camera and smiles, hair moving in the breeze, slow push in' is the whole art. The camera chips append a move to anything you've typed, and Suggest motion reads your still and proposes three variants you can swap in with a click.
Model notes from real runs
Surge anchors the first frame almost pixel-perfect and runs $0.10/second. LTX reconstructs the frame rather than locking it — expect about 75% similarity on the opening frame, in exchange for its longer 6–20s clips. If the still must survive untouched, start with Surge.