Reference images tell a video model what a subject looks like from more than one side. Shooting them deliberately makes a large difference.
Angles
Take a front view, a three-quarter view and a side view. These three cover most of the turns a subject makes in a short clip.
Light
Use the same soft, even light for every shot. Strong shadows in one picture and none in another confuse the model about the subject's real shape.
Background
A plain, uncluttered background keeps the model focused on the subject. Busy backgrounds tend to leak into the result.
Framing
Fill most of the frame with the subject, but leave a small margin so nothing important is cut off.
Consistency
Do not change clothing, labels or accessories between shots. Anything that differs will flicker between versions in the generated video.
Tools that accept several references alongside a prompt, for example image to video AI, benefit most from this preparation.