A prompt has four slots, not one sentence
Write the prompt as four separate decisions before you merge them. Subject: who and what is in frame. Action: the one movement that defines the shot. Camera: how the frame itself behaves. Light: the source and quality of illumination.
- Subject — age-appropriate, adult, described by build and wardrobe rather than by name.
- Action — exactly one primary verb. Two competing actions produce a smear.
- Camera — static, slow push, orbit, handheld. Choose one.
- Light — direction plus quality: window light, hard flash, neon from screen left.
One action per shot
A five-second clip cannot resolve two simultaneous motions. If you ask for a turn and a walk, the model averages them into a drift. Pick the motion that carries the story and let the rest of the frame stay quiet.
If you cannot mime the shot in one gesture, the prompt is asking for two shots.
Words that move the camera
Camera language is the highest-leverage vocabulary in adult video prompting, because it changes the whole frame rather than one body part. Useful terms include slow push in, pull back, handheld follow, static tripod, slight orbit, over-the-shoulder and low angle.
Words that do nothing
Quality adjectives such as cinematic, 8k masterwork and ultra detailed add almost nothing to a video model and consume attention budget that belongs to the action. Replace three adjectives with one concrete physical detail — the reflection of a screen on skin, for example.
Iterate one slot at a time
When a result disappoints, change exactly one of the four slots and regenerate. Changing the action and the camera together tells you nothing about which one helped. Single-slot iteration is slower per round and faster overall.