Video agent prompt is easiest to handle when you decide a scene-by-scene request with purpose, visual action, narration, timing, and transition. The practical route is to record that decision, apply the checks below, and then use the Video Agent when the input or brief is ready. This guide covers specify each scene's purpose, visual, narration, duration and transition without repeating the full commercial feature list.
Make the first decision before you start
Write one scene as one job. State what changes on screen, what the audience hears, and how the scene hands off to the next. Keep duration in the video settings when the interface exposes that control.
The goal is a reviewable input, not a perfect promise. If the source fails a rights, clarity, or fit check, replace it before you ask a generation tool to interpret it.
Use the scene prompt as a working sheet
Write the answers in plain language. For example, name the one visible action, the one detail that must survive, and the one condition that would make you stop. This keeps the video agent prompt task focused instead of turning it into a collection of decorative instructions.
Copy a scene-by-scene prompt
Create a 3-scene product explainer for [audience]. Scene 1 - Purpose: show the problem. Visual: [subject] performs [action] in [setting]. Narration: "[one approved sentence]." Duration: [short / medium / long]. Scene 2 - Purpose: show one proof point. Visual: [specific detail] changes or becomes visible. Narration: "[one supportable sentence]." Handoff: carry [prop, person, or screen direction] forward. Scene 3 - Purpose: give one next step. Visual: [clear end action]. Narration: "[single instruction]." Review: keep the approved wording and rights-cleared inputs; flag any scene that changes the message.
The bracketed values are the part to customize. Keep each scene to one job so a reviewer can tell which line or visual needs repair.
Apply the sheet in a short review loop
- 1
- State the job. Write the viewer, the message, and the single outcome this page must support. 2
- Choose the smallest useful input. Remove unrelated details, unsupported claims, and material you do not have permission to use. 3
- Use the Pippit route at the handoff. Bring the prepared input or brief to the Video Agent and select the workflow that matches the decision you made. 4
- Review the first useful result. Compare it with the check sheet. Keep what supports the job, edit what is local, and stop when a rights or continuity issue cannot be repaired safely.
For this page, the next action is deliberately narrow: specify each scene's purpose, visual, narration, duration and transition. A broader sibling task belongs in the other guides collected on the related Video Agent guides.
Know when the result is ready to move on
Accept the work when the central decision is visible, the input remains authorized, and the result does not introduce a new problem that changes the message. If one check fails, return to that field instead of adding more adjectives or more steps. Human review still owns claims, rights, pronunciation, selection, and publication.
FAQs
How should a video agent prompt be structured?
Give each scene a purpose, visual action, narration, duration intent, and handoff. The starter prompt above makes the fields explicit.
Should each scene have its own prompt?
Yes, when scenes have different purposes or need separate review. Keep shared audience, message, and continuity anchors outside the individual scene rows.
Where should I set video duration?
State duration intent in the scene row or the relevant workflow setting. Review the actual timing after generation instead of assuming the request was followed.
Continue with the finished workflow
When the sheet is complete, use the Video Agent for the production step. If you need another decision in this topic, browse the related Video Agent guides.