Describe a shot, not a story
Each generation is one 5-second clip. That is one shot — a single moment with a subject, a setting and some movement — not a sequence of scenes or a plot.
Start with the plain sentence you would say to a camera operator: “a golden retriever running through shallow surf at sunset.” That alone gives the model something clear to work with.
Avoid stacking events. A prompt asking for someone to enter a room, pick up a phone, and then walk outside is more than 5 seconds can show, and the model will usually pick one part or blur them together.