Plan a Video Prompt Like a Shot
A video prompt is easier to control when it reads like a production brief: one subject, one action, one camera idea, and a clear sound plan.
Describe What Changes Over Time
Subject and Action
Name who or what is present, the single main action, its pace, and the state at the end of the shot.
Scene
Specify location, time, weather, practical details, background activity, and spatial relationships.
Camera
Choose framing, angle, movement, lens feel, focus behavior, and whether the shot is continuous or cut.
Look
Set lighting, palette, contrast, medium, texture, and genre without stacking contradictory styles.
Audio
Call out dialogue, ambience, effects, music, timing, and any sounds that must not appear.
Exclusions
Prevent text, watermarks, duplicate subjects, unwanted cuts, camera shake, or changes to key features.
[Framing] of [subject] [performing one action] in [scene]. The action begins with [start state], develops through [beat], and ends with [end state]. Camera: [angle, movement, lens feel, focus]. Look: [light, color, texture, style]. Audio: [dialogue, ambience, effects, music]. Keep [continuity requirements]. Exclude [unwanted elements].
- Prototype one shot before asking for a sequence.
- Review motion, continuity, composition, and audio as separate dimensions.
- Change only the weakest block in the next prompt.
- Use model-specific controls for extension, first or last frames, and conversational edits when available.
Original Sources
Synthesized from UlazAI's Veo 3 Prompting Guide and Google's current video generation documentation. Check Google's documentation before production because model capabilities and recommended defaults can change.