text-to-video-prompting
Installation
SKILL.md
Text-to-Video Prompting
Image prompting describes a frame. Video prompting describes a frame plus what changes over time, and the second part is where almost all failures live.
A prompt that reads like an image prompt gets you either a near-static clip or motion the model invented.
The four things a video prompt must state
Most prompts state one or two and let the model guess the rest. Guessed motion is the drifting, morphing output people complain about.
| Element | State it as | Example |
|---|---|---|
| Subject | What is in frame, concretely | "A ceramic mug, steam rising" |
| Subject motion | What the subject does over the clip | "steam thickens, then drifts left" |
| Camera | The move, or explicitly none | "camera locked, no movement" |
| Look | Lighting, lens, grade, mood | "soft morning window light, shallow depth of field" |
Naming the camera is the highest-leverage addition. Without it the model picks a move, and the move it picks is usually a slow drift that makes the clip feel unintentional.