Topic
AI Video Prompts
A short reference on AI video prompts: the structure I use, why reference images matter more than adjectives, and where to read the official model documentation.
An AI video prompt is a written instruction given to a video generation model describing subject, action, camera, lighting and mood.
Good prompts read like a shot list, not like a wish. Vague adjectives produce generic footage; concrete camera and lighting terms produce usable footage.
The parts that matter
Subject and action first, then camera movement, then lens and lighting, then mood. Each one narrows what the model can return.
Reference images
A single reference image usually fixes more than another paragraph of description.
Iteration
Change one variable per attempt. Rewriting the whole prompt makes it impossible to learn what worked.
Trusted sources
- Google DeepMind — Veo
Official documentation of the video generation model behind much of today's AI video.
- OpenAI — Sora
OpenAI's own description of its text-to-video model and its limits.
- OpenAI — Prompt engineering guide
The reference I check when a prompt keeps returning the wrong result.