Luvio AI · 光趣 AILuvio AI · 光趣 AI
How to write AI image and video prompts
Use subjects, settings, actions, composition and light to write clearer image and video prompts, with practical examples for Luvio AI.
A prompt communicates a production goal. Make it clear and checkable, then refine one part at a time after reviewing the result.
Image prompts: describe one frame
Try subject, setting, composition, light and style. For example: a watchmaker in a dark blue work shirt sits at a wooden table with an open mechanical watch, medium close-up, soft side light from a window, realistic photography. Each detail should support the image you need.
Video prompts: add action and camera movement
Extend the scene with action and camera direction. For example: the watchmaker slowly picks up the watch, pauses and looks toward the window; the camera holds a medium close-up with a gentle push in; afternoon side light remains steady. Avoid too many actions or locations in one short clip.
References: say what stays and what changes
With a compatible model, identify which appearance or composition should stay and describe the desired action or environment changes. Avoid vague requests or contradictory styles, framing and lighting instructions.
Change one factor at a time
If the subject is unclear, refine its description. If the action is wrong, simplify it. If framing is off, clarify shot size and subject position. Record what each revision is meant to improve. Actual capabilities and limits depend on the selected model.
Getting started
Write one production goal
Say what the image or clip needs to communicate. Put the key subject and action first.
Add necessary visual detail
Include setting, composition, light and style. For video, add action order and camera movement.
Iterate from the result
Check the most important part first, keep what works and adjust specific factors that need improvement.
Common questions
Are longer prompts always better?
No. Contradictory or irrelevant details make results harder to assess. Prioritize the subject, action and visual goal.
Does every model support negative prompts?
No. Use them only when the selected model provides a corresponding field. Follow the settings and limits shown in the workspace.