Emotion words like "alert" or "shocked" make video models act for the back row. Writing a beat-by-beat physical sequence with one involuntary tell gave us a restrained, believable glance instead of a startled stare.
Use one clean product photo as reference image 1, ask for one camera move and one action, and lock the product's look. Two unedited 9:16 tests, the exact prompts, and what to fix when the model drifts.
Seedance 2.0 options make 4 to 15-second clips with up to 9 images, 3 videos and 3 audio files as references; Seedance 2.5 goes to 30 seconds with 30, 10 and 10. A side-by-side table, how SoulNexus enforces the limits, and a quick way to choose.
One sheet with a head close-up and full-body front, side and back views on white gives a video model every angle it needs. Here is the layout, the prompt we used, the checks to run before you trust it, and how to attach it to every shot.