Describe the beat, camera move, and lighting; Voor AI turns that brief into short-form motion you can drop into edits, UGC tests, or concept reels. The goal is controllable iteration: adjust language, regenerate, and compare outputs side by side.
Open with subject and action, then add environment, time of day, and lens cues. Closing with pacing (“slow dolly in”, “handheld micro-jitter”) helps temporal models lock onto motion instead of inventing unrelated motion blur.
If dialogue or on-screen text matters, state it verbatim in quotes. Models treat quoted strings as higher-priority tokens than descriptive prose.
When you already like a key still—perhaps from text to image—use it as the first frame in image to video so identity and wardrobe stay anchored. Pure text runs are best for exploration; reference-guided runs are best for continuity.
Version prompts in your ticket system the same way you version design files. Note model name, aspect ratio, and safety settings so QA can reproduce issues without guessing.
Plan audio elsewhere for now: finalize picture first, then lay dialog, music, and SFX in your NLE where metering and loudness standards are reliable.