Turn a finished still into three seconds of motion by asking for less, not more.
The problem
The instinct with an image-to-video model is to ask for a lot: a camera move, a light change, liquid pouring, the cap lifting. What comes back is a product that morphs. Labels crawl, edges breathe, the whole object subtly reshapes itself over the clip.
Short, single-variable motion is what survives. One camera move, or one thing in the scene moving, never both. Three to five seconds. The still you already approved does the heavy lifting; the motion just has to not ruin it.
Copy this
“Only the camera moves” is the single most useful sentence in motion work. It gives the model one degree of freedom instead of many.
Four seconds because failures compound over time · a clip that holds for four will not necessarily hold for eight, and you rarely need eight.
The explicit no-morphing clause targets the label, which is the first thing to go and the thing a client will notice immediately.
What to watch for
- The label crawling or re-forming mid-clip · scrub frame by frame through the last second, where drift concentrates.
- The product drifting in scale as the camera moves, which reads as the object growing rather than the camera approaching.
- Motion that starts clean and degrades; cut the clip short at the point it breaks rather than re-rolling the whole thing.
- Generate motion from the approved still, never from a re-render of it. Same rule as master.
The drill
Take one hero still and produce three four-second clips from it: a slow push-in, a slow orbit, and a static frame where only the light shifts. Watch each at full size, then again at 2x speed where drift is easier to spot. Keep the one that holds. You now have motion for paid from an asset you'd already delivered.
